CN118393870A - Methods for selecting data for training artificial intelligence, methods for generating training datasets, training datasets - Google Patents

Abstract

The present invention relates to a method for selecting data for training artificial intelligence for controlling a handheld machine tool, comprising: providing a plurality of data sets of measured values of at least one operating variable of the handheld machine tool; implementing a setting for determining the operating state of the handheld machine tool and/or predicting an event time point on the data sets, and outputting the determination results and/or prediction results from a plurality of model structures; determining the result deviation between the determination results and/or prediction results of the plurality of model structures; selecting at least one data set, which, when executing the plurality of model structures, causes the result deviation of the determination results and/or prediction results to reach or exceed a predetermined limit value. Determining the result deviation between the determination results and/or prediction results of the plurality of model structures; selecting at least one data set, which, when executing the plurality of model structures, causes the determination results and/or prediction results to have a result deviation that reaches or exceeds a predetermined limit value. Determining the result deviation between the determination results and/or prediction results of the plurality of model structures; selecting at least one data set, which, when executing the plurality of model structures, causes the determination results and/or prediction results to have a result deviation that reaches or exceeds a predetermined limit value.

Description

Methods for selecting data for training artificial intelligence, methods for generating training data sets, training data sets

Technical Field

The present invention relates to a method for selecting data for training artificial intelligence. In addition, the present invention relates to a method for generating a training data set. In addition, the present invention relates to a training data set.

Background technique

Methods for selecting data for training artificial intelligence are known from the prior art.

Summary of the invention

The object of the present invention is to provide an improved method for selecting data for training an artificial intelligence, a method for generating a training data set and a training data set.

This object is achieved by an improved method for selecting data for training an artificial intelligence, a method for generating a training data set and a training data set according to the invention. Advantageous embodiments are listed below.

According to one aspect, a method for selecting data for training an artificial intelligence for controlling a handheld machine tool is provided, the method comprising:

providing a plurality of data sets of measured values of at least one operating variable of the handheld power tool, wherein the data sets are based on a plurality of measurements of the operating variable of the handheld power tool or of a plurality of handheld power tools of the same type;

executing a plurality of different model structures on the data set, which are provided for determining an operating state of the hand-held power tool and/or for predicting an event time at which the hand-held power tool changes from one operating state to another operating state, and

Outputting the obtained results and/or predicted results from the multiple model structures;

Obtaining result deviations between the obtained results and/or predicted results of the plurality of model structures; and

At least one data set is selected which, when executing the plurality of model structures, has led to an ascertainment result and/or a prediction result having a result deviation which reaches or exceeds a predefined limit value.

The technical advantage is that an improved method for selecting data for training an artificial intelligence for controlling a handheld machine tool can be provided. To this end, a plurality of data sets of measured values of at least one operating variable of the handheld machine tool are first received. Subsequently, a plurality of model structures are applied to the data sets. Here, the model structures are provided for determining the operating state of the handheld machine tool and/or predicting an event time at which the handheld machine tool changes from one operating state to another operating state. Thus, by applying the model structures to the plurality of data sets, the operating state of the handheld machine tool and/or the predicted event time are determined by each model structure. Subsequently, the corresponding determination results of the model structures (i.e., the determined operating states or the predicted event times) are compared with each other, and the result deviations between the determination results or the determined operating states or the predicted event times are determined. Subsequently, a data set from the plurality of data sets is selected for generating a training data set, for which the corresponding determination results or prediction results of different model structures reach or exceed a predefined limit value. This makes it possible to select, for the training data set, in particular, data sets that lead to different results when tested by the model structure. This makes it possible to generate particularly heterogeneous training data sets. Furthermore, by taking into account corresponding data sets that lead to different results in the model structure, data sets that do not allow for a clear determination or prediction of results can be taken into account in the training data set. This makes it possible to improve the training of artificial intelligence for determining operating states or predicting event times by not only using clear data sets in the training, but also taking into account data sets that may lead to less clear results.

According to one specific embodiment, the model structure is designed as a data-based model and/or as artificial intelligence.

This can achieve the technical advantage that a particularly convincing simulation of the determination of operating states or predictions of event times by the artificial intelligence to be trained can be achieved through the model structure. This can make it possible to select a convincing data set for the training data set of the artificial intelligence to be trained.

According to one embodiment, the artificial intelligence is configured as a neural network and/or a support vector machine and/or a decision tree having different structures.

This provides the technical advantage that the model structure is constructed similarly or comparably to the artificial intelligence to be trained, which ensures that data sets that lead to different results when implementing the model structure will also lead to similar ambiguous prediction results in the training of the artificial intelligence to be trained. This makes it possible to achieve a training data set that is as heterogeneous as possible.

According to one specific embodiment, the result deviation is designed as a variance.

This makes it possible to achieve the technical advantage that by taking into account the variance, a precise ascertainment of different ascertainment results for a plurality of model structures or result deviations of the prediction results is made possible.

According to one specific embodiment, the data set includes labeled and/or unlabeled measured values of the operating variable.

This allows the technical advantage of achieving a high degree of flexibility in the selection of data sets. In particular, the selection of data sets can be carried out on data sets with unlabeled measured values. This allows the selection to be made directly when receiving a data set, without requiring complex processing of the data records in the form of labeled measured values.

According to one embodiment, the ascertainment result includes: no operating state is ascertained and/or the prediction result includes: no event time is predicted.

This can achieve the technical advantage that a precise detection of deviations in the ascertainment results or prediction results is achieved. A deviation exists when, for example, one model structure outputs an unambiguous ascertainment of an operating state or an unambiguous prediction of an event time, while another model structure cannot ascertain any operating state or cannot unambiguously predict an event time.

According to one embodiment, the method further comprises:

Taking into account the selected data set, a retraining of a plurality of model structures constructed as artificial intelligence is carried out.

Thus, the following technical advantages can be achieved: by retraining a plurality of model structures constructed as artificial intelligence under consideration of the selected data set (in the previous application of the plurality of model structures on the corresponding data set, the selected data set resulted in deviations in the results of the search or prediction), the training state of the model structure on the current training data set can be updated. Thus, when the model structure is reapplied to the newly received data set, only such data sets can be identified or selected: the data sets result in different search results or prediction results caused by the model structure. The data set consistent with the previously selected data set will not result in different results again after retraining the model structure. Therefore, it can be achieved through retraining that only data sets that are actually different from the data sets received in the training data set can result in different search results or prediction results caused by the model structure. Thus, it can be avoided that redundancy is received in the training data set. Here, retraining describes the training of an already trained artificial intelligence based on a new training data set.

According to one specific embodiment, the plurality of data sets are subsets of a comprehensive data set of measured values of the operating variable, wherein the data sets are selected for provision according to a selection criterion.

This allows the technical advantage to be achieved that redundancies can be filtered out of the existing data set by dividing the existing data set and carrying out a plurality of model structures on selected subsets of the existing data set, in order to thereby generate a training data set that is as heterogeneous as possible.

According to one specific embodiment, in order to provide a plurality of data sets, the data sets are received from a handheld power tool or from a plurality of handheld power tools of the same type.

This achieves the technical advantage that data sets from a plurality of different handheld power tools of the same type can be taken into account.

According to one embodiment, the control parameters include one or more items from the following list: engine speed, engine current, engine power of the engine of the handheld machine tool, and/or wherein the operating parameters include one or more items from the following list: engine current, engine position angle, engine rotation speed, voltage of a voltage source of the handheld machine tool, movement and/or vibration of or in the handheld machine tool, and/or wherein the operating state is one or more items from the following list: load range (the handheld machine tool operates within the load range), intensity of vibration in the handheld machine tool and/or on the machined workpiece and/or in a user of the handheld machine tool, temperature in the handheld machine tool and/or on the workpiece, operating mode (the handheld machine tool operates in the operating mode), working progress of the handheld machine tool, material of the workpiece, presence of a form fit between the handheld machine tool and the workpiece.

This can achieve the technical advantage that the handheld power tool can be precisely controlled taking into account the output target value of the control parameter. Here, the engine speed, engine current or engine power of the engine of the handheld power tool are reliable control parameters based on which the handheld power tool is controlled.

Furthermore, the operating variables can provide meaningful measured variables for determining the operating state. By measuring the motor current of the handheld power tool, the motor position angle, the motor rotational speed, the operating voltage of its voltage source, or the movements and/or vibrations of or in the handheld power tool, meaningful information can be obtained, which can be used to determine the operating state of the handheld power tool.

For example, in the above example, by measuring the motor current, it can be identified whether the screw to be screwed in has been screwed into the workpiece to be processed in a locked manner. When the form fit is achieved, the change in the motor current and the motor rotation speed or the motor speed can be identified, so that the operating state can be accurately determined based on this. For example, it can also be identified based on the movement signal whether the handheld machine tool, such as a screwdriver, has moved to the next work location and thus whether the previous working phase (i.e., the previous tightening process) has ended. This achieves a reset setting of the operating state corresponding to the then new working phase (i.e., the then new tightening process) that is matched in time.

In addition, when controlling the handheld power tool in the form of the second target value of the control parameter, different operating states can be taken into account: the handheld power tool can be in these different operating states or the handheld power tool can enter these different operating states. Therefore, the method according to the invention can take into account a wide range of operating states, thereby providing a control method that can be used universally.

According to one aspect, a method for generating a training data set for training artificial intelligence is provided, the method comprising:

implementing a method according to any one of the preceding embodiments for selecting data for training an artificial intelligence for controlling a handheld power tool;

The selected datasets are merged into a training dataset.

As a result, the following technical advantage can be achieved: an improved method for generating a training data set for training an artificial intelligence is provided, wherein for this purpose a method for selecting data having the technical advantages described above is implemented.

According to one aspect, a training data set for training an artificial intelligence for controlling a handheld power tool is provided, wherein the training data set is generated according to a method for generating a training data set for training an artificial intelligence.

According to one aspect, a computing unit is provided which is configured to carry out a method for selecting data for training an artificial intelligence for controlling a handheld power tool and/or a method for generating a training data set according to one of the above embodiments.

According to one aspect, a computer program product comprising instructions is provided, which, when the program is executed by a data processing unit, causes the data processing unit to implement a method for selecting data for training an artificial intelligence for controlling a hand-held machine tool and/or a method for generating a training data set according to any of the above embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

An embodiment of the present invention is explained with reference to the following drawings. In the drawings:

FIG. 1 shows a schematic illustration of a handheld power tool according to one embodiment;

FIG. 2 shows a further schematic illustration of the hand-held power tool, in which individual functional sequences of the hand-held power tool are illustrated;

FIG3 shows the time course of operating parameters of the handheld power tool and the rotation angle of the engine of the handheld power tool;

FIG. 4 shows a schematic illustration of a hand-held power tool according to one embodiment, wherein the hand-held power tool is shown in different operating states;

FIG. 5 shows a schematic illustration of a drive control of a handheld power tool according to one embodiment;

FIG. 6 shows a diagrammatic representation of a system for selecting data for training artificial intelligence according to one embodiment;

FIG. 7 shows a flow chart of a method for selecting data for training artificial intelligence according to one embodiment;

FIG8 shows a flow chart of a method for generating a training data set for training artificial intelligence according to one embodiment;

FIG. 9 shows a schematic diagram of an artificial intelligence which is configured to be used in controlling a handheld power tool; and

FIG. 10 shows a schematic diagram of a computer program product.

Detailed ways

FIG. 1 shows a schematic illustration of a portable power tool 100 according to one specific embodiment.

The exemplary handheld power tool 100 comprises a motor 101 with a motor control unit 103. The handheld power tool 100 further comprises a computing unit 105 on which a state determination module 107 is installed and can be executed. The computing unit 105 with the state determination module 107 is provided for executing the method according to the invention in order to control the handheld power tool 100. The handheld power tool 100 further comprises a voltage source 111 and a current measuring device 113. The handheld power tool 100 further comprises a trigger switch 109, by means of which the user can control the handheld power tool 100. In addition, the handheld power tool 100 comprises a regulating device 115, by means of which different operating modes of the handheld power tool 100 can be adjusted. Finally, the handheld power tool 100 comprises a tool 117, by means of which a corresponding working process can be executed by the execution of the handheld power tool 100.

The handheld power tool 100 can be designed, for example, as a screwdriver or a battery-operated screwdriver. For this purpose, the tool 117 can be designed, in particular, as a receptacle for a replaceable screwdriver rotary blade.

The illustrated motor control unit 103 may include, in particular, the associated power components of the motor control unit. The motor 101 may also include a corresponding transmission, which is not explicitly shown in FIG. 1 .

The motor 101 can be designed, for example, as a mechanically or electrically commutated DC motor. The corresponding transmission can be designed as a planetary gear.

According to the invention, the power part of the engine controller 103 can convert the control signal into the voltage change process or current change process required for the engine 101, for example, by PWM pulse width modulation. For this purpose, the control signal can first be converted into a corresponding digital signal, and then the corresponding converted signal can be transmitted via a suitable data bus (for example, I2C or SPI). In the case of an electrically commutated DC motor, a corresponding rotating field can be generated, which can be tracked synchronously with the rotation of the rotor. The engine controller can implement voltage-driven or speed-driven regulation. In the case of voltage-driven regulation, as the load (torque) increases, the engine speed decreases when the operating current increases. Information about the engine speed or dynamic rotation angle can be obtained system-inherently based on phase changes. In addition or alternatively, a continuous rotation angle sensor (not shown in FIG. 1) can be used to identify the rotor position. Such a rotor position signal can be transmitted to the calculation unit 105 in order to control the handheld power tool 100. This signal transmission can again be carried out by PWM, I2C, SPI or analog.

According to one specific embodiment, instead of the rotational angle sensor, an algorithm is implemented in the control unit, which infers the rotational angle from the measured motor current and motor voltage, which contain signal components resulting from the voltage induced back by the rotor. In this specific embodiment, the rotational angle sensor can be functionally replaced by the algorithm implementation.

The current supply via the voltage source 111 can be provided by a plurality of battery elements (eg lithium-ion batteries). Overcharging, overcurrent and deep discharge can be prevented by a corresponding battery management system.

According to one embodiment, trigger switch 109 may be implemented as a potentiometer, which provides an analog control signal corresponding to a linear actuation of trigger switch 109 to computing unit 103 for controlling handheld power tool 100. By actuating trigger switch 109, corresponding user inputs may be provided for controlling handheld power tool 100.

The current measuring device 113 can determine the battery current of the voltage source 111, which is dominated by the motor current of the electric motor 101. The controller usually has a current consumption of less than 200 mA. Low-ohmic resistors or Hall sensors can be used as measuring elements. With the help of an amplifier circuit and level adjustment, an analog signal proportional to the current can be provided to the computing unit 103 as a control for the handheld power tool 100.

The adjusting device 115 can be designed as a rotary potentiometer, a rocker switch or a double switching element. It is necessary that the threshold value required for the automatic function can be realized in an influencing manner by the user.

The computing unit 101 may include a microcontroller with conventional wiring (voltage regulator, clock source, electromagnetic compatibility (EMC) measures) and communication devices (Bluetooth, 4G, WLAN). The microcontroller may include an analog-to-digital converter and a digital interface in order to generate or detect signals of the trigger switch 109, the control device 115, the current measuring device 113, the supply voltage, the engine speed of the motor 101, and the control signal. The microcontroller (not explicitly shown in FIG. 1 ) may implement the state determination module 107 in order to implement the method according to the present invention to control the handheld power tool 100.

The handheld power tool 100 can be designed as a screwdriver, a rotary impact screwdriver, or a simple battery screwdriver. Alternatively, the handheld power tool 100 can be designed as an electric drill, an impact drill, an electric hammer, or a drilling tool.

FIG. 2 shows a further schematic illustration of handheld power tool 100 , in which individual functional sequences of handheld power tool 100 are explained.

2 shows a further schematic illustration of an embodiment of a handheld power tool 100 according to the present invention. The handheld power tool 100 comprises a motor 101 , a computing unit 105 arranged on a printed circuit board 127 , an energy source 109 and a tool 117 .

The illustrated diagram shows some of the operating principles of the handheld power tool 100. Some of the components of the handheld power tool 100 from FIG. 1 are shown, while other components or parts are not shown in order to keep the illustrated diagram as simple as possible. However, the illustrated handheld power tool 100 can include all the components shown in FIG. 1.

As already described above, the computing unit 105 can be constructed as a microcontroller or can include such a microcontroller, and the state determination module 107 shown in Figure 1 is installed on the computing unit according to the present invention. In addition to the computing unit, a speed sensor 129 (with the help of which the engine speed of the engine 101 can be measured) and a vibration sensor 131 (with the help of which the vibration of the handheld machine tool 100 can be measured) and an inverter 133 are also installed on the circuit board 127.

The user of the handheld power tool 100 can transmit a user input 139 to the computing unit 105 via the trigger switch 109 by means of an electrical signal transmission 147. The power of the handheld power tool 100 can be adjusted via the user input 139, which includes, for example, a trigger level of the trigger switch 109. In addition, the user input 139 can define a rotation direction of the motor 101, which defines, for example, a screwing direction or a drilling direction, or specify an operating mode of the handheld power tool 100, which describes a screwing process with or without an impact function.

Based on the user input 139, the computing unit 105 controls the control of the handheld power tool 100 and outputs corresponding control signals to the inverter 133 via an electrical signal transmission 147. The inverter 133 outputs a corresponding electrical energy transmission 141 to the motor 101. The motor 101 realizes a force transmission 135 to the tool 117 via a corresponding force-torque transmission 143, by means of which a workpiece 137 can be machined.

According to the invention, the method for controlling the handheld power tool 100, which is implemented by the computing unit 105, uses measured values of operating variables, based on which the operating state of the handheld power tool 100 is ascertained. The operating variables may be, for example, the motor current, the motor power, and the speed or torque of the motor 101. In the embodiment shown, the motor speed of the motor 101 is taken into account as an operating variable, based on which the operating state of the handheld power tool 100 is determined by the method according to the invention. These are measured by the speed sensor 129 shown, which detects the movement 145 of the motor 101. Furthermore, in the embodiment shown, the vibration/movement 145 of the handheld power tool 100 or the tool 117 or the workpiece 137 is taken into account as an operating variable. Here, the vibration/movement is measured by the vibration sensor 131. The corresponding measurement signals are forwarded by the speed sensor 129 and the vibration sensor 131 to the computing unit 105 for further processing.

The vibration sensor 131 may be designed, for example, as an acceleration sensor.

According to the present invention, for controlling handheld power tool 100, state determination module 107 is implemented to analyze the measured rotational speed or the detected vibration and to ascertain the current operating state of handheld power tool 100. Based on the ascertained operating state, the control of handheld power tool 100 is adjusted accordingly.

For a more detailed explanation of the method according to the invention for controlling a handheld power tool 100 , reference is made to the following description of the figures.

The force transmission device 135 of the hand-held power tool 100 can be realized as a direct drive or via a gear mechanism.

Furthermore, the drive of the handheld power tool 100 , which is indicated only schematically in FIGS. 1 and 2 , can include different drive options such as, for example, a striker, a hammer mechanism or a chisel mechanism.

As already mentioned, the power of the engine 101 can be transmitted to the drive of the handheld power tool 100 via a transmission. The transmission can be designed, for example, as a control transmission with two, three or more gears. The transmission can be connected to a slip clutch, which in turn has a direct connection to the drive. An alternative solution without a slip clutch is also conceivable, in which the transmission is directly connected to the drive.

The vibration or movement measured by the vibration sensor 131 may include a movement of the handheld power tool 100, for example, triggered by a user of the handheld power tool 100. In addition, the movement or vibration may be caused by a motor, a transmission or a drive. Alternatively, the movement or vibration measured by the vibration sensor 131 may be caused by the movement of a bit on the screwdriver head or may be caused by the force applied by the screwdriver to the workpiece 137 to be processed.

FIG. 3 shows the time profile of an operating variable 119 of the handheld power tool 100 and of an angle of rotation α _rot of the motor 101 of the handheld power tool 100 .

Graph a) shows the time profile of an operating variable 119 of a handheld power tool 100. In the embodiment shown, operating variable 119 indicates the motor current of motor 101 of handheld power tool 100. In the embodiment shown, handheld power tool 100 is designed as a screwdriver, and the shown profile of operating variable 119 shows the time profile of the motor current during screwing, in which a self-tapping screw is screwed into a workpiece 137 made of wood or a similar material.

The temporal profile of operating variable 119 shows a time series 123 consisting of a plurality of time-sequential measured values 121 of the motor current. Measured values 121 are recorded by corresponding current sensors within handheld power tool 100 during operation of handheld power tool 100 (i.e., when screwing a screw into a workpiece).

Here, FIG. a) illustrates a typical course of the motor current I when a self-drilling screw is driven into wood or similar material. The motor current I describes the torque output by the motor 101 via a torque constant in Newton meters per ampere. The torque generated by the starting current is applied to accelerate the rotor of the electric motor 101 at the beginning of the tightening operation. This results in a peak value of the motor current I at the beginning of the time series 123. Immediately after the start of the rotor, the number of revolutions is largely constant, which results in an almost horizontal course of the motor current I.

In a region, the initial peak value of the motor current I and its almost horizontal course are detected, which is characterized in FIG. a) by operating state A. In this operating state A, a large part of more than 90% of the applied torque is converted into actual mechanical work on the screw or the workpiece, and the screw is screwed into the workpiece accordingly.

In diagram a), two event times 125 , 126 are also marked at which the transition between operating state A firstly into operating state B and then into further operating state C takes place.

In the region of event time 126 , the motor current I in operating state A shows a uniform increase. This is due to the further screwing of the screw into the workpiece, for which a greater and greater torque is required with increasing screwing depth, and an increased motor current I is required to generate this greater and greater torque.

Starting from the event time 126, a steeper rise in the motor current I occurs compared to the relatively gently increasing rise in operating state A. This increasingly steep rise in the motor current I is caused by the conical screw head of the self-tapping screw lying flat on the surface of the workpiece to be machined. At the time of the event time 126, the conical screw head of the self-tapping screw hits the surface of the workpiece. Operating state B is therefore characterized by screwing the conical screw head into the workpiece. Due to the conical shape of the screw head, an increased torque is required to screw the screw head into the workpiece, which leads to a steep rise in the motor current I.

In contrast, event time 125 describes the transition from operating state B, in which the screwing of the conical screw head into the workpiece is described, to operating state C, in which the screw head is completely screwed into the workpiece. In this case, event time 125 describes the point at which the screw head is flush with the surface of the workpiece. Furthermore, this point in time is also referred to as the flush point. The point in time marked by event time 126, at which the conical screw head comes into contact with the surface of the workpiece, is also referred to as the pre-flush point.

In the embodiment shown, a flattened profile of the motor current I is shown in operating state C. This can occur due to shearing of the head or damage to the material of the workpiece when the screw is tightened further beyond the flush point.

Figure b) shows the time variation of the rotation angle α _rot . The time variation shows the signal of the rotation angle transmitter by way of example. In the example shown, the sensor has a single-value range of 360°. This is not necessary; a smaller single-value range is also sufficient. Usually, the single-value range of the sensor can be coupled with the number of revolutions of the engine 101. The absolute rotation angle information about the complete tightening process from the beginning (by the user operating the trigger switch 109) to the desired shutdown can be obtained based on the rotation angle sensor signal by means of so-called phase expansion. For this purpose, it can be known with the help of knowledge about the continuous rotation of the engine 101 that the transition from positive 180° to, for example, negative 179° actually corresponds to a rotation angle of positive 181°.

In diagram b), the rotation time t _rot is also marked. At the rotation time t _rot , a complete rotation of the motor 101 occurs. The knowledge of the rotation angle α _rot or the rotation angle α _rot of the motor 101 can then be used to predict the event time 125, 126. When the event time 126 described above is reached, at which the conical screw head lies flat on the workpiece surface, the event time 125 can be predicted with knowledge of the rotation angle, at which the screw is flush with the surface of the workpiece. Diagram b) also shows the rotation angle difference Δ _rot as the difference between the rotation angles α _rot at the event times 125, 126.

The situation shown in FIG. 3 is intended only to illustrate a possible application of the method according to the invention by way of example. Other applications in which the handheld power tool 100 is not designed as a screwdriver but is designed, for example, as a drill or chisel or jigsaw are also intended to be covered by the invention. More or fewer and differently designed operating states (A, B, C) than those described in the present example are also conceivable.

For example, the handheld power tool 100 can be configured as a screwdriver. Here, the possible operating states A, B, C can describe the screwing depth of the screw to be screwed in, the tightening in a certain tightening mode (impact mode, screwing in/out), the operating efficiency, the defined torque, the selection of the screw used or the recoil in the case of high torque.

Furthermore, the handheld power tool 100 can be designed as a drilling machine. Here, the operating states A, B, C can include the drilling mode (impact drilling, etc.), the orientation of the handheld power tool 100 relative to the surface of the workpiece to be processed, the type of insert tool (metal drill, wood drill, stone drill), the drilling depth, the orientation of the handheld power tool 100 relative to a predetermined drilling position, the drilling of cables or pipes, the demolding of the drill hole, the occurrence of drilling dust, or the recoil when the handheld power tool 100 is tightened suddenly.

Furthermore, the handheld power tool 100 can be designed as a percussion drill or an electric hammer. The operating states A, B, C can indicate the operating mode, the progress of the work, the risk of damage, vibrations and noise or the lack of contact force.

Furthermore, the handheld power tool 100 may be designed as a chisel. The possible operating states A, B, C may indicate an incorrect insertion tool, an incorrect grip of the handheld power tool 100 or a work process.

Furthermore, independently of the design of the hand-held power tool 100, possible operating states A, B, C for work safety/comfort can include, for example, vibration/noise generation, vibration monitoring, torque control with respect to a maximum torque, simple control of the hand-held power tool 100, correct clamping of the tool in the hand-held power tool 100 or recognition of situations such as a fall from a ladder, drilling a hole in a cable or pipe.

By correspondingly training the artificial intelligence of the state determination module 107 , the listed operating states A, B, C or similar operating states which may occur during the operation of the handheld machine tool 100 can be determined based on the measured values 121 of the operating variables 119 during the operation of the handheld machine tool 100 .

In this case, operating variable 119 may include, for example, motor current I or a motor position angle, a motor rotational speed, a voltage of a voltage source, movements and/or vibrations within handheld power tool 100 or similar measurable parameters of handheld power tool 100 .

FIG. 4 shows a schematic illustration of a portable power tool 100 according to one specific embodiment, wherein the portable power tool 100 is shown in different operating states A, B, C.

FIG. 4 illustrates the operating states A, B, C of the handheld power tool 100 illustrated in FIG. 3 in a diagrammatic representation. The handheld power tool 100 is again designed as a screwdriver and the operating states A, B, C depict such a tightening process in which a screw 169 is screwed into a workpiece 137. The screw 169 can be, for example, a self-tapping screw, and the workpiece 137 can be made of wood. For this purpose, the handheld power tool 100 also has a screw drill 168.

FIG. a ) shows operating state A in which the screw 169 is screwed into the workpiece 137 .

FIG. b ) depicts operating state B, which is characterized in that the conical screw head 170 abuts against the surface 167 of the workpiece 137 and is screwed into the workpiece 137 in the further process.

FIG. c ) depicts operating state C, which is characterized in that the screw 169 is flush with the surface 167 of the workpiece 137 and has thus reached the flush point.

As explained above, diagram a) of FIG. 3 shows the time course of the motor current of motor 101 of handheld power tool 100 in different operating states A, B, C during screwing of screw 169 into workpiece 137 .

In operating state A, in which screw 169 is screwed uniformly into workpiece 137 , motor current I exhibits a largely horizontal profile which has a slight gradient with increasing screw-in depth.

In operating state B (which is characterized by the fact that the conical screw head 170 is adjacent to the surface 167 of the workpiece 137 and the screw head 170 is screwed into the workpiece 137), the motor current I shows a steep rise, which is due to the increased torque that is required to screw the conical screw head 170 into the workpiece 137.

Operating state C is characterized in that screw head 170 is flush with surface 167 , whereas FIG. 3 shows a flattened profile of motor current I, which may be due to damage to the screw or to workpiece 137 .

4, the handheld power tool 100 also has an external sensor 171. In the embodiment shown, the external sensor 171 is designed as a camera sensor and enables the recording of camera data, with the aid of which the tightening process of the screw 169 into the workpiece 137 can be depicted. As will be explained in more detail in the following figures, the corresponding camera data can be used to train the state determination module 107 or the artificial intelligence 149 of the state determination module 107.

FIG. 5 shows a schematic illustration of a drive controller 197 of a handheld power tool 100 according to one specific embodiment.

FIG. 5 shows a control chain of the handheld power tool 100 .

According to the invention, the control chain comprises an inner control loop 193 and an outer control loop 191. In this case, inner control loop 193 is used to control a drive of handheld power tool 100 based on a user input 173 of a user.

In order to adjust the drive of the handheld power tool 100 based on the user input 173 of the user only via the internal adjustment ring 193, the user first performs the user input 173. This can be performed, for example, by operating the trigger switch 109. In this regard, the first target value of the control parameter of the handheld power tool 100 can be defined by the user input of the user. Here, the control parameter can include, for example, the engine speed, the engine power, the torque defined by the engine speed. Alternatively, the control parameter can include the direction of rotation (such as, for example, when screwing in or out a screw) or the operating mode (such as, for example, the impact mode or the hammer mode). Here, the user input 173 indicates the value of the control parameter.

An actual value of a control parameter, which describes the actual state of an actuator 195 of a drive controller 197 of the handheld power tool 100 , is recorded via a sensor measurement 175 of the operating variable 119 .

Sensor measurements 175 of operating variables 119 may include measurements of engine current, engine speed, engine power, vibrations of the engine or handheld power tool 100 or other relevant operating variables, using which operating states A, B, C are determined.

The first target value 173 input by the user and the actual value of the sensor measurement 175 of the control variable are transmitted to the internal control loop 193 by means of digital signal preprocessing 177 .

Corresponding control signals are output via an internal control circuit 193 to an actuator 195 in order to control the handheld power tool 100 .

The outer control loop 191 now serves to take into account the operating states A, B, C in which the handheld power tool 100 is in operation when controlling the handheld power tool 100 .

For this purpose, after digital signal preprocessing 197, a model inference 183 is performed on the first target value of the control parameter input 173 by the user and in particular on the actual value of the sensor measurement 175 of the control parameter. The previously described embodiment of the state determination module 107 plays a role in the model inference 183. In this case, the state determination module 107 is configured to recognize the operating state A, B, C in which the handheld power tool 100 is located based on the information sensor measurement 175 or the corresponding sensor data of the control parameter. Alternatively or additionally, the state determination module 107 can be configured to predict the event time 125, 126 at which the transition between the different operating states A, B, C of the handheld power tool 100 occurs based on the sensor data of the sensor measurement 175 of the operating variable 119.

State determination module 107 implemented in model reasoning 183 may be designed as a correspondingly trained artificial intelligence that is trained to ascertain operating states A, B, C or to predict event times 125 , 126 based on measured values of operating variables 119 .

The information ascertained by state determination module 107 in the form of model reasoning 183 about the present operating states A, B, C or predicted event times 125 , 126 is made available to an external control loop 191 after further processing 187 .

Therefore, the external control loop 191 is provided for defining corresponding second target values for the control parameters based on the information of the model reasoning 183 about the existing operating states A, B, C or the predicted event times 125, 126. In this case, the second target values for the control parameters defined by the external control loop 191 are coordinated with the respectively existing operating states A, B, C or the respectively predicted event times 125, 126. Taking into account the second target values of the control parameters generated by the external control loop 191, the control of the portable power tool 100 can be optimally adapted to the respectively existing operating states A, B, C or the respectively predicted event times 125, 126.

The second target value for the control variable generated by control loop 191 is then provided to inner control loop 193 .

According to the present invention, the internal control loop 193 is now configured to calculate an output target value, taking into account a first target value for a control parameter provided by the user in the user input 173 and a second target value for a control parameter provided by the external control loop 191 taking into account the existing operating state A, B, C or the predicted event time point 125, 126, and to control the actuator 195 of the hand-held machine tool 100 based on the output target value.

Here, the output target value can be calculated by the inner control loop 193, for example, as the product of the first target value of the user input 173 and the second target value of the outer control loop 191. Thus, by means of the product of the first and second target values, the first target value of the user input 173 can be sensitized by the second target value of the outer control loop 191 (which is determined in relation to the existing operating state A, B, C or the predicted event time 125, 126), i.e. adjusted according to the respective operating state A, B, C or the expected event time 125, 126. Alternatively, the output target value can be defined as the minimum and maximum of the first and second target values. Thus, the output target can be defined as the value of the first and second target values that best matches the respectively determined operating state A, B, C.

Alternatively, the output target value can be defined by the first target value of the user input 173 when the second target value of the external control loop 191 is less than a predefined threshold value, and the output target value can be defined as the predefined target value when the second target value is greater than or equal to the predefined threshold value. The predefined target value can be given as a constant value of a control parameter, which is set, for example, in the presetting of the portable power tool 100 as a function of the respectively existing operating state A, B, C or the previous event time 125, 126. The predefined threshold value can be adapted, for example, empirically to the respectively occurring operating state A, B, C and to any second target value by corresponding measurements.

Alternatively, in the case where the second target value of the external control loop 191 is less than a predefined threshold value, the output target value can be defined as the product of the first target value of the user input 173 and the predefined first target value, and in the case where the second target value of the external control loop 191 is greater than or equal to the predefined threshold value, the output target value can be defined as the product of the first target value of the user input 173 and the predefined second target value. Thus, depending on the second target value determined by the external control loop 191, the first target value of the user input 173 can be sensitized or adjusted in the form of predefined first and second target values (the predefined first and second target values are respectively defined as constant values of the control parameter and can be adjusted according to the respectively existing operating state or the expected event time point) according to the respectively existing operating state A, B, C or the expected event time point 125, 126. The predefined first and second target values and the predefined threshold value can in turn be determined empirically for the possible operating states A, B, C.

Therefore, taking into account the user input 173 made by the user of the hand-held power tool 100 and the corresponding first target value of the control parameter and taking into account the second target value of the control parameter determined in relation to the existing operating state A, B, C or the expected event time point 125, 126, the hand-held power tool 100 can be controlled in the form of an output target value determined by the internal control loop 193.

In the embodiment of Figures 3 and 4, the hand-held power tool 100 is constructed as a screw machine, and in this embodiment, a tightening situation is described: in the tightening situation, a self-tapping screw 169 is screwed into a wooden workpiece 137, and in this embodiment, different operating states A, B, and C depict: in operating state A, the screw 169 is screwed into the workpiece 137, in operating state B, the conical screw head 170 is screwed into the workpiece 137, and in operating state C, the screw head 170 is flush with the surface 167 of the workpiece 137. In this embodiment, the control parameter can be, for example, the number of revolutions of the engine 101 of the hand-held power tool 100.

User input 173 may include, for example, a signal of trigger switch 109 actuated by a user, by means of which a respective first target value for the engine speed is defined.

Operating variable 119 ascertained in sensor measurement 175 may be given, for example, by motor current I of motor 101 of portable power tool 100 .

The state determination module 107, which uses the model reasoning 183 to the measured values of the motor current I, can be correspondingly configured to determine different operating states A, B, C based on the motor current I according to diagram A of FIG3 . In addition, the state determination module 107 can be configured to predict corresponding event times 125 , 126 based on the measured values 121 .

The process described above and in particular the determination of the second target value by the external control loop 191 or the determination of the output target value by the internal control loop 193 are explained below in an application scenario in which, according to the embodiment in Figures 3 and 4, the hand-held machine tool 100 is constructed as a screw machine and the screw 169 is screwed into the workpiece 137 by the screw machine.

The user input 173 based on the actuation of the trigger switch 109 can define a correspondingly high value of the engine speed, for example, as a first target value. On the other hand, the state determination module 107 determines based on the sensor measurement 175 of the engine current that the handheld power tool 100 is already in the operating state C of FIG. 3 , in which the flush point has been reached and the screw head 170 is flush with the surface 167 of the workpiece 137 . Based on this, the outer control circuit 191 calculates a much lower value for the engine speed as the second target value in order to prevent damage to the screw or the workpiece 137 , which may be a concern in the case of a high engine speed of the user input 173 . Thus, by determining the output target value by the inner control circuit 193 , a clearly too high target value for the engine speed of the user input 173 can be adjusted downward by the much lower second target value for the engine speed calculated by the outer control circuit 193 , in order to control the handheld power tool 100 accordingly to the existing operating state and, if necessary, to switch it off.

For this purpose, the second target value and the predefined first and second target values can assume the value 0. The output target value can likewise be adjusted downward to the value 0, whereby the handheld power tool 100 can be stopped.

In the specific embodiment shown, the information of model reasoning 183 can also be displayed in a status display 189 .

Furthermore, a reset operation can be performed on the model reasoning, in which the model implementation is reset to initial values. This can be performed, for example, during individual tightening processes. For example, the state determination module 107 can be reset for each new tightening process (in each new tightening process, a single screw is screwed into or out of the workpiece 137). Alternatively or additionally, a reset can also be performed when the handheld power tool 100 is switched on or off.

For this purpose, a reset preprocessing 181 is first performed based on the sensor measurement 175 and based on this a reset decision 185 is made. Furthermore, the reset decision 185 can be made taking into account the results of the model reasoning 183 .

The information from model reasoning 183 and reprocessing 187 may be provided in digital or quasi-analog form.

FIG. 6 shows a diagrammatic representation of a system for selecting data for training artificial intelligence 149 according to one specific embodiment.

In the embodiment shown, the system for selecting a data set includes a computing unit 199. A plurality of model structures 150, 152, 154 are installed on the computing unit 199. The model structure 154 is provided for determining operating variables A, B, C of the handheld machine tool 100 or predicting event time points 125, 126 based on the measured values 121 of the operating variables 119 of the handheld machine tool. In the embodiment shown, the model structure is constructed as artificial intelligence 144, 146, 148. In particular, the artificial intelligence can be constructed as an artificial neural network with different structures or as a support vector machine or a decision tree.

Here, the operating variable 119 may be, for example, a motor current, a motor position angle, a motor speed, a voltage of a voltage source of the portable power tool 100, or a similar physically measurable variable. Here, the operating states A, B, C may be a load range in which the portable power tool is operated, the intensity of vibrations in the portable power tool and/or on the machined workpiece 137 and/or on a user of the portable power tool 100, the temperature in the portable power tool 100 and/or on the workpiece 137, an operating state in which the portable power tool 100 is operated, the progress of the work of the portable power tool 100, the material of the workpiece 137, the presence of a force connection between the portable power tool 100 and the workpiece 137, or a similar state of the portable power tool 100.

In order to select data sets 401, 403, 405 of training data set 400 for training artificial intelligence 149, according to the present invention, a plurality of data sets 401, 403, 405 of measured values 121 of operating variables 119 are received by computing unit 199. Data sets 401, 403, 405 may be, for example, measured values of operating variables 119 of one handheld machine tool 100 or of a plurality of handheld machine tools 100. Data sets 401, 403, 405 may be provided, for example, in the form of queue data from handheld machine tool 100.

Alternatively, the plurality of data sets 401, 403, 405 may be from a subset E of the comprehensive data set 407. The comprehensive data set 407 may be constructed as a data lake of a cloud storage, for example.

Therefore, for the purpose of data set selection, different model structures 150, 152, 154 are applied to data sets 401, 403 of the plurality of data sets. By applying the model structures 150, 152, 154, the operating states A, B, C of the handheld power tool 100 and/or the predicted event times 125, 126 are determined by the model structures 150, 152, 154 based on their training based on the corresponding measured values 121 of the operating variables 119. Here, the model structures 150, 152, 154 are respectively implemented on individual data sets 401, 403, 405. Therefore, each model structure 150, 152, 154 outputs a corresponding determination result in the form of the determined operating state A, B, C or a prediction result in the form of the predicted event time 125, 126 for each data set 401, 403, 405. Here, the results of obtaining a single model structure 150, 152, 154 may also include the following result: one or more model structures 150, 152, 154 cannot obtain a single value of the operating state A, B, C. Similarly, the prediction results of a single model structure 150, 152, 154 may include: one or more model structures cannot predict any single value of the event time point 125, 126.

In order to select data sets 401, 403, 405, different model structures 150, 152, 154 are now applied to the corresponding data sets 401, 403, 405, and the ascertainment or prediction results of the individual model structures 150, 152, 154 are compared with each other. Then, data sets 401, 403, 405 are selected for training data set 400, and model structures 150, 152, 154 have generated different prediction results or ascertainment results for the training data sets 401, 403, 405. Here, the deviation of the different ascertainment or prediction results of multiple model structures 150, 152, 154 can be calculated, for example, by the variance of the individual ascertainment or prediction results.

To generate the training data set 400, one or more of the received data sets 401, 403, 405 are then selected for which the implementation of the model structure 150, 152, 154 leads to an ascertainment result or a prediction result having a result deviation that reaches or exceeds a predefined limit value. Therefore, the ascertainment result or the prediction result has a variance that reaches or exceeds a predefined limit value.

Next, the selected data sets 401, 403, 405 are merged into the training data set 400. In addition to the selected data sets 401, 403, 405, the training data set 400 may also have further data sets.

According to one embodiment, retraining 409 of the model structures 150 , 152 , 154 on the newly generated training data set 400 including the selected data set 403 is then performed.

Next, the newly trained model structures 150 , 152 , 154 may be applied to newly received data sets 401 , 403 , 405 in order to again select data sets for expansion of the training data set 400 .

For this purpose, new data sets 401, 403, 405 of handheld power tool 100 may be received. Alternatively or additionally, further data sets 401, 403, 405 of comprehensive data set 407 may be selected. In this case, data sets 401, 403, 405 may be selected, for example, from a further subset D. This may be continued until all data sets 401, 403, 405 of comprehensive data set 407 have been checked by model structure 150, 152, 154 and the corresponding data set 403 has been selected for training data set 400.

In this case, data sets 401 , 403 , 405 may include labeled and/or unlabeled measured values 121 of operating variable 119 .

In the embodiment shown, exactly one data set 403 is selected which, when implementing the model structures 150, 152, 154, leads to an ascertained result (i.e., an ascertained operating state A, B, C) and/or a predicted result (i.e., a predicted event time 125, 126) of a result deviation that is greater than or equal to a predetermined limit value. However, this is not intended to limit the invention.

The present invention is based on the idea of selecting data sets 401, 403, 405 that lead to different results between the model structures 150, 152, 154 by implementing a plurality of model structures 150, 152, 154. In this regard, data sets 401, 403, 405 that may lead to contradictory prediction results/determination results may be considered in the training data set. By accepting these unclear or difficult to analyze data sets in the training data set 400, the training of artificial intelligence can be improved.

FIG. 7 shows a flow chart of a method 200 for selecting data for training artificial intelligence 149 according to one specific embodiment.

In order to select data for a training data set, a plurality of data sets 401, 403, 405 of measured values 121 of at least one operating variable 119 of a handheld power tool 100 are first received in a first method step 201. In this case, data sets 401, 403, 405 may be based on a plurality of measurements of operating variables 119 of a handheld power tool 100 or of a plurality of handheld power tools 100 of the same type.

In a further method step 203, a plurality of different model structures 150, 152, 154 are implemented on a plurality of data sets 401, 403, 405. Model structures 150, 152, 154 are directed to ascertaining operating states A, B, C of handheld power tool 100 and/or predicting event times 125, 126 based on measured values 121 of operating variables 119.

In a further method step 205 , result deviations between the ascertainment results and/or prediction results of a plurality of model structures 150 , 152 , 154 are ascertained.

In a further method step 207 , at least one data set 403 is selected which, when executing a plurality of model structures 150 , 152 , 154 , leads to an ascertainment result and/or a prediction result having a result deviation which reaches or exceeds a predefined limit value.

In a further method step 209 , model structures 150 , 152 , 154 are retrained taking into account selected data set 403 .

FIG. 8 shows a flow chart of a method 300 for generating a training data set 400 for training an artificial intelligence 149 according to one embodiment.

In order to generate training data set 400 for training artificial intelligence 149 , in a first method step 301 , method 200 according to one of the above-described specific embodiments for selecting data for training artificial intelligence 149 for controlling handheld power tool 100 is firstly carried out.

In a further method step 303 , the selected data sets are merged into a training data set 400 . This merging may include adding the selected data set to an already existing training data set 400 .

The artificial intelligences 144, 146, 148 of the model structures 150, 152, 154 can be constructed identically to the artificial intelligence 149 of the handheld power tool 100. However, according to the present invention, the artificial intelligences 144, 146, 148 can be different instances of artificial intelligence. However, after the training or retraining 409 is completed, the artificial intelligences 144, 146, 148 can also be used to determine the operating states A, B, C or predict the event times 125, 126 in the handheld power tool 100 according to the above-described embodiments.

FIG. 9 shows a schematic illustration of an artificial intelligence 149 which is provided to be used in controlling the handheld power tool 100 .

State determination module 107 may include a correspondingly trained artificial intelligence 149, which is trained to determine the current operating state A, B, C or predict event times 125, 126 based on the measured values of operating variable 119. State determination module 107 may also be designed to determine a second target value or to output a target value.

FIG. 9 shows an embodiment of such an artificial intelligence 149 , which can be used to determine operating states A, B, C or to predict event times 125 , 126 .

In the specific embodiment shown, artificial intelligence 149 is designed as an artificial neural network and in particular as a long short-term memory (LSTM) network.

In the specific embodiment shown, the artificial neural network comprises an input layer 153 for receiving input data 151. The input data may comprise sensor data of the operating variable 119 in a correspondingly preprocessed form.

Furthermore, the artificial neural network comprises two dense layers 155 and two pooling layers 157, which are arranged one after the other in an alternating manner. Furthermore, the artificial neural network comprises two long short-term memory layers 159, between which an information loss layer 161 is arranged. Finally, the artificial neural network further comprises two dense layers and an output layer 163.

In data processing, downsampling 164 is first performed by the input layer 153, the two first dense layers 155 and the two pooling layers 157. The subsequent two LSTM layers 159 and the information loss layer 161 in the middle implement feature extraction 165. The last two dense layers 155 and the output layer 163 implement prediction 166.

Unlike the embodiment shown, the artificial intelligence 149 used can also be structured in a different model architecture, which can perform a regression or classification based on the time series 123 of the measured values 121 of the operating variable 119. The model architecture of the artificial intelligence 149 used assumes that the corresponding model can be provided in a format that can be implemented on the microcontroller of the handheld power tool 100.

For the embodiments described here, a Tensorflow/Keras model with the architecture described below can be used. After training, the model can first be converted into a Tensorflow-Lite format, which can in turn be converted into C code for a microcontroller with the help of a TVM converter.

The three input channels may be, for example, the motor current I of the motor 101, the trigger voltage of the trigger switch 109, and the motor revolutions per second.

The architecture used can be constructed as follows:

The absolute number of parameters used: 7,801

Of these, trainable parameters: 7,801

Non-trainable parameters: 0

The first dense layer 155 may be configured to have a 6X8 kernel and 8 biases. The second dense layer 155 may be configured to have an 8X16 kernel and 16 biases. The first LSTM layer 159 may be configured to have a 16X128 kernel, a 32X128 loop kernel, and a 128 bias. The second LSTM layer 159 may be configured to have a 32X32 kernel, an 8X32 loop kernel, and a 32 bias. The third dense layer 155 may be configured to have an 8X4 kernel and 4 biases. The fourth dense layer 155 may be configured to have a 4X1 kernel and 1 bias.

The dense layer 155 and the LSTM layer 159 may be constructed with a TanH activation function.

10 shows a schematic illustration of a computer program product 500 which includes commands which, when the program is executed by a data processing unit, cause the data processing unit to implement a method 200 for selecting data for training an artificial intelligence 149 for controlling a handheld machine tool 100 and/or a method 300 for generating a training data set 400 according to any of the above embodiments.

In the embodiment shown, computer program product 500 is stored on a storage medium 501. In this case, storage medium 501 may be any desired storage medium known from the prior art.

Claims (14)

1. A method (200) for selecting data for training an artificial intelligence (149) for controlling a hand-held power tool (100), the method comprising:

Providing (201) a plurality of data sets (401, 403, 405) of measured values (121) of at least one operating parameter (119) of the hand-held power tool (100), wherein the data sets (401, 403, 405) are based on a plurality of measurements of the operating parameter (119) of the hand-held power tool (100) or of a plurality of hand-held power tools (100) of the same type;

-implementing (203) a plurality of different model structures (150, 152, 154) on the data set, which are provided for determining an operating state (a, B, C) of the hand-held power tool (100) and/or for predicting an event time point (125, 126) at which the hand-held power tool (100) changes from one operating state (a, B, C) into another operating state (a, B, C), and

Outputting the result of the determination and/or the prediction from a plurality of model structures (150, 152, 154);

-determining (205) a result deviation between the determination results and/or the prediction results of the plurality of model structures (150, 152, 154); and

At least one dataset (403) is selected (207), which, when the plurality of model structures (150, 152, 154) are implemented, already results in a determination and/or prediction result with a result deviation that meets or exceeds a predefined limit value.

2. The method (200) of claim 1, wherein the model structure (150, 152, 154) is configured as a data-based model and/or as artificial intelligence (144, 146, 148).

3. The method (200) of claim 2, wherein the artificial intelligence (144, 146, 148) is configured as a neural network having different structures and/or support vector machines and/or decision trees.

4. The method (200) of any of the preceding claims, wherein the resulting deviation is configured as a variance.

5. The method (200) according to any one of the preceding claims, wherein the dataset comprises marked and/or unmarked measured values (121) of the operating parameter (119).

6. The method (200) of any of the preceding claims, wherein the evaluating the results comprises: no operating states (a, B, C) are ascertained, and/or wherein the prediction result comprises: no event time points (125, 126) are predicted.

7. The method (200) according to any one of the preceding claims, further comprising:

retraining of a plurality of model structures (150, 152, 154) configured as artificial intelligence (144, 146, 148) is performed taking into account the selected dataset (403).

8. The method (200) according to any one of the preceding claims, wherein the plurality of data sets (401, 403, 405) is a subset of a comprehensive data set (405) of measured values (121) of the operating parameter (191), and wherein for the providing (201) the data sets (401, 403, 405) are selected according to a selection criterion.

9. The method (200) according to any one of the preceding claims, wherein, for providing (201) the plurality of data sets (401, 403, 405), the data sets (401, 403, 405) are received from one hand-held power tool (100) or a plurality of hand-held power tools (100) of the same type.

10. The method (200) according to any one of the preceding claims, wherein the operating parameters (119) comprise one or more of the following list: the motor current, the motor position angle, the motor rotational speed, the voltage of the voltage source of the hand-held power tool (100), and/or wherein the operating state (A, B, C) comprises one or more from the list of: the load range in which the hand-held power tool (100) is operated, the intensity of vibrations in the hand-held power tool (100) and/or on a processed workpiece (137) and/or in a user of the hand-held power tool (100), the temperature in the hand-held power tool (100) and/or on the workpiece (137), the operating mode in which the hand-held power tool (100) is operated, the operating schedule of the hand-held power tool (100), the material of the workpiece (137), the presence of a force fit between the hand-held power tool (100) and the workpiece (137).

11. A method (300) for generating a training dataset (400) for training artificial intelligence (149), the method comprising:

Implementing (301) a method (200) according to any one of the preceding claims 1 to 10 for selecting data for training an artificial intelligence (149) for controlling a hand-held power tool (100);

The selected data sets (403) are combined (303) into a training data set (400).

12. Training data set (400) for training an artificial intelligence (149) for controlling a hand-held power tool (100), wherein the training data set (400) is generated in accordance with the method (300) for generating the training data set (400) for training the artificial intelligence (149) according to claim 11.

13. -A computing unit (199) configured to implement the method (200) for selecting data for training an artificial intelligence (149) for controlling a hand-held power tool (100) according to any one of the preceding claims 1 to 10, and/or to implement the method (300) for generating a training data set (400) according to claim 11.

14. Computer program product (500) comprising instructions which, when the program is executed by a data processing unit, cause the data processing unit to execute the method (200) for selecting data for training an artificial intelligence (149) for controlling a hand-held power tool (100) according to any of the preceding claims 1 to 10 and/or to execute the method (300) for generating a training data set (400) according to claim 11.