CN120761345A - A smoke transmittance detector and its fault diagnosis method and system - Google Patents

Abstract

The present invention belongs to the field of fault diagnosis technology, and specifically relates to a flue gas transmittance detector, a fault diagnosis method, and a system therefor. The method comprises: S1, obtaining a suction process parameter for representing the suction process of the flue gas transmittance detector, an operating state parameter for representing the operating process of the flue gas transmittance detector, and the flue gas transmittance in each detection channel of the flue gas transmittance detector; S2, inputting the suction process parameter and the flue gas transmittance in each detection channel into a pre-trained transmittance prediction model in the form of time series data to obtain the future flue gas transmittance; S3, if the future flue gas transmittance output by the transmittance prediction model is abnormal, inputting the operating state parameter into a pre-trained fault diagnosis model to obtain the fault type of the flue gas transmittance detector. The present invention solves the technical problem of the complex fault diagnosis process of the flue gas transmittance detector in the prior art.

Description

Flue gas transmittance detector and fault diagnosis method and system thereof

Technical Field

The invention belongs to the technical field of fault diagnosis, and particularly relates to a smoke transmittance detector and a fault diagnosis method and system thereof.

Background

The smoke transmittance detector is a detection device for detecting smoke transmittance of cigarettes, is driven by a servo motor, simulates a suction curve of standard flow by driving the motion of a needle cylinder piston, and uses a digital laser sensor to emit light beams for smoke transmittance detection. Each channel is provided with a sensor at three positions of the needle cylinder, and aerosol transmittance at each position in each sucking process is detected respectively. The standard flow suction curve is simulated by the motion of the syringe piston, driven by a servo motor. The digital laser sensor is used for correlation beam to detect the transmission of smoke.

In the long-term use process of the smoke transmittance detector, various faults possibly occur due to environmental factors, mechanical abrasion or improper operation, and the accuracy of a detection result is affected. Such failures include, but are not limited to, light sensor anomalies, suction line blockage, lead screw motor failures, pneumatic element failures, clamp port damage, connection air path leaks, and the like. Diagnosis and treatment of such faults often require specialized technicians and complex detection equipment, which are time-consuming, labor-consuming, and costly, and do not allow for real-time fault detection. Therefore, the fault diagnosis of the existing smoke transmittance detector has the problems of complex process, time consumption and labor consumption.

Disclosure of Invention

The invention aims to provide a smoke transmittance detector and a fault diagnosis method and system thereof, which are used for solving the technical problem of repeated fault diagnosis process of the smoke transmittance detector in the prior art.

In order to solve the technical problems, the technical scheme of the fault diagnosis method of the smoke transmittance detector provided by the invention is that the fault diagnosis method of the smoke transmittance detector comprises the following steps:

S1, acquiring a suction process parameter for representing a suction process of a smoke transmittance detector, an operation state parameter for representing an operation process of the smoke transmittance detector and smoke transmittance in each detection channel of the smoke transmittance detector;

S2, inputting the parameters of the suction process and the transmittance of the smoke in each detection channel into a pre-trained transmittance prediction model in a time sequence data form to obtain the transmittance of the future smoke;

S3, if the transmittance of the future smoke output by the transmittance prediction model is abnormal, the operation state parameters and the transmittance of the smoke in each detection channel are input into a pre-trained fault diagnosis model, and the fault state type of the smoke transmittance detector is obtained.

The technical scheme has the beneficial effects that the technical scheme of the fault diagnosis method of the smoke transmittance detector belongs to the creation of improved inventions. According to the invention, fault diagnosis is carried out by a data driving method, the transmittance of each detection channel at the future time or for the future number of ports can be predicted according to the parameters of the suction process and the current transmittance of each channel, and fault diagnosis can be carried out according to the operation state parameters of the operation process of the smoke transmittance detector. The transmittance prediction model and the fault diagnosis model are trained in advance through a large amount of data so as to learn the relationship between the parameters of the pumping process and the transmittance, the operation state parameters and the fault types at the future time or for the future number of ports. And after judging whether the abnormality occurs or not through the transmittance prediction model, further determining the fault type through the fault diagnosis model. In the fault diagnosis process of the invention, the specific fault type can be determined by only acquiring various data of the smoke transmittance detector in the operation process without other professional detection equipment and tools. The invention solves the technical problem of complex fault diagnosis process of the smoke transmittance detector in the prior art.

Further, the aspiration process parameters include aspiration volume and aspiration compensation coefficients for modifying the aspiration process.

Further, the operating state parameter includes a flue gas temperature.

Further, the pumping process parameters further include one of diffusion time, pumping interval, diffusion multiple, trigger interval, trigger time, preheating time for preheating each detection channel, and channel temperature, or a combination of two or three or more.

The transmission prediction model comprises a transmission characteristic extraction module and a time attention module, wherein the transmission characteristic extraction module is used for extracting the pumping process parameters and time sequence characteristics in the transmission of the smoke in each detection channel through an LSTM, and the time attention module is used for identifying and weighting key time nodes in the time sequence characteristics output by the transmission characteristic extraction module so as to obtain the transmission of the smoke to be detected according to the weighted characteristics.

Further, the fault diagnosis model comprises a transducer network model.

Further, the fault state type comprises a normal state, detection of air leakage of the square tube, abnormal state of the clamping opening and abnormal thermocouple.

The invention also provides a technical scheme of the fault diagnosis system of the smoke transmittance detector, which comprises a processor, wherein the processor is used for executing a computer program to realize the steps of the fault diagnosis method of the smoke transmittance detector.

The invention also provides a technical scheme of the smoke transmittance detector, which comprises a fault diagnosis unit, wherein the fault diagnosis unit comprises a processor, and the processor is used for executing a computer program to realize the steps of the fault diagnosis method of the smoke transmittance detector.

Drawings

FIG. 1 is a schematic diagram showing the transmittance prediction in the fault diagnosis method of the smoke transmittance detector according to the present invention;

FIG. 2 is a schematic diagram of fault diagnosis in an embodiment of a fault diagnosis method of the smoke transmittance detector according to the present invention;

FIG. 3 is a schematic diagram of a network structure of MV-LSTM in an embodiment of a fault diagnosis method for a smoke transmittance detector according to the present invention;

fig. 4 is a confusion matrix corresponding to a transducer model in an embodiment of a fault diagnosis method of the smoke transmittance detector of the present invention.

Detailed Description

According to the invention, fault diagnosis is carried out by a data driving method, the transmittance of each detection channel at the future time or for the future number of ports can be predicted according to the parameters of the suction process and the current transmittance of each channel, and the fault diagnosis can be carried out by the operation state parameters of the operation process of the gas transmittance detector. The transmittance prediction model and the fault diagnosis model are trained in advance through a large amount of data so as to learn the relationship between the current transmittance of the pumping process parameters and each channel and the transmittance, the running state parameters and the fault type which are the transmittance of the future time or the future number of ports. And after judging whether the abnormality occurs or not through the transmittance prediction model, further determining the fault type through the fault diagnosis model. In the fault diagnosis process of the invention, the specific fault type can be determined by only acquiring various data of the smoke transmittance detector in the operation process without other professional detection equipment and tools. The invention solves the technical problem of complex fault diagnosis process of the smoke transmittance detector in the prior art.

The fault diagnosis method implementation mode of the smoke transmittance detector comprises the following steps:

A fault diagnosis method for a smoke transmittance detector, the method comprising:

S1, acquiring a suction process parameter for representing a suction process of the smoke transmittance detector, an operation state parameter for representing an operation process of the smoke transmittance detector and smoke transmittance in each detection channel of the smoke transmittance detector.

The smoke transmittance detector comprises a trigger cylinder, a lifting cylinder, a pipe clamping cylinder, a needle cylinder motor and a square pipe motor.

The trigger cylinder is mainly used for controlling the starting of the detection process in the smoke transmittance detector. When a detection instruction is received, the cylinder is triggered to act, other related components are pushed to enter a working state, for example, a suction device is started to start to suck a smoke sample. Its normal operation is a precondition for the start of the detection process.

The lifting cylinder is used for adjusting the height position of the light quantity sensor inside the detector. When detecting different types of flue gas or adapting to different detection environments, the height of the detection component needs to be adjusted to ensure the accuracy of detection. For example, for fumes of different concentrations or compositions, adjusting the detection component height may optimize the detection effect. The pipe clamping cylinder is used for clamping or loosening the connecting pipeline and guaranteeing the tightness of the gas path. In the sucking and detecting process, the pipe clamping cylinder ensures that the gas path cannot leak, and the influence of the mixing of external air on the detecting result is prevented. The gas path leakage can cause inaccuracy of the detected smoke concentration, thereby affecting the reliability of the detection result. Syringe motor-typically used to drive a syringe-type suction device-controls the speed and amount of suction. Parameters such as rotation speed and steering of the needle cylinder motor can directly influence the suction capacity and the suction efficiency, so that the detection result is influenced. For example, unstable motor rotation speed may cause fluctuation in the amount of suction, causing abnormal changes in the detected transmittance. And the square tube motor drives a piston in a square tube for containing smoke to be detected in the light transmittance detector to move so as to adjust the space size in the square tube or control the air flow state (including the flow direction) in the square tube. Specifically, as shown in FIG. 1, the pumping process parameters include pumping capacity, diffusion time, pumping interval, diffusion times, pumping compensation coefficient, pumping port count, trigger count, channel transmittance F/M/B, and channel temperature.

The suction capacity refers to the amount of smoke drawn by the smoke transmittance detector in one suction process. In the working principle of the smoke transmittance detector, the suction capacity is determined by the structure and control parameters of the suction device. The suction capacity is controlled, for example, by adjusting the power of the suction pump or controlling the opening of a valve. It is one of the key factors affecting the accuracy of the detection results, as different suction capacities can lead to differences in the detected smoke concentration. If the suction capacity is unstable, deviation of the detected transmittance may occur, thereby affecting the judgment of the smoke concentration. Improper suction capacity control is one of the common causes of detection errors and is therefore an input parameter that must be present and of great importance.

The diffusion time refers to the time required by the extracted flue gas to be uniformly diffused in the flue gas transmittance detector. This is closely related to the structural design inside the smoke transmittance detector and the physical properties of the smoke. Reasonable diffusion time can ensure that the flue gas is fully and uniformly mixed, so that the detected transmittance can accurately reflect the actual concentration of the flue gas. If the diffusion time is too short, the smoke is insufficiently mixed, and the detection result may deviate, and if the diffusion time is too long, the detection efficiency may be affected. In the prior art, accurate control and monitoring of diffusion time is an important step in improving detection accuracy, so it is an important input parameter, but in some cases, the importance of the method can be appropriately reduced if uniform mixing of the flue gas can be ensured by other means compared with the pumping capacity.

Suction interval: the time interval between two suction operations. The proper suction interval can ensure the stable operation of the detector, and the influence on the detection result caused by too frequent suction or too long suction interval is avoided. For example, too frequent pumping may cause unstable pressure inside the device, affecting the pumping effect, and too long pumping intervals may not reflect the change in flue gas concentration in time. The arrangement of the suction interval needs to comprehensively consider the working efficiency of the equipment and the change condition of the smoke concentration. It is an important factor affecting the running stability of the device and the accuracy of the detection result, and belongs to an important input parameter, but the importance of the device can be reduced when the detection environment is relatively stable.

In the flue gas transmittance detector, a plurality of detection channels at different positions are usually arranged around a square tube, and the channel transmittance F/M/B respectively represents flue gas transmittance values measured by the different channels. These values directly reflect the degree of light transmission of the smoke at different locations or under different conditions. By comparing the transparencies of different channels, whether the distribution of the smoke inside the detector is uniform or not and whether local abnormal conditions exist or not can be judged. For example, if the transmittance of a channel differs significantly from that of other channels, this may mean that the detection area corresponding to the channel has problems, such as sensor failure or local blockage. They are key parameters that directly reflect the state of the flue gas, are critical to fault diagnosis, and are necessary and important input parameters.

And the suction compensation coefficient is used for correcting the suction process due to the fact that various interference factors exist in the actual detection process, such as equipment aging, ambient temperature and pressure change and the like, so that the suction effect is different from an ideal state. The method is a parameter obtained by calculation through an algorithm according to the characteristics and actual running conditions of equipment. For example, by combining historical operation data of the equipment and environmental parameters monitored in real time, a specific formula is utilized to calculate a suction compensation coefficient so as to adjust the suction amount, improve the suction accuracy and further improve the reliability of a detection result. In the prior art, reasonable calculation and application of the suction compensation coefficient can effectively improve detection accuracy, and the suction compensation coefficient is an important input parameter, and if the parameter is missing, a large error can exist in a detection result.

In particular, the operating state parameters include a flue gas temperature.

S2, inputting the parameters of the suction process and the transmittance of the smoke in each detection channel into a pre-trained transmittance prediction model in a time sequence data mode, and obtaining the transmittance of the future smoke.

In the embodiment, the transmissivity prediction model adopts MV-LSTM (Multi-View LSTM), and the method combines the advantages of time sequence analysis and deep learning, can effectively process and analyze a large amount of data generated in the running process of the detector, and realizes real-time monitoring of equipment state and early warning of faults.

The collected data first needs to be preprocessed before the model is built. This includes steps of data cleansing, normalization, etc. to ensure quality and consistency of the data. Because the data generated by the smoke transmittance detector has time series characteristics, the data needs to be constructed into a time series data form so as to adapt to the input requirement of a neural network model. The time series data is processed by using a long and short time memory network (LSTM) and an attention mechanism, and key features are extracted. LSTM is a special cyclic neural network, which can effectively solve the problems of gradient disappearance and gradient explosion of the traditional RNN when processing long sequence data. The attention mechanism can make the model pay more attention to important features when processing data, so that the performance of the model is improved. And building a MV-LSTM neural network model. MV-LSTM can process multivariate time series data, and is suitable for multi-parameter monitoring scenes.

The model is trained using the training set data while checking the model performance on the validation set, selecting the best model. In the training process, the model is optimally performed on the verification set by adjusting the model parameters and an optimization algorithm, so that the generalization capability of the model is ensured.

And (3) carrying out real-time monitoring on the transmittance detection value by using the trained model, and immediately giving out early warning once abnormality is found. The model predicts whether the transmittance detection value is within a normal range by analyzing the real-time data, and if the transmittance detection value exceeds a preset threshold value, triggers an early warning mechanism and performs fault diagnosis (i.e. the following S3).

As shown in fig. 3, the input layer of the MV-LSTM network receives the preprocessed and converted multivariate data (i.e., the above-mentioned pumping process parameters) in the form of a time series, and the input layer is followed by a hidden layer, where the hidden layer includes a plurality of stacked LSTM units and a time attention mechanism, and the LSTM units have memory cells and a gating mechanism, so that long-term dependency in long-series data can be effectively processed, and the problems of gradient disappearance and gradient explosion that easily occur when the conventional recurrent neural network processes long-series data are overcome. And the LSTM units are progressive layer by layer, and deep feature extraction is continuously carried out on the data. And finally, an output layer which outputs a prediction result of whether the transmittance detection value is abnormal or not based on the characteristics extracted by the hidden layer.

The MV-LSTM attention mechanism employs a weighting-based calculation approach. When processing time series data, the model calculates the association degree of the time series data and a leachable weight vector according to the data of each time step, so as to obtain the importance score of the data of each time step. The data for all time steps are then weighted according to these scores so that the model can be more focused on those time instants that are critical to the anomaly early warning when processing the data. For example, when the running state of the equipment is abnormal, the data at the moment is given a higher weight, so that the importance of the data to early warning judgment is highlighted.

And the output of MV-LSTM is used for indicating whether the transmittance detection value is abnormal. If the output result is abnormal, the current detected transmittance is beyond the normal range, and potential faults or abnormal operation conditions of the equipment can exist, so that the equipment needs to be timely concerned and processed. The result informs the running state of the user equipment in an intuitive way, and provides timely early warning information for the user so as to take corresponding measures.

Specifically, the transmittance of the MV-LSTM output includes the time domain transmittance at the next time of each detection channel and the mouth number domain transmittance of the next pumping mouth number.

The invention uses the number of times of extraction of the smoke transmittance detector as the number of openings

S3, if the future smoke transmittance (namely, the time domain transmittance and/or the mouth number domain transmittance) output by the transmittance prediction model is abnormal, the smoke transmittance in each detection channel of the operation state parameters is input into a pre-trained fault diagnosis model, and the fault state type of the smoke transmittance detector is obtained.

The threshold value is determined by determining a normal threshold range of light transmittance including an upper threshold value and a lower threshold value based on a large amount of history data at the time of normal operation in advance in combination with factors such as device characteristics and operation environment.

In the process of monitoring the transmittance in real time, the current channel transmittance data (such as channel transmittance F/M/B) are continuously obtained. When the transmittance value of a certain channel exceeds the normal threshold range (or within a certain time window) continuously for a plurality of times, such as continuously lower than the lower threshold or higher than the upper threshold, the abnormal transmittance is determined.

And once the light transmittance is judged to be abnormal, immediately triggering an early warning mechanism, outputting an abnormal early warning result and prompting that equipment operation may have problems. Meanwhile, the reliability of early warning can be further confirmed by combining other parameters (such as whether the suction capacity is changed abnormally synchronously or not).

According to the proposal of the research and development personnel of the smoke transmittance detector, vulnerable parts with direct influence on the detection result are selected as target parts of fault diagnosis, namely a detection square tube, a clamping opening and a thermocouple, as shown in figure 2.

The common neural network models CNN-LSTM and Bi-LSTM, transformer for processing time series data are selected to be used for fault diagnosis tasks of the base rod forming experimental device, and the model with highest diagnosis accuracy is selected to be the final fault diagnosis model of the base rod forming experimental device.

The time step of the preprocessed cigarette segmentation data is 270s, the experimental sample distribution is shown in table 1, and each model is trained under the data background. The cross entropy function is used as a loss function of each model, the training set is utilized to train the model, the model is checked on the verification set, and the model with the best performance on the verification set is saved.

Table 1 experimental sample distribution

After model training is completed, the trained model is tested by using the test set data, and the model diagnosis performance is evaluated. And selecting Accumey (Accuracy), recall (Accuracy), precision (Accuracy) and F1-Measure values (the harmonic average of the Accuracy and the Recall) as performance evaluation indexes of the fault diagnosis model.

The above models were compared, and the results are shown in Table 2, and the transducer model was selected as the final failure diagnosis model, since the transducer effect was found to be the best.

TABLE 2 Performance evaluation results of test set of fault diagnosis models

The fault diagnosis model of the present embodiment adopts a transducer network model. The method can effectively process and analyze a large amount of data generated in the running process of the detector, and early warning of faults is realized. The transducer model test set diagnostic confusion matrix is shown in figure 4.

In the embodiment, the UI interface is built to display the output results of the transmittance prediction model and the fault diagnosis model and the fault tree, and the association among the input layer, the middle layer and the output results is displayed. Meanwhile, the output signal of the diagnosis model is related to a twin model of the smoke transmittance detector shown in fig. 4, so that the visual effect is enhanced, and the fault diagnosis information can be understood visually.

The embodiment also provides corresponding fault processing suggestions and maintenance measures according to the fault diagnosis result. The model not only can diagnose faults, but also can provide possible solutions according to historical data and expert knowledge base. This helps the user to more intuitively understand the fault diagnosis result and maintenance advice. According to feedback in practical application, the model structure and parameters are continuously optimized, and the accuracy and efficiency of fault diagnosis are improved.

According to the embodiment, the transmittance prediction model and the fault diagnosis model are required to be continuously adjusted and optimized according to actual running conditions. And integrating the transmittance prediction model, the fault diagnosis model and the transmittance detector system to realize automatic abnormality early warning and fault diagnosis functions. The system integration is a key step of applying the model to actual equipment, and needs to consider factors such as hardware compatibility, software interfaces and the like.

Fault diagnosis system implementation of smoke transmittance detector:

A fault diagnosis system of a smoke transmittance detector comprising a processor for executing a computer program to implement the steps of the fault diagnosis method of a smoke transmittance detector as described above. The specific method for diagnosing the failure of the smoke transmittance detector is described in detail in the above embodiments of the method for diagnosing the failure of the smoke transmittance detector, and will not be repeated here.

In particular, the processor may be a CPU, but may also be other general purpose processor, digital signal processor (DIGITAL SIGNAL Processer, DSP), application Specific Integrated Circuit (ASIC), field programmable gate array (Field Programmable GATE ARRAY, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like. A general purpose processor may be a microprocessor or any conventional processor or the like. The processor may also be a processor supporting an advanced reduced instruction set machine (ADVANCED RISC MACHINES, ARM) architecture.

Flue gas transmittance detector embodiment:

A smoke transmittance detector comprising a fault diagnosis unit comprising a processor for executing a computer program to implement the steps of the fault diagnosis method of the smoke transmittance detector as described above. The specific method for diagnosing the failure of the smoke transmittance detector is described in detail in the above embodiments of the method for diagnosing the failure of the smoke transmittance detector, and will not be repeated here.

In particular, the processor may be a CPU, but may also be other general purpose processor, digital signal processor (DIGITAL SIGNAL Processer, DSP), application Specific Integrated Circuit (ASIC), field programmable gate array (Field Programmable GATE ARRAY, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like. A general purpose processor may be a microprocessor or any conventional processor or the like. The processor may also be a processor supporting an advanced reduced instruction set machine (ADVANCED RISC MACHINES, ARM) architecture.

The invention has the following characteristics:

the invention develops an abnormality early warning and equipment fault diagnosis model of a smoke transmittance detector. The model realizes real-time monitoring of the running state of the smoke transmittance detector and early warning of faults through advanced data processing and machine learning technologies. The method can process and analyze a large amount of data generated in the running process of the detector, extract key characteristics, and accurately diagnose possible faults of equipment by utilizing the constructed neural network model, so that the efficiency and the accuracy of fault diagnosis are greatly improved. The invention also comprises a user-friendly visual interface which can display the output result of the model and the fault tree, so that the user can intuitively understand the result of fault diagnosis. The model also has self-optimizing capability, and can be continuously adjusted and improved according to feedback in practical application so as to improve the accuracy and efficiency of diagnosis. Through the system integration with the transmittance detector, the model realizes automatic abnormality early warning and fault diagnosis functions, provides an efficient and reliable technical support for environmental monitoring, remarkably reduces maintenance cost and enhances the reliability of equipment.

The present invention applies multivariate data to LSTM networks and designs a mechanism of attention to time series. The traditional LSTM network is large in multiprocessing univariate time series data, and the model creatively introduces multivariate data, so that the model can comprehensively consider the influence of a plurality of operation parameters on transmittance, and more comprehensively capture the equipment operation state information. Meanwhile, attention mechanisms focused on time dimension can dynamically adjust attention degrees of different time step data. The accuracy of abnormality early warning is remarkably improved. Through comprehensive analysis of the multivariate data, the model can timely find potential abnormal conditions in the running process of the equipment, and the problem of missing report caused by the fact that single parameter changes are not timely perceived is avoided. The attention mechanism further enhances the sensitivity of the model to abnormal data at key time points, so that early warning is more timely and accurate.

It should be noted that the foregoing description is only a preferred embodiment of the present invention, and the present invention is not limited to the preferred embodiment, but may be modified without inventive effort or equivalent substitution of some of the technical features thereof by those skilled in the art, even though the present invention has been described in detail with reference to the foregoing embodiment. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (9)

1. The fault diagnosis method of the smoke transmittance detector is characterized by comprising the following steps of:

S1, acquiring a suction process parameter for representing a suction process of a smoke transmittance detector, an operation state parameter for representing an operation process of the smoke transmittance detector and smoke transmittance in each detection channel of the smoke transmittance detector;

S2, inputting the parameters of the suction process and the transmittance of the smoke in each detection channel into a pre-trained transmittance prediction model in a time sequence data form to obtain the transmittance of the future smoke;

S3, if the transmittance of the future smoke output by the transmittance prediction model is abnormal, the operation state parameters and the transmittance of the smoke in each detection channel are input into a pre-trained fault diagnosis model, and the fault state type of the smoke transmittance detector is obtained.

2. The method of claim 1, wherein the pumping process parameters include pumping capacity and pumping compensation coefficients for modifying the pumping process.

3. The method of claim 1, wherein the operating state parameter comprises a temperature of the flue gas.

4. The method according to claim 2, wherein the pumping process parameters further include one or a combination of two or three or more of a diffusion time, a pumping interval, a diffusion multiple, a triggering interval, a triggering time, a preheating time for preheating each detection channel, and a channel temperature.

5. The fault diagnosis method of the smoke transmittance detector according to claim 1, wherein the transmittance prediction model comprises a transmittance feature extraction module and a time attention module, wherein the transmittance feature extraction module is used for extracting the pumping process parameters and time sequence features in the smoke transmittance in each detection channel through LSTM, and the time attention module is used for identifying and weighting key time nodes in the time sequence features output by the transmittance feature extraction module so as to obtain the transmittance of the smoke to be detected according to the weighted features.

6. The method of claim 1, wherein the fault diagnosis model comprises a transducer network model.

7. The method according to claim 1, wherein the fault status type includes a normal status, a detected square tube leak, a clamping port status abnormality, and a thermocouple abnormality.

8. A fault diagnosis system for a smoke transmittance detector, comprising a processor, wherein the processor is configured to execute a computer program to implement the steps of the fault diagnosis method for a smoke transmittance detector according to any one of claims 1 to 7.

9. A smoke transmittance detector comprising a fault diagnosis unit comprising a processor, characterized in that the processor is configured to execute a computer program to implement the steps of the fault diagnosis method of the smoke transmittance detector according to any one of claims 1-7.