CN115186370A - Engineering forklift transfer learning system based on deep learning training model - Google Patents

Abstract

The invention discloses an engineering forklift migration learning system based on a deep learning training model. Each processing module; the data collection module is used to collect various data information of the engineering forklift; the model building module is used to build a deep learning training model; the training learning module uses the deep learning training model built by the model building module, combined with the data collection module. The collected data information of the forklift realizes the training and learning of the forklift; the invention compares and analyzes the working state and service life of the steel fork of the engineering forklift through the data analysis module, and obtains the data information of the service life of the engineering forklift and the optimal replacement time, which is convenient The steel fork should be replaced in time for the engineering forklift in actual use to ensure the safety of the engineering forklift in use.

Description

An engineering forklift transfer learning system based on deep learning training model

technical field

The invention relates to the technical field of migration learning systems, in particular to an engineering forklift migration learning system based on a deep learning training model.

Background technique

The development of the modern logistics industry has led to a rapid increase in the production of mobile cranes and transport machinery, and the most widely used forklifts in the field of mobile cranes and transport machinery. A forklift is a special vehicle that is equipped with forks and can lift loads to a target height. Sometimes forklifts are also classified as construction machinery. As a vehicle, forklifts and battery trucks, tractors, dump trucks, and AGV trolleys are equivalent to industrial vehicles or loading and unloading vehicles. It is widely used in factories, warehouses, ports, airports and other places to realize mechanized loading and unloading, stacking and short-distance handling, which greatly improves production efficiency and is an indispensable equipment in the modern logistics industry.

Because the existing engineering forklift does not have a transfer learning system, the engineering forklift cannot be maintained in time when it is in use, resulting in damage to the engineering forklift when it is in use, resulting in potential safety hazards.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an engineering forklift migration learning system based on a deep learning training model, through the data analysis module to carry out data comparative analysis on the working state and service life of the steel fork of the engineering forklift, and obtain the data information of the service life of the engineering forklift and the maximum service life of the engineering forklift. To optimize the replacement time, the output module outputs the data results, which facilitates the timely replacement of the steel fork in the actual use of the engineering forklift, and ensures the safety of the engineering forklift in use.

The object of the present invention can be realized through the following technical solutions:

An engineering forklift migration learning system based on a deep learning training model, comprising a central processing module, the central processing module is used to process the data information of the entire system, and sends instructions to each processing module for controlling each processing module of the system;

The data acquisition module is used to collect various data information of the engineering forklift;

The model building module is used to build a deep learning training model;

The training and learning module uses the deep learning training model built by the model building module, combined with the forklift data information collected by the data acquisition module, and combines the data information with the deep learning training model to realize the training and learning of the forklift.

As a further scheme of the present invention: a data analysis module is also included, which collects statistics on the learning data obtained by the training and learning module, analyzes the physical data information collected in the data acquisition module, and obtains the optimal data information actually used by the forklift through comparative analysis;

The output module outputs the data information structure of the system.

As a further scheme of the present invention: the data collection module includes physical data collection and simulation data collection, the physical data collection is the collection of various data information in the actual working process of the engineering forklift under the real environment, and the simulation data collection is the computer simulation environment. The collection of various data of forklift in the simulation state.

As a further scheme of the present invention: the data acquisition module collects the energy consumption data and energy recovery data of the engineering forklift, and is used for learning and training the energy recovery of the engineering forklift.

As a further solution of the present invention: the energy recovery of the engineering forklift includes the recovery of the potential energy of the working device of the forklift and the recovery of the braking energy of the traveling device of the forklift.

As a further solution of the present invention, the data collection module collects the service life data of the engineering forklift, and is used for learning and training the service life of the engineering forklift, so as to determine the replacement time of the wearing parts of the forklift.

As a further scheme of the present invention: the specific workflow of the transfer learning system is:

The first stage is the engineering forklift simulation training stage. The engineering forklift simulation software is used to simulate the computer simulation samples, and the computer simulation samples are used to train the preset deep learning training model, and the first stage deep learning training model is obtained;

The second stage is the physical sample training stage. The first stage deep learning training model and the preset deep learning training model are modeled, and the fused deep learning training model is trained with physical samples, and the second stage deep learning training model is obtained. , where the physical sample is the physical data information sample collected by the data acquisition module;

The third stage is the mixed sample training stage. The deep learning training model of the second stage is modeled with the preset deep learning training model, and the mixed samples are obtained by mixing computer simulation samples and physical samples to train the fused deep learning training model. The final deep learning training model is obtained, and the training process ends.

Beneficial effects of the present invention:

(1) The data analysis module is used to compare and analyze the working status and service life of the steel fork of the engineering forklift, and obtain the data information of the service life of the engineering forklift and the optimal replacement time. The output module outputs the data results, which is convenient for the engineering forklift in the actual Replace the steel fork in time to ensure the safety of the engineering forklift when in use.

(2) The energy-saving data information of the engineering forklift is obtained through the system, and an energy storage device is designed in the hydraulic system of the forklift to store the potential energy released during the descending process of the lifting device and use it when ascending, so as to improve the energy utilization efficiency, and at the same time To achieve the purpose of making the system run smoothly, work reliably and safely, the accumulator is used to recover the energy that should have been converted into heat energy and dissipated by the braking system, and when the forklift accelerates again, it is released to cooperate with the engine to drive the forklift. Realize energy recovery and reuse.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings.

Fig. 1 is the structural block diagram of the system of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Example 1

Referring to FIG. 1, the present invention is an engineering forklift migration learning system based on a deep learning training model, including a central processing module, which is used to process the data information of the entire system and send instructions to each processing module. , used for each processing module of the control system; the data acquisition module is used to collect various data information of the engineering forklift, wherein the data acquisition module includes physical data acquisition and simulation data acquisition, and the physical data acquisition is the actual work of the engineering forklift in the real environment. The collection of various data and information in the process, the collection of simulation data is computer simulation environment, the collection of various data of engineering forklifts in the simulation state, by comparing and integrating physical data information and simulation data information, so as to improve the learning of engineering forklifts Authenticity of training.

The model building module is used to build a deep learning training model. Through the deep learning training model, the forklift is trained and learned, so as to realize the integration and statistics of various data information of the forklift, and obtain the data information of the actual work of the engineering forklift.

The training and learning module uses the deep learning training model built by the model building module, combined with the forklift data information collected by the data acquisition module, and combines the data information with the deep learning training model to realize the training and learning of the forklift, and the simulation training of the forklift. Learning data.

The data analysis module makes statistics on the learning data obtained by the training and learning module, analyzes the physical data information collected in the data acquisition module, and obtains the optimal data information actually used by the forklift through comparative analysis.

The output module outputs the data information structure of the system.

The physical data collection collects the energy consumption data information of ordinary engineering forklifts in actual work and the energy recovery data information of engineering forklifts with energy recovery mechanisms, and sends the actual energy consumption and energy recovery data information to the central processing module through the data acquisition module. At the same time, the computer is used to build the engineering forklift simulation working environment, and the simulated energy loss data information and simulated energy recovery data information of the engineering forklift in the simulation environment are collected through the simulation data collection. The data acquisition module will simulate the energy loss data information and simulated energy recovery data. The data information is sent to the central processing module, and the model building module builds the corresponding deep learning training model according to the physical data information and simulation data information collected by the data acquisition module. The training learning module uses the deep learning training model to train and learn the engineering forklift. Through a large number of learning and training, the energy loss of the engineering forklift at work, and the best energy recovery data after the installation of the energy recovery mechanism are obtained. The data information of the energy recovery effect, and finally output the best energy recovery effect data information through the output module to complete the improvement of the actual engineering forklift.

The energy recovery of the forklift includes the recovery of the potential energy of the forklift working device and the recovery of the braking energy of the forklift traveling device.

The recovery of the working potential energy of the engineering forklift is as follows: when the steel fork rises, the power source provides hydraulic energy through the hydraulic pump to supply the goods to rise, which is converted into the gravitational potential energy of the goods; when it falls, the gravitational potential energy of the goods is converted into hydraulic energy. The energy storage device is designed in the hydraulic system of the forklift to store the potential energy released during the descending process of the lifting device and use it when ascending, so as to improve the energy utilization efficiency, and at the same time achieve the purpose of making the system run smoothly, reliably and safely.

The recovery of the walking braking energy by the engineering forklift is as follows: when the vehicle is braking, the energy is transferred in the reverse direction. Compared with the forward transfer from the engine to the wheels when the vehicle is running normally, the driving kinetic energy of the vehicle is transmitted through the wheels, transmission system, generator/wheel. The hydraulic motor is converted into electrical energy/hydraulic energy and stored in the accumulator, and is released to assist the engine output to do work during the subsequent driving. When the forklift is not emergency braking, the accumulator can be used to convert the brake system that should be used. The energy lost for heat energy is recovered, and released when the forklift accelerates again, it cooperates with the engine to drive the forklift, and realizes energy recovery and reuse.

Embodiment 2

This embodiment is used for learning and training the service life of the engineering forklift, so as to determine the replacement time of the forklift wearing parts and ensure the safety performance of the forklift during use. This embodiment takes the steel fork of the forklift as an example for description.

The physical data collection collects the working data of the steel fork and the corresponding service life and other data information when the engineering forklift is actually working, and sends the actual physical data information of the engineering forklift to the central processing module through the data acquisition module. The computer is based on the theory of the engineering forklift. The data builds a simulation working environment, collects the simulation data information of the engineering forklift and the service life information of the steel fork in the simulation environment through the simulation data collection, and sends the corresponding simulation data information to the central processing module, and the model building module collects the data according to the data. The physical data information and simulation data information collected by the module build a corresponding deep learning training model, and then the training learning module uses the deep learning training model to train and learn the engineering forklift, and obtain the steel fork when the engineering forklift is working through a large amount of learning and training. Through the data analysis module, the working status and service life of the steel fork of the engineering forklift are compared and analyzed, and the data information of the service life of the engineering forklift and the optimal replacement time are obtained. The output module outputs the data results, which is convenient for the engineering forklift in the In actual use, the steel fork should be replaced in time to ensure the safety of the engineering forklift in use.

For the replacement frequency of the service life of the steel fork, the specific collection and training process is as follows:

The first stage is the engineering forklift simulation training stage. The engineering forklift simulation software is used to simulate the computer simulation samples, and the computer simulation samples are used to train the preset deep learning training model, and the first stage deep learning training model is obtained;

The second stage is the physical sample training stage. The first stage of deep learning training model and the preset deep learning training model are modeled, and physical samples are used to train the fused deep learning training model, and the second stage deep learning training model is obtained. , where the physical sample is the physical data information sample collected by the data acquisition module;

The third stage is the mixed sample training stage. The deep learning training model of the second stage is modeled with the preset deep learning training model, and the mixed samples are obtained by mixing computer simulation samples and physical samples to train the fused deep learning training model. The final deep learning training model is obtained, and the training process ends.

In the process of training and learning, set the number of fork feeds of the engineering forklift as n, the weight of a single fork feed as ma, the total weight of fork feed as M, the critical weight of single fork feed breakage as mb, and the total service life of the steel fork. is S, and the replacement time of the steel fork is t.

Then M=n×ma

The service life of the steel fork is related to the total weight of the forklifted goods and the number of forks.

In the formula, K is the fixed coefficient of the service life of the steel fork of the engineering forklift, and the value of the coefficient is related to the material and specification of the steel fork.

In order to ensure the normal use of the steel fork, the weight of the steel fork in a single fork should be less than the breaking critical weight of the steel fork, that is,

ma≤g×mb

In the formula, g is the safety factor of the steel fork, and the value of g ranges from 0.5 to 0.9, and its specific value is related to the material and specification of the steel fork.

In order to improve the safety performance of the engineering forklift in use and prevent the steel fork from breaking and causing safety accidents during the use of the forklift, the replacement time t of the steel fork should be less than the total service life S of the steel fork, that is

t≤f×S

In the formula, f is the safety factor of the steel fork, and its specific value is related to the material and specification of the steel fork.

The invention also provides a migration learning method for a forklift, which divides the source domain and the target domain according to different migration learning tasks, constructs a migration learning network, and initializes network hyperparameters.

constructing the transfer learning network based on the feature extractor and the classifier;

The migration learning network provided by the embodiment of the present invention is composed of a feature extractor and a label classifier. The feature extractor is used to extract the features of the forklift input sample set, and the classifier is used to predict and classify the input forklift sample set.

Specifically, in order to improve the performance of the preset transfer learning network, the preset transfer learning network is pre-trained by using the labeled ImageNet dataset before performing the transfer learning. The sub-network formed by the ResNet-50 model serves as the feature extractor of the transfer learning network in the embodiment of the present invention, and the two fully connected layers following the feature extractor serve as the label classifier.

Input the respective data samples of the source domain and the target domain into the preset transfer learning network and forward propagation to obtain the network prediction label; use the stochastic gradient descent method to train the entire network, and use back propagation to update the network parameters until the model converges Or stop training when the maximum number of iterations is reached.

Based on a preset loss function, the transfer learning network is learned.

Specifically, in the training process of the transfer learning network, the knowledge of the source domain is transferred to the target domain, which is usually called transfer learning. For the training of deep transfer learning network in the traditional sense, a loss function is usually introduced, and the loss function measures the difference between the predicted value and the actual value; in the deep transfer learning process of the neural network, a loss function can also be introduced, in Measuring the difference between the predicted value and the true value also needs to be able to measure the effect of transfer learning.

However, the data in the source domain and the target domain obey different probability distributions, and only using the loss function in the traditional sense as the loss function in the transfer learning process cannot make the transfer learning achieve a good effect.

On the basis of the above embodiment, the preset loss function includes a source domain classification error rate loss function, a conditional entropy loss function of the target domain prediction output, and an entropy loss function of the target domain predicted category distribution. Transfer learning networks for learning, including:

The loss function of the classification error rate of the source domain, the conditional entropy loss function of the predicted output of the target domain, and the entropy loss function of the predicted category distribution of the target domain are determined to determine the loss function of the deep transfer learning network, and then update the parameters of the deep transfer learning network. The transfer learning network is adapted to the target domain, and a good classification effect can be achieved.

According to the above loss function, the objective function and optimization objective provided by the embodiment of the present invention can be constructed:

where θ represents the network parameters, S represents any batch of samples in the source domain sample set, T represents any batch of samples in the target domain sample set; L _s ( ) represents the source domain classification error rate loss function, and L _e ( ). ) represents the conditional entropy loss function of the predicted output of the target domain, L _d (·) represents the entropy loss function of the predicted category distribution of the target domain; λ and β are adjustable trade-off parameters.

It can be understood that the learning process is a process of continuously updating parameters. When the transfer learning network converges or reaches a preset number of learning times, the learning stops.

After the transfer learning process, a transfer learning network with better generalization performance can be obtained. After saving the final network model and training results, the unlabeled target domain sample set is introduced into the network model to obtain more accurate target domain sample set labels. The trained network can be used to predict unlabeled samples in the target domain, instead of manually labeling unknown data with high accuracy.

An embodiment of the present invention has been described in detail above, but the content is only a preferred embodiment of the present invention, and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still belong to the scope of the patent of the present invention.

Claims (7)

1. An engineering forklift transfer learning system based on a deep learning training model is characterized by comprising a central processing module, wherein the central processing module is used for processing data information of the whole system and sending instruction commands to each processing module for controlling each processing module of the system;

the data acquisition module is used for acquiring various data information of the engineering forklift;

the model building module is used for building a deep learning training model;

the training learning module utilizes the deep learning training model built by the model building module, combines the forklift data information collected by the data collection module, combines the data information and the deep learning training model, and realizes the training and learning of the forklift.

2. The engineering forklift transfer learning system based on the deep learning training model as claimed in claim 1, further comprising a data analysis module for counting the learning data obtained by the training learning module, analyzing the physical data information collected by the data collection module, and comparing and analyzing to obtain the optimal data information actually used by the forklift;

and the output module is used for outputting the data information structure of the system.

3. The deep learning training model-based engineering forklift migration learning system according to claim 2, wherein the data acquisition module comprises physical data acquisition and simulation data acquisition, the physical data acquisition is the acquisition of various data information of the engineering forklift in an actual working process under a real environment, and the simulation data acquisition is the acquisition of various data of the engineering forklift in a simulation state under a computer simulation environment.

4. The engineering forklift transfer learning system based on the deep learning training model as claimed in claim 3, wherein the data acquisition module is used for acquiring energy loss data and energy recovery data of the engineering forklift for learning and training energy recovery of the engineering forklift.

5. The engineering forklift transfer learning system based on the deep learning training model as claimed in claim 4, wherein the engineering forklift energy recovery comprises forklift working device potential energy recovery and forklift running device braking energy recovery.

6. The engineering forklift migration learning system based on the deep learning training model as claimed in claim 2, wherein the data acquisition module is used for acquiring service life data of the engineering forklift and carrying out learning training on the service life of the engineering forklift so as to determine the replacement time of the quick-wear part of the forklift.

7. The engineering forklift transfer learning system based on the deep learning training model according to any one of claims 1 to 6, characterized in that the transfer learning system has a specific work flow:

the first stage is an engineering forklift simulation training stage, a computer simulation sample is obtained by utilizing engineering forklift simulation software, and a preset deep learning training model is trained by adopting the computer simulation sample to obtain a first-stage deep learning training model;

the second stage is a physical sample training stage, the first stage deep learning training model and a preset deep learning training model are subjected to model fusion, and the fused deep learning training model is trained by adopting physical samples to obtain a second stage deep learning training model, wherein the physical samples are physical data information samples collected by a data collection module;

and the third stage is a mixed sample training stage, model fusion is carried out on the deep learning training model in the second stage and a preset deep learning training model, a mixed sample obtained by mixing a computer simulation sample and a physical sample is adopted to train the fused deep learning training model to obtain a final deep learning training model, and the training process is finished.

Images