Deep learning driven adaptive fault injection method and system

Abstract

The invention discloses a deep learning driven adaptive fault injection method and system, and relates to the technical field of fault injection, and the system comprises a real data collection module, a fault classification module, a fault injection module, a deep learning module, a system evaluation module and a model optimization module. The real data acquisition module collects fault data of equipment through a sensor, a log system and the like; through the training and optimization of a deep learning model, the system can extract key features from a large amount of complex sensor data in a test result, reduces common false alarm and missing alarm phenomena in a traditional method, provides more accurate fault prediction and root cause analysis, improves the monitoring accuracy, facilitates the discovery of potential problems in advance, and improves the reliability of the system. In addition, reasonable allocation of resources is improved, unnecessary maintenance operation and resource waste are avoided, a fault report can be generated and sent to an engineer according to the occurrence probability and influence of different fault modes, and the engineer is helped to improve work.

Description

Technical Field

[0001] The present invention relates to the technical field of fault injection, and specifically to an adaptive fault injection method and system driven by deep learning. Background Art

[0002] In a controlled environment, preset faults or abnormal conditions are artificially introduced to observe and analyze the response mechanism, error handling process, and recovery ability of the system. The theoretical basis of fault injection technology covers multiple disciplinary fields such as software engineering, system reliability analysis, and fault-tolerant computing. It requires testers to have professional technical knowledge and also use tools such as statistics and probability theory to accurately design fault injection schemes to ensure the effectiveness and accuracy of testing. Fault injection systems are also starting to develop towards automation and intelligence to simulate more realistic fault scenarios; Existing fault injection systems rely on predefined rules or simple simulation methods, cannot be dynamically optimized according to actual data, have a low accuracy in predicting faults, will miss potential fault situations and cannot handle complex fault situations; In view of the above technical defects, a solution is proposed. Summary of the Invention

[0003] In view of the deficiencies of the prior art, the present invention provides an adaptive fault injection method and system driven by deep learning.

[0004] To achieve the above objectives, the present invention is realized through the following technical solutions: An adaptive fault injection method and system driven by deep learning, including: a real data acquisition module, a fault classification module, a fault injection module, a deep learning module, a system evaluation module, and a model optimization module; The real data acquisition module collects the fault data of the device through sensors, log systems, etc., including information such as communication faults, mechanical faults, and software faults, and transmits the data to the fault classification module; The fault classification module uses generative adversarial networks (GANs) to classify the fault data, generates communication fault, software fault, and hardware fault data, and transmits it to the fault injection module; The fault injection module, according to different fault types, uses tools to simulate network faults, controls sensors to simulate software faults, and simulates hardware faults, and transmits the test results to the deep learning module; After preprocessing the fault data, the deep learning module is trained through a convolutional neural network (CNN) and a recurrent neural network (RNN). The trained model identifies faults and generates a fault report and sends it to the engineer; The system evaluation module uses the softmax function to calculate the fault occurrence probability and other parameters, generates an evaluation result and transmits it to the model optimization module; The model optimization module receives the evaluation results and adjusts the deep learning model parameters according to the evaluation results.

[0005] The real data acquisition module collects real equipment failure data, including communication failures, mechanical failures, and software failures. Specifically, sensors, log recording systems, and fault diagnosis tools of the equipment collect failure data that occurs during the actual operation of different equipment. The failure data includes failure types, failure frequencies, impact ranges, diagnostic data of the equipment, sensor data in the production environment, and external environment monitoring data. The collected failure data is transmitted to the fault classification module.

[0006] The fault classification module classifies the collected failure data into different types of fault modes. These fault types include communication failures, software failures, and hardware failures. The generative adversarial network (GANs) is used to extract important parameters from the classified failure data to generate communication failure data, software failure data, and hardware failure data, and then transmit them to the fault injection module.

[0007] The fault injection module includes a communication failure processing unit, a software failure processing unit, and a hardware failure processing unit. The fault injection module receives the communication failure data, software failure data, and hardware failure data and distributes them to the communication failure processing unit, software failure processing unit, and hardware failure processing unit. The communication failure processing unit uses the software tool Pumba to insert artificial delays in the network according to the parameters in the communication failure data, evaluates the data processing ability of the system under high-delay conditions, verifies the delay tolerance, and uses the network simulation tool Mininet to set the packet loss rate on the network device to test the retransmission mechanism of the application layer of the system and the recovery ability of the network protocol. The software failure processing unit controls the data flow of the sensor through software, deliberately injects incorrect data or simulates the disconnection state of the sensor, and evaluates whether the system can automatically switch to the backup sensor or repair the sensor data through an algorithm. The hardware failure processing unit uses the fault injection device to simulate the motor load change and adjust the friction of the transmission component according to the parameters in the hardware failure data to verify the fault recovery ability of the system, and uses the power management device to simulate the instantaneous increase or decrease of the voltage to test the UPS system of the system terminal and the standby battery switching mechanism. The fault injection module transmits the test result data after testing by the communication failure processing unit, software failure processing unit, and hardware failure processing unit to the deep learning module.

[0008] The deep learning module receives the test result data and preprocesses the fault data. The preprocessing includes data cleaning, annotation, and normalization. The model training module injects the preprocessed data into the convolutional neural network CNN. The convolutional neural network CNN extracts local features through the convolutional layer and reduces the data dimension through the pooling layer. In the convolutional layer, the convolutional kernel size is set to (3, 3), the stride is set to (1, 1), and the padding method is padding = same, so that the size of the output is the same as that of the input. The activation function ReLU is selected. The pooling layer uses MaxPooling, and the pooling size is set to (2, 2). For time series data, the recurrent neural network RNN and the long short-term memory network LSTM are used. The number of units in the LSTM layer is set to 128, and dropout = 0.2 is applied. The network structure adjusts the structures of the input layer, hidden layer, and output layer according to the characteristics of the data input. The shape of the input layer is (batch_size, time_steps, features). The hidden layer includes multiple convolutional layers or LSTM layers. The output layer uses softmax for classification tasks according to the task classification and uses a linear activation function for regression tasks. The loss function is selected according to the task type. Categorical_crossentropy is used for classification tasks, and mean_squared_error is used for regression tasks. The Adam optimizer is selected as the optimization algorithm. The learning rate is set to 0.001, beta_1 is set to 0.9, and beta_2 is set to 0.999. During the hyperparameter tuning process, the learning rate is set to 0.001, the batch size is set to 32, and the number of training epochs is set to 50; The trained model is used to re-analyze the fault data, identify different types of faults, and calculate the occurrence probabilities of various faults using the softmax activation function. The features are extracted from the middle layer of the model and combined with the data analysis method LIME to analyze the causes of the faults. Further, according to the occurrence probabilities, types of the faults, and the importance of the equipment, production impact, and safety risks, the urgency of the faults is evaluated. Optimization suggestions are generated using data mining and rule engines, and a fault report is generated through LaTeX and sent to the engineers. The report includes the fault type, occurrence probability, impact assessment, cause analysis, and optimization suggestions.

[0009] The system evaluation module: The model evaluation module is used to evaluate the performance of the model. The effects of the model in tasks such as fault prediction and classification are measured by monitoring various indicators of the system, including CPU occupancy, memory usage, and response time. The precision, recall, and F1 values are calculated using the precision_score function of scikit-learn, and cross-validation is performed using cross_val_score and cross_validate. The evaluation indicators are response time, recovery time, and system stability. The evaluation results are generated and transmitted to the model optimization module.

[0010] The model optimization module continuously optimizes the deep learning model according to the evaluation results, uses the feedback of the model training results to adjust the system parameters, optimizes the model by adjusting the learning rate, batch size, regularization coefficient, and optimization algorithm, and the system dynamically adjusts the network structure according to the feedback, modifies the number of nodes in the hidden layer, increases or decreases the number of layers, and further introduces BatchNormalization to accelerate training and improve stability.

[0011] The specific method of the deep learning-driven adaptive fault injection method is as follows: S1. The data acquisition module collects device fault data, including communication, mechanical, and software faults, and transmits the fault data to the fault classification module; S2. The fault classification module classifies the fault data and transmits the classified fault data to the fault injection module; S3. The fault injection module simulates different types of faults and transmits the test results to the deep learning module; S4. The deep learning module uses the test results to train the deep learning model and analyzes the faults with the trained model to generate a fault report; S5. The evaluation module monitors the system performance, evaluates the model performance, and transmits the evaluation results to the model optimization module; S6. The model optimization module adjusts the model parameters according to the evaluation results.

[0012] The present invention provides a deep learning-driven adaptive fault injection method and system. Compared with the prior art, it has the following beneficial effects: Through the training and optimization of the deep learning model, the system of the present invention can extract key features from a large amount of complex sensor data in the test results, reduce false alarms and missed alarms common in traditional methods, provide more accurate fault prediction and root cause analysis, improve the monitoring accuracy, which helps to detect potential problems in advance. In addition, it improves the reasonable allocation of resources, avoids unnecessary maintenance operations and resource waste, and can generate a fault report according to the occurrence probability and impact of different fault modes and send it to the engineer to help the engineer improve the work. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with 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, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] Please refer to Figure 1 , this application provides an adaptive fault injection method and system driven by deep learning, including a real data acquisition module, a fault classification module, a fault injection module, a deep learning module, a system evaluation module, and a model optimization module; The real data acquisition module collects the fault data of the device through sensors, log systems, etc., including communication faults, mechanical faults, software faults, etc., and transmits the data to the fault classification module; The fault classification module uses Generative Adversarial Networks (GANs) to classify the fault data, generates communication fault data, software fault data, and hardware fault data, and transmits them to the fault injection module; The fault injection module, according to different fault types, uses tools to simulate network faults, controls sensors to simulate software faults, and simulates hardware faults, and transmits the test results to the deep learning module; After preprocessing the fault data, the deep learning module is trained through a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN). The trained model identifies faults and generates fault reports to be sent to engineers; The system evaluation module uses the softmax function to calculate the probability of fault occurrence and other parameters, and generates evaluation results to be transmitted to the model optimization module; The model optimization module receives the evaluation results and adjusts the parameters of the deep learning model according to the evaluation results.

[0016] The real data acquisition module collects real device fault data, including communication faults, mechanical faults, and software faults. Specifically, sensors, log recording systems, fault diagnosis tools, etc. of the device collect fault data that occurs during the actual operation of different devices. The fault data includes fault types, fault occurrence frequencies, impact ranges, diagnostic data of the device, sensor data in the production environment, and external environment monitoring data, and transmits the collected fault data to the fault classification module.

[0017] The fault classification module classifies the collected fault data into different types of fault patterns. These fault types include communication faults, software faults, and hardware faults. Generative Adversarial Networks (GANs) are used to extract important parameters from the classified fault data to generate communication fault data, software fault data, and hardware fault data, and transmit them to the fault injection module.

[0018] The generative adversarial network is used in specific classification, enabling the fault classification module to have stronger adaptability. By continuously comparing and optimizing the generated data with the real data, the model can better fit the actual fault patterns, thereby improving the accuracy and reliability of fault classification.

[0019] The fault injection module includes a communication fault handling unit, a software fault handling unit, and a hardware fault handling unit. The fault injection module receives communication fault data, software fault data, and hardware fault data and distributes them to the communication fault handling unit, the software fault handling unit, and the hardware fault handling unit. The communication fault handling unit uses the software tool Pumba to insert artificial delays in the network according to the parameters in the communication fault data, evaluates the data processing ability of the system under high-delay conditions, verifies the delay tolerance, and uses the network simulation tool Mininet to set the packet loss rate on network devices to test the retransmission mechanism of the system application layer and the recovery ability of network protocols. The software fault handling unit controls the data flow of the sensor through software, deliberately injects incorrect data or simulates the disconnection state of the sensor, and evaluates whether the system can automatically switch to a backup sensor or repair the sensor data through an algorithm. The hardware fault handling unit uses a fault injection device to simulate motor load changes and adjust the friction of transmission components according to the parameters in the hardware fault data to verify the fault recovery ability of the system, and uses a power management device to simulate an instantaneous increase or decrease in voltage to test the terminal UPS system and the standby battery switching mechanism of the system. The fault injection module transmits the test result data after testing by the communication fault handling unit, the software fault handling unit, and the hardware fault handling unit to the deep learning module.

[0020] The deep learning module receives the test result data and preprocesses the fault data. The preprocessing includes data cleaning, annotation, and normalization. The model training module injects the preprocessed data into the convolutional neural network CNN. The convolutional neural network CNN extracts local features through the convolutional layer and reduces the data dimension through the pooling layer. In the convolutional layer, the convolutional kernel size is set to (3, 3), the stride is set to (1, 1), the padding method is padding = same, and the output has the same size as the input. The activation function selects ReLU. The pooling layer uses MaxPooling, and the pooling size is set to (2, 2). For time series data, the recurrent neural network RNN and the long short-term memory network LSTM are used. The number of units in the LSTM layer is set to 128, and dropout = 0.2 is applied. The network structure adjusts the structures of the input layer, hidden layer, and output layer according to the characteristics of the data input. The shape of the input layer is (batch_size, time_steps, features). The hidden layer includes multiple convolutional layers or LSTM layers. The output layer uses softmax for classification tasks according to task classification and uses a linear activation function for regression tasks. The loss function is selected according to the task type. Categorical_crossentropy is used for classification tasks, and mean_squared_error is used for regression tasks. The optimization algorithm selects the Adam optimizer. The learning rate is set to 0.001, beta_1 is set to 0.9, and beta_2 is set to 0.999. During the hyperparameter tuning process, the learning rate is set to 0.001, the batch size is set to 32, and the number of training epochs is set to 50; The trained model re-analyzes the fault data, identifies different types of faults, and calculates the occurrence probabilities of various faults using the softmax activation function. It extracts features from the middle layer of the model and combines with the data analysis method LIME to analyze the causes of faults. Further, according to the occurrence probabilities, types of faults, importance of the equipment, production impact, and safety risks, it evaluates the urgency of the faults, generates optimization suggestions using data mining and rule engines, and generates a fault report in LaTeX and sends it to the engineer. The report includes fault types, occurrence probabilities, impact assessments, cause analyses, and optimization suggestions.

[0021] Combined with the importance of the equipment, production impact, and safety risks, the model will evaluate the urgency of the faults and determine which faults need to be processed first. For faults with a higher urgency, the model generates optimization suggestions through data mining and rule engines.

[0022] System Evaluation Module: The model evaluation module is used to evaluate the performance of the model. It measures the effectiveness of the model in tasks such as fault prediction and classification by monitoring various system metrics including CPU occupancy, memory usage, and response time. It calculates precision, recall, and F1 values through the precision_score function of scikit - learn, and uses cross_val_score and cross_validate for cross - validation. The evaluation metrics are response time, recovery time, and system stability, and the generated evaluation results are transmitted to the model optimization module.

[0023] The model optimization module continuously optimizes the deep - learning model according to the evaluation results. It uses the feedback of the model training results to adjust system parameters, and optimizes the model by adjusting the learning rate, batch size, regularization coefficient, and optimization algorithm. The system dynamically adjusts the network structure according to the feedback, modifies the number of nodes in the hidden layer, increases or decreases the number of layers, and further introduces BatchNormalization to accelerate training and improve stability.

[0024] The specific method of the deep - learning - driven adaptive fault injection method is as follows: S1. The data acquisition module collects device fault data, including communication, mechanical, and software faults, and transmits the fault data to the fault classification module; S2. The fault classification module classifies the fault data and transmits the classified fault data to the fault injection module; S3. The fault injection module simulates different types of faults and transmits the test results to the deep - learning module; S4. The deep - learning module uses the test results to train the deep - learning model and analyzes the faults with the trained model to generate a fault report; S5. The evaluation module monitors the system performance, evaluates the model performance, and transmits the evaluation results to the model optimization module; S6. The model optimization module adjusts the model parameters according to the evaluation results.

[0025] Specific working process: The real - data acquisition module collects the fault data of the device through sensors, logging systems, etc. The collected fault data includes fault types, occurrence frequencies, and influence scopes, and is transmitted to the fault classification module. The fault classification module classifies the collected fault data, generates different types of fault data, and transmits the classified data to the fault injection module. The fault injection module simulates corresponding faults through different processing units according to the fault types: the communication - fault processing unit simulates network latency and packet loss, the software - fault processing unit simulates sensor - data errors, and the hardware - fault processing unit simulates motor - load changes and transmission - component faults. The test results after injection are transmitted to the deep - learning module. The deep - learning module pre - processes the injected test results and uses a convolutional neural network for training. The trained model is used to identify faults, generate fault reports, and send them to engineers. The evaluation module monitors the performance of the model, uses precision and recall evaluation metrics, generates evaluation results, and transmits them to the model - optimization module. The model - optimization module adjusts the parameters of the deep - learning model according to the evaluation results and introduces BatchNormalization to accelerate training and improve stability.

[0026] Furthermore, through the training and optimization of the deep - learning model in the present invention, the system can extract key features from a large amount of complex sensor data in the test results, reduce false alarms and missed alarms common in traditional methods, provide more accurate fault prediction and root - cause analysis. Improving monitoring accuracy helps to detect potential problems in advance. In addition, it improves the rational allocation of resources, avoids unnecessary maintenance operations and resource waste, and can generate fault reports according to the occurrence probabilities and influences of different fault modes and send them to engineers to help engineers improve their work.

[0027] Some of the data in the above formula are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well - known to those skilled in the art.

[0028] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that modifications or equivalent replacements can be made to the technical method of the present invention without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A deep learning-driven adaptive fault injection system, characterized in that, Including: A real data acquisition module, a fault classification module, a fault injection module, a deep learning module, a system evaluation module, and a model optimization module; The real data acquisition module collects the fault data of the device through sensors, log systems, etc., including communication faults, mechanical faults, software faults, etc., and transmits the data to the fault classification module; The fault classification module uses Generative Adversarial Networks (GANs) to classify the fault data, generates communication fault data, software fault data, and hardware fault data, and transmits them to the fault injection module; The fault injection module, according to different fault types, uses tools to simulate network faults, controls sensors to simulate software faults, and simulates hardware faults, and transmits the test results to the deep learning module; After preprocessing the fault data, the deep learning module is trained through Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). The trained model identifies faults and generates fault reports to be sent to engineers; The system evaluation module uses the softmax function to calculate the probability of fault occurrence and other parameters, and generates evaluation results to be transmitted to the model optimization module; The model optimization module receives the evaluation results and adjusts the parameters of the deep learning model according to the evaluation results.

2. The deep learning-driven adaptive fault injection system according to claim 1, wherein The real data acquisition module collects real device fault data, including communication faults, mechanical faults, and software faults. Specifically, the device's sensors, log recording systems, fault diagnosis tools, etc. collect the fault data that occurs during the actual operation of different devices. The fault data includes fault types, fault occurrence frequencies, influence ranges, device diagnostic data, sensor data in the production environment, and external environment monitoring data, and transmits the collected fault data to the fault classification module.

3. The deep learning-driven adaptive fault injection system according to claim 1, wherein The fault classification module classifies the collected fault data into different types of fault patterns. These fault types include communication faults, software faults, and hardware faults. It uses Generative Adversarial Networks (GANs) to extract important parameters from the classified fault data to generate communication fault data, software fault data, and hardware fault data, and transmits them to the fault injection module.

4. The deep learning-driven adaptive fault injection system according to claim 1, characterized in that The fault injection module includes a communication fault processing unit, a software fault processing unit, and a hardware fault processing unit. The fault injection module receives communication fault data, software fault data, and hardware fault data and distributes them to the communication fault processing unit, the software fault processing unit, and the hardware fault processing unit. The communication fault processing unit uses the software tool Pumba to insert artificial delays in the network according to the parameters in the communication fault data, evaluates the data processing ability of the system under high-delay conditions, verifies the delay tolerance, and uses the network simulation tool Mininet to set the packet loss rate on network devices to test the retransmission mechanism of the system application layer and the recovery ability of network protocols. The software fault processing unit controls the data flow of the sensor through software, deliberately injects incorrect data or simulates the disconnection state of the sensor, and evaluates whether the system can automatically switch to a standby sensor or repair the sensor data through an algorithm. The hardware fault processing unit uses a fault injection device to simulate motor load changes and adjust the friction of transmission components according to the parameters in the hardware fault data to verify the fault recovery ability of the system, and uses a power management device to simulate an instantaneous increase or decrease in voltage to test the terminal UPS system and the standby battery switching mechanism of the system. The fault injection module transmits the test result data after testing by the communication fault processing unit, the software fault processing unit, and the hardware fault processing unit to the deep learning module.

5. The deep learning-driven adaptive fault injection system according to claim 1, wherein The deep learning module accepts the test result data and preprocesses the fault data. The preprocessing includes data cleaning, annotation, and normalization. The model training module injects the preprocessed data into a Convolutional Neural Network (CNN). The CNN extracts local features through convolutional layers and reduces the data dimension through pooling layers. In the convolutional layer, the kernel size is set to (3, 3), the stride is set to (1, 1), and the padding method is padding=same, so that the output has the same size as the input. The activation function ReLU is selected. The pooling layer uses MaxPooling, and the pooling size is set to (2, 2). For time series data, a Recurrent Neural Network (RNN) and a Long Short-Term Memory Network (LSTM) are used. The number of units in the LSTM layer is set to 128, and dropout=0.2 is applied. The network structure adjusts the structure of the input layer, hidden layer, and output layer according to the characteristics of the data input. The shape of the input layer is (batch_size, time_steps, features). The hidden layer includes multiple convolutional layers or LSTM layers. The output layer uses softmax for classification tasks and a linear activation function for regression tasks according to the task classification. The loss function is selected according to the task type, categorical_crossentropy for classification tasks and mean_squared_error for regression tasks. The optimization algorithm selects the Adam optimizer, the learning rate is set to 0.001, beta_1 is set to 0.9, and beta_2 is set to 0.

999. During the hyperparameter tuning process, the learning rate is set to 0.001, the batch size is set to 32, and the number of training epochs is set to 50; The trained model is used to re-analyze the fault data, identify different types of faults, and calculate the occurrence probability of various faults using the softmax activation function. Features are extracted from the middle layer of the model and combined with the data analysis method LIME to analyze the causes of the faults. Further, according to the occurrence probability, type of the faults, and the importance of the equipment, production impact, and safety risks, the urgency of the faults is evaluated. Optimization suggestions are generated using data mining and rule engines, and a fault report is generated using LaTeX and sent to the engineers. The report includes the fault type, occurrence probability, impact assessment, cause analysis, and optimization suggestions.

6. The deep learning-driven adaptive fault injection system according to claim 1, wherein The system evaluation module: The model evaluation module is used to evaluate the performance of the model. The effectiveness of the model in tasks such as fault prediction and classification is measured by monitoring various indicators of the system, including CPU occupancy, memory usage, and response time. The precision, recall, and F1 values are calculated using the precision_score function of scikit-learn, and cross-validation is performed using cross_val_score and cross_validate. The evaluation metrics are response time, recovery time, and system stability. The evaluation results are generated and transmitted to the model optimization module.

7. The deep learning-driven adaptive fault injection system according to claim 1, wherein The model optimization module continuously optimizes the deep learning model according to the evaluation results, uses the feedback of the model training results to adjust the system parameters, and optimizes the model by adjusting the learning rate, batch size, regularization coefficient, and optimization algorithm. The system dynamically adjusts the network structure according to the feedback, modifies the number of nodes in the hidden layer, increases or decreases the number of layers, and further introduces BatchNormalization to accelerate training and improve stability.

8. Deep learning-driven adaptive fault injection method and system, characterized in that The specific method of the adaptive fault injection method driven by deep learning is as follows: S1. The data acquisition module collects device fault data, including communication, mechanical, and software faults, and transmits the fault data to the fault classification module; S2. The fault classification module classifies the fault data and transmits the classified fault data to the fault injection module; S3. The fault injection module simulates different types of faults and transmits the test results to the deep learning module; S4. The deep learning module uses the test results to train the deep learning model and analyzes the faults with the trained model to generate a fault report; S5. The evaluation module monitors the system performance, evaluates the model performance, and transmits the evaluation results to the model optimization module; S6. The model optimization module adjusts the model parameters according to the evaluation results.

Images