Multi-model microcontroller signal processing device, method and related products

Abstract

The present application provides a multi-model microcontroller signal processing device, method and related products, which can be applied to fault diagnosis of 32-bit microcontrollers. The device includes: a first pin for inputting a first fault signal; a second pin for inputting a second fault signal; a processor including a first signal processing module and a second signal processing module, the first pin being directly connected to the first signal processing module, the second pin being directly connected to the second signal processing module, the first signal processing module being used to receive the first fault signal input by the first pin, and inputting the first fault signal into the first signal processing model for signal processing, the second signal processing module being used to receive the second fault signal input by the second pin, and inputting the intermediate data and the first classification data in the first signal processing model, as well as the second fault signal, into the second signal processing model for signal processing.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a multi-model microcontroller signal processing device, method, and related products. Background Art

[0002] Currently, 32-bit microcontrollers are widely used in heating, ventilation, and air conditioning (HVAC), automotive exterior and interior lighting, and door, window, wiper, and seat control. However, 32-bit microcontrollers often have a wide variety of internal and external faults, leading to an urgent need to improve the diagnosis of these faults. Summary of the Invention

[0003] To solve the above technical problems, the present application provides a multi-model microcontroller signal processing device, method and related products, which can effectively improve the diagnosis effect of internal faults and external faults of 32-bit microcontrollers.

[0004] In a first aspect, a multi-model microcontroller signal processing device is provided, comprising:

[0005] A first pin, used for inputting a first fault signal;

[0006] The second pin is used to input a second fault signal;

[0007] The processor includes a first signal processing module and a second signal processing module. The above-mentioned first pin is directly connected to the first signal processing module, and the above-mentioned second pin is directly connected to the second signal processing module. The first signal processing module is used to receive the first fault signal input by the above-mentioned first pin, and input the first fault signal into the first signal processing model for signal processing. The second signal processing module is used to receive the second fault signal input by the above-mentioned second pin, and input the intermediate data and at least one of the first classified data in the first signal processing model, as well as the second fault signal into the second signal processing model for signal processing.

[0008] Among them, the intermediate data is the data generated during the signal processing process, the first classified data is the result of the first signal processing model performing signal processing based on the first fault signal, and the dependency relationship between at least one of the intermediate data and the first classified data and the second fault signal is obtained by calculating the correlation measure.

[0009] In the above scheme, using at least one of the intermediate data and the first classification data in the first signal processing model as the second signal processing model in the additional information input device enables the second signal processing model to extract more features during the signal processing of the second fault signal, including features contained in the second fault signal itself and features of information related to the second fault signal. Furthermore, the second signal processing model can integrate the second fault signal and the additional information to conduct in-depth mining of the input data and obtain higher-quality features. Therefore, the second signal processing model processes the second fault signal based on a larger number of higher-quality features, and can obtain a more accurate classification result. This improves the accuracy of fault diagnosis and enhances the effectiveness of fault diagnosis when there are many types of faults.

[0010] In some possible implementations, the first signal processing model and the second signal processing model each include a neural network model.

[0011] In the above scheme, the first signal processing model and the second signal processing model adopt a neural network model, which can utilize the feature extraction ability and anti-interference ability of the neural network model, and provide a convenient implementation method for effectively improving the fault diagnosis effect of the multi-model microcontroller signal processing device.

[0012] In some possible implementations, the first signal processing model and the second signal processing model both include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

[0013] In the above solution, the first signal processing model uses a self-attention model to determine the spatial correlation of information extracted from the first fault signal, thereby determining a self-similarity measure for the first fault signal. The second signal processing model uses a self-attention model to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model and the second fault signal, thereby determining a self-similarity measure for the second fault signal.

[0014] In some possible implementations, the initial state of the second signal processing model is implemented based on at least one of the intermediate state and the classification state of the first signal processing model.

[0015] Among them, the above-mentioned intermediate state represents the state of the first signal processing model when the above-mentioned first signal processing model performs signal processing on the above-mentioned first fault signal, and the above-mentioned classification state represents the state of the first signal processing model when the above-mentioned first signal processing model calculates the first classification data corresponding to the above-mentioned first fault signal.

[0016] In the above scheme, the initial state of the second signal processing model is realized according to at least one of the intermediate state and classification state of the first signal processing model, which can connect the operations implemented by the first signal processing module and the second signal processing module, thereby facilitating the extraction of information correlation between the data from the first signal processing model and the second fault signal.

[0017] In some possible implementations, the processor further includes a backup module, which is used to replace the first signal processing module and the second signal processing module when a failure occurs.

[0018] In the above solution, a backup module is provided in the processor to ensure that the multi-model microcontroller signal processing device can continue to operate normally when a disorder occurs.

[0019] In a second aspect, a multi-model microcontroller signal processing method is provided, comprising:

[0020] The multi-model microcontroller signal processing device obtains a first fault signal through a first pin and obtains a second fault signal through a second pin;

[0021] The multi-model microcontroller signal processing device inputs the first fault signal into the first signal processing model in the first signal processing module for signal processing to obtain intermediate data and first classification data;

[0022] The multi-model microcontroller signal processing device inputs the intermediate data, at least one of the first classification data, and the second fault signal into the second signal processing model in the second signal processing module for signal processing.

[0023] Among them, the above-mentioned intermediate data is the data generated in the above-mentioned signal processing process, the above-mentioned first classification data is the result of signal processing performed by the above-mentioned first signal processing model based on the above-mentioned first fault signal, and the dependency relationship between at least one of the above-mentioned intermediate data and the first classification data and the above-mentioned second fault signal is obtained by calculating the correlation measure.

[0024] In the above scheme, using at least one of the intermediate data and the first classification data in the first signal processing model as the second signal processing model in the additional information input device enables the second signal processing model to extract more features during the signal processing of the second fault signal, including features contained in the second fault signal itself and features of information related to the second fault signal. Furthermore, the second signal processing model can integrate the second fault signal and the additional information to conduct in-depth mining of the input data and obtain higher-quality features. Therefore, the second signal processing model processes the second fault signal based on a larger number of higher-quality features, and can obtain a more accurate classification result. This improves the accuracy of fault diagnosis and enhances the effectiveness of fault diagnosis when there are many types of faults.

[0025] In some possible implementations, the first signal processing model and the second signal processing model each include a neural network model.

[0026] In the above scheme, the first signal processing model and the second signal processing model adopt a neural network model, which can utilize the feature extraction ability and anti-interference ability of the neural network model, and provide a convenient implementation method for effectively improving the fault diagnosis effect of the multi-model microcontroller signal processing device.

[0027] In some possible implementations, the first signal processing model and the second signal processing model both include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

[0028] In the above solution, the first signal processing model uses a self-attention model to determine the spatial correlation of information extracted from the first fault signal, thereby determining a self-similarity measure for the first fault signal. The second signal processing model uses a self-attention model to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model and the second fault signal, thereby determining a self-similarity measure for the second fault signal.

[0029] In some possible implementations, the initial state of the second signal processing model is implemented based on at least one of the intermediate state and the classification state of the first signal processing model.

[0030] Among them, the above-mentioned intermediate state represents the state of the first signal processing model when the above-mentioned first signal processing model performs signal processing on the above-mentioned first fault signal, and the above-mentioned classification state represents the state of the first signal processing model when the above-mentioned first signal processing model calculates the first classification data corresponding to the above-mentioned first fault signal.

[0031] In the above scheme, the initial state of the second signal processing model is realized according to at least one of the intermediate state and classification state of the first signal processing model, which can connect the operations implemented by the first signal processing module and the second signal processing module, thereby facilitating the extraction of information correlation between the data from the first signal processing model and the second fault signal.

[0032] In some possible implementations, the above method further includes:

[0033] When the first signal processing module or the second signal processing module fails, the backup module is used to replace the first signal processing module or the second signal processing module.

[0034] In the above solution, a backup module is provided in the processor to ensure that the multi-model microcontroller signal processing device can continue to operate normally when a disorder occurs.

[0035] In a third aspect, a multi-model microcontroller signal processing device is provided, comprising: an acquisition module, a first processing module, and a second processing module.

[0036] The acquisition module is used to acquire the first fault signal and the second fault signal;

[0037] The first processing module is used to input the first fault signal into a first signal processing model for signal processing to obtain intermediate data and first classification data;

[0038] The second processing module is used to input at least one of the intermediate data and the first classification data, and the second fault signal into a second signal processing model for signal processing.

[0039] Among them, the above-mentioned intermediate data is the data generated in the above-mentioned signal processing process, the above-mentioned first classification data is the result of signal processing performed by the above-mentioned first signal processing model based on the above-mentioned first fault signal, and the dependency relationship between at least one of the above-mentioned intermediate data and the first classification data and the above-mentioned second fault signal is obtained by calculating the correlation measure.

[0040] In some possible implementations, the first signal processing model and the second signal processing model each include a neural network model.

[0041] In some possible implementations, the first signal processing model and the second signal processing model both include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

[0042] In some possible implementations, the initial state of the second signal processing model is implemented based on at least one of the intermediate state and the classification state of the first signal processing model.

[0043] Among them, the above-mentioned intermediate state represents the state of the first signal processing model when the above-mentioned first signal processing model performs signal processing on the above-mentioned first fault signal, and the above-mentioned classification state represents the state of the first signal processing model when the above-mentioned first signal processing model calculates the first classification data corresponding to the above-mentioned first fault signal.

[0044] In some possible implementations, the above device further includes a backup module.

[0045] The backup module is used to replace the first processing module or the second processing module when a failure occurs in the first processing module or the second processing module.

[0046] In a fourth aspect, a chip is provided, comprising a processor and a memory, wherein the processor is configured to execute instructions stored in the memory so that the chip performs any method according to the second aspect.

[0047] In a fifth aspect, a computer-readable storage medium is provided, comprising computer program instructions. When the computer program instructions are executed by a computing device, the computing device executes the method according to any one of the second aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a schematic structural diagram of a multi-model microcontroller signal processing device provided by the present application;

[0049] Figure 2 This is a flow chart of a multi-model microcontroller signal processing method provided by the present application;

[0050] Figure 3 This is a structural diagram of a multi-model microcontroller signal processing device provided by this application. DETAILED DESCRIPTION

[0051] The embodiment of the present invention is described below with reference to the accompanying drawings. For further explanation, the following description is made by taking a 32-bit microcontroller applied to a car as an example.

[0052] See also Figure 1 , Figure 1 This is a schematic diagram of the structure of a multi-model microcontroller signal processing device provided by this application. Figure 1 As shown, the multi-model microcontroller signal processing device 100 includes a first pin 101 , a second pin 102 , and a processor 103 .

[0053] The first pin 101 is used to input a first fault signal 104 to the processor 103. The first fault signal 104 may originate from an internal fault of the 32-bit microcontroller, such as undervoltage, overvoltage, core voltage overload, flash memory undervoltage, temperature anomaly, a single-bit fault in static random access memory (SRAM), a multi-bit fault in static random access memory (SRAM), a single-bit fault in flash memory, a multi-bit fault in flash memory, an error correction memory (ECM) fault, etc. The first fault signal 104 may also originate from an external fault of the 32-bit microcontroller, such as a vehicle interior temperature control fault, a vehicle interior lighting fault, a door fault, a window fault, etc.

[0054] Second pin 102 is used to input a second fault signal 105 to processor 103. Second fault signal 105 can originate from an internal fault of the 32-bit microcontroller, such as undervoltage, overvoltage, core voltage overload, flash memory undervoltage, temperature anomaly, a single-bit static random access memory (SRAM) fault, a multi-bit SRAM fault, a single-bit flash memory fault, a multi-bit flash memory fault, an error correction memory fault, etc. Second fault signal 105 can also originate from an external fault of the 32-bit microcontroller, such as a vehicle interior temperature control fault, a vehicle interior lighting fault, a door fault, a window fault, etc. Furthermore, second fault signal 105 and first fault signal 104 may originate from different faults.

[0055] The processor 103 includes a first signal processing module 106 and a second signal processing module 107 .

[0056] The first signal processing module 106 is connected to the first pin 101 and is used to receive the first fault signal 104 input by the first pin 101 and input the first fault signal 104 into the first signal processing model 108 in the first signal processing module 106 for signal processing to obtain first classification data 109 .

[0057] In this application, most fault data exhibits self-similarity, meaning that a local portion of the fault data is highly similar (or correlated) to another portion of the fault data. For example, if the fault source is a periodic fault, the fault data will exhibit periodic changes. Alternatively, if the fault source evolves from a single-point fault to a coupled fault, the representation and changes of the fault data in the coupled fault will include the representation and changes of the fault data in the single-point fault.

[0058] Therefore, in the first signal processing model 108, the model relies on a self-similarity measure of the first fault signal 104. This self-similarity measure is a measure of the similarity between different data segments in the first fault signal 104. This measure incorporates the interactions between different data segments and is calculated using segments of any length from the first fault signal 104. Therefore, the signal processing implemented by the first signal processing model 108 is based on the spatial correlation of the information extracted from the first fault signal 104. In this way, the first signal processing model 108 is enriched with a representation of similarity information between all possible pairs of segments in the first fault signal 104. These relationships are considered a salient and dense representation of the relevant information contained in the first fault signal 104.

[0059] The categories in the first signal processing model 108 include: low voltage, high voltage, core voltage too high, flash memory voltage too low, temperature abnormality, static random access memory single-bit failure, static random access memory multi-bit failure, flash memory single-bit failure, flash memory multi-bit failure, error correction memory failure, car interior temperature control failure, car interior lighting opening and closing failure, door opening and closing failure, window opening and closing failure, etc.

[0060] The first signal processing module 106 inputs the first fault signal 104 into the first signal processing model 108 to calculate the probability that the first fault signal 104 belongs to each category and selects the category with the highest probability as the first classification data 109. In some possible implementations, the first signal processing model 108 may be a neural network model. Specifically, the neural network model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to receive the first fault signal 104, and cut it into segments of arbitrary length, and input these segments into the hidden layer. Furthermore, the input layer can be used to first normalize the first fault signal 104, and then cut the processed data into segments of arbitrary length, and input these segments into the hidden layer; the hidden layer is used to use the segments output by the input layer as input data, extract the similarity information between all possible segments, and calculate the output data of the layer, which is used to represent the spatial correlation of the information extracted from the first fault signal 104; the output layer is used to use the output data of the hidden layer as input data, and use normalized exponential functions, logistic functions and other functions to calculate, to obtain the probability that the first fault signal 104 belongs to each category, and select the category with the highest probability as the first classification data 109. In particular, when the neural network model has multiple hidden layers, the first hidden layer is used to use the segments output by the input layer as input data, extract the similarity information between all possible segments pairs, and calculate the output data of this layer; the second hidden layer is used to use the output data of the first hidden layer as input data, further extract the similarity information, and obtain the output data of this layer; the third hidden layer is used to use the output data of the second hidden layer as input data, and further extract the similarity information to obtain the output data of this layer; and so on, until the last hidden layer calculates the output data of this layer; and the output layer is used to use the output data of the last hidden layer as input data, and use normalized exponential function, logistic function and other functions to calculate, to obtain the probability that the first fault signal 104 belongs to each category, and select the category with the highest probability as the first classification data 109.

[0061] In some possible implementations, the first signal processing model 108 may include a long short-term memory model, a self-attention model, and a classifier. Specifically, the first signal processing module 106 inputs the first fault signal 104 into the long short-term memory model and calculates a time series result. Subsequently, the first signal processing module 106 inputs the time series result into the self-attention model to obtain high-dimensional feature data, wherein the self-attention model is used to determine the spatial correlation of the information extracted from the first fault signal 104, including a self-attention sub-model and a fully connected layer, and the calculation process is to input the time series result into the self-attention sub-model to obtain multiple attention values, and then input these attention values ​​into the fully connected layer to obtain high-dimensional feature data. Finally, the first signal processing module 106 inputs the high-dimensional feature data into a classifier, such as a Softmax classifier or a logistic classifier, to obtain the probability that the first fault signal 104 belongs to each category, and selects the category with the highest probability as the first classification data 109.

[0062] It should be understood that the aforementioned LSTM model, self-attention model, and classifier can be three different neural network models, or they can be integrated into the same neural network model. When the LSTM model, self-attention model, and classifier are integrated into the same neural network model, the LSTM model, self-attention model, and classifier can be different layers in the neural network model.

[0063] The second signal processing module 107 is connected to the second pin 102, and is used to receive the second fault signal 105 input from the second pin 102, and input the data from the first signal processing model 108 and the second fault signal 105 into the second signal processing model 111 in the second signal processing module 107 for signal processing to obtain second classification data 112 with higher accuracy.

[0064] After the first signal processing model 108 calculates the first fault signal 104 to obtain the first classified data 109, the second signal processing model 111 extracts relevant information from the data from the first signal processing model 108 and the second fault signal 105. In the present application, different faults of the 32-bit microcontroller are correlated. For example, when the 32-bit microcontroller generates a voltage overload fault, the current flowing through it increases. Due to the thermal effect of the current, the heat generated at this time increases, thereby causing a temperature abnormality. Therefore, the second fault signal 105 indicating the temperature abnormality and the data from the first signal processing model 108 indicating the voltage overload are correlated, such as a timing correlation.

[0065] In statistics and related fields, correlation is used to describe whether variables are related. A correlation function is a real-valued function that quantifies the correlation between variables. While there is no single definition for correlation metrics, parameters such as the Pearson correlation coefficient, mutual information, and the Spearman rank correlation coefficient are used to measure the dependency between variables. In signal processing, correlation metrics can be used to summarize the dependency between the data from the first signal processing model 108 and the second fault signal 105.

[0066] The second signal processing model 111 calculates a self-similarity measure for the second fault signal 105 by exploiting the dependencies between the data from the first signal processing model 108 and the second fault signal 105. This measure is calculated for arbitrary-length segments of data formed by combining the data from the first signal processing model 108 and the second fault signal 105 with high dependencies. Because the second signal processing model 111 requires higher-accuracy second classified data 112, its input is enriched by representing the correlations between the data from the first signal processing model 108 and the second fault signal 105. These relationships represent a global description of the input data in both space and time, with the second fault signal 105 describing the spatial characteristics of the input data and the combination of the data from the first signal processing model 108 and the second fault signal 105 describing the temporal characteristics of the input data. Therefore, the signal processing implemented by the second signal processing model 111 is based on the correlations between the information extracted from the data from the first signal processing model 108 and the second fault signal 105.

[0067] The first signal processing module 106 inputs the first fault signal 104 into the first signal processing model 108 for signal processing. This generates both intermediate data 110 generated during the signal processing process and first classified data 109 representing the classification result after the signal processing is completed. Therefore, both the intermediate data 110 and the first classified data 109 can be considered data from the first signal processing model 108. In particular, when the first signal processing model 108 is a neural network model, the intermediate data 110 can be one or more of the following: a fragment of the input layer output, output data of the hidden layer, and the probability that the first fault signal 104 belongs to each classification. When the neural network model has multiple hidden layers, the output data of the hidden layer includes at least one of the output data of all hidden layers. When the first signal processing model 108 includes a long short-term memory model, a self-attention model, and a classifier, the intermediate data 110 can be one or more of the following: a fragment of the input layer output, time series results, high-dimensional feature data, and the probability that the first fault signal 104 belongs to each classification.

[0068] In some possible implementations, the second signal processing model 111 may be a neural network model. Specifically, the neural network model includes an input layer, a hidden layer, and an output layer, wherein the input layer is used to receive data formed by combining the data from the first signal processing model 108 and the second fault signal 105 with a high dependency relationship. For example, the second fault signal 105 is spliced ​​with the data from the first signal processing model 108 to generate combined data, and then the combined data is cut into segments of any length, and these segments are input into the hidden layer. Furthermore, the input layer can be used to first normalize the second fault signal 105 and the data from the first signal processing model 108, and then combine the normalized data to form The combined data is then divided into segments of arbitrary length, and these segments are input into the hidden layer; the hidden layer is used to use the segments output by the input layer as input data, extract the similarity information between all possible segments pairs, and calculate the output data of the layer, which is used to represent the correlation between the data from the first signal processing model 108 and the information extracted from the second fault signal 105; the output layer is used to use the output data of the hidden layer as input data, and use normalized exponential functions, logistic functions and other functions to calculate, to obtain the probability that the second fault signal 105 belongs to each category, and select the category with the highest probability as the second classification data 112. In particular, when the neural network model has multiple hidden layers, the first hidden layer is used to use the segments output by the input layer as input data, extract the similarity information between all possible segments pairs, and calculate the output data of this layer; the second hidden layer is used to use the output data of the first hidden layer as input data, further extract the similarity information, and obtain the output data of this layer; the third hidden layer is used to use the output data of the second hidden layer as input data, and further extract the similarity information to obtain the output data of this layer; and so on, until the last hidden layer calculates the output data of this layer; and the output layer is used to use the output data of the last hidden layer as input data, and use normalized exponential function, logistic function and other functions to calculate, to obtain the probability that the second fault signal 105 belongs to each category, and select the category with the highest probability as the second classification data 112.

[0069] In some possible implementations, the second signal processing model 111 may include a long short-term memory model, a self-attention model, and a classifier. Specifically, the second signal processing module 107 inputs data formed by combining the data from the first signal processing model 108 and the second fault signal 105, which have a high dependency, into the long short-term memory model to calculate a time series result. Subsequently, the second signal processing module 107 inputs the time series result into the self-attention model to obtain high-dimensional feature data. The self-attention model is used to determine the correlation between the information extracted from the data from the first signal processing model 108 and the second fault signal 105. It includes a self-attention sub-model and a fully connected layer. The calculation process is to input the time series result into the self-attention sub-model to obtain multiple attention values, and then input these attention values ​​into the fully connected layer to obtain high-dimensional feature data. Finally, the second signal processing module 107 inputs the high-dimensional feature data into a classifier, such as a Softmax classifier or a logistic classifier, to obtain the probability that the second fault signal 105 belongs to each category, and selects the category with the highest probability as the second classified data 112.

[0070] It should be understood that the aforementioned LSTM model, self-attention model, and classifier can be three different neural network models, or they can be integrated into the same neural network model. When the LSTM model, self-attention model, and classifier are integrated into the same neural network model, the LSTM model, self-attention model, and classifier can be different layers in the neural network model.

[0071] In some possible implementations, the initial state of the second signal processing model 111 may be an intermediate state, a classified state, or a combination of the intermediate state and the classified state, when the first signal processing model 108 performs signal processing on the first fault signal 104. For example, the combination result is obtained by multiplying a state matrix representing the intermediate state with a state matrix representing the classified state. The intermediate state represents the state of the first signal processing model 108 during the signal processing of the first fault signal 104 by the first signal processing model 108, and the classified state represents the state of the first signal processing model 108 after the first signal processing model 108 calculates the first classified data 109 corresponding to the first fault signal 104.

[0072] In some possible implementations, the processor 103 further includes a backup module to replace the first signal processing module 106 or the second signal processing module 107 when a failure occurs.

[0073] Based on the aforementioned multi-model microcontroller signal processing device 100 , a multi-model microcontroller signal processing method provided by the present application is described in detail below.

[0074] See also Figure 2 , Figure 2 This is a flow chart of a multi-model microcontroller signal processing method provided by this application. Figure 2 As shown, the multi-model microcontroller signal processing method provided by the present application includes:

[0075] S201: The multi-model microcontroller signal processing device obtains a first fault signal through a first pin and obtains a second fault signal through a second pin.

[0076] Multi-mode microcontroller signal processing devices can be Figure 1 In the multi-model microcontroller signal processing device 100, the first pin may be Figure 1 In the first pin 101 of the multi-model microcontroller signal processing device 100, the first fault signal may be Figure 1 The first fault signal 104, the second pin can be Figure 1 The second fault signal may be the second pin 102 of the microcontroller signal processing device 100 of the multi-model. Figure 1 The second fault signal 105 in.

[0077] The first fault signal may be caused by an internal fault or an external fault of the 32-bit microcontroller. The second fault signal may be caused by an internal fault or an external fault of the 32-bit microcontroller. Furthermore, the second fault signal and the first fault signal may be caused by different faults.

[0078] S202: The multi-model microcontroller signal processing device inputs the first fault signal into the first signal processing model in the first signal processing module for signal processing to obtain intermediate data and first classification data.

[0079] The first signal processing module may be Figure 1 The first signal processing module 106 of the multi-model microcontroller signal processing device 100, the first signal processing model can be Figure 1 In the first signal processing model 108 of the multi-model microcontroller signal processing device 100, the intermediate data can be Figure 1 The intermediate data 110 of the microcontroller signal processing device 100 of the multi-model, the first classification data can be Figure 1 The first classification data 109 of the multi-model microcontroller signal processing device 100 is shown.

[0080] After the multi-model microcontroller signal processing device acquires the first fault signal via the first pin, it inputs the first fault signal into the first signal processing model within the first signal processing module for signal processing. The first signal processing model relies on a self-similarity metric for the first fault signal. This metric integrates the interactions between different data components and is calculated using segments of the first fault signal of arbitrary length. Therefore, the signal processing implemented by the first signal processing model is based on the spatial correlation of the information extracted from the first fault signal. The classification in the first signal processing model refers to the classification in the first signal processing model 108 described above. The first signal processing model can calculate the probability that the first fault signal belongs to each classification and select the classification with the highest probability as the first classification data. The data generated during the calculation process is intermediate data.

[0081] In some possible implementations, the first signal processing model may be a neural network model comprising an input layer, a hidden layer, and an output layer. The input layer is configured to receive the first fault signal, segment it into segments of arbitrary length, and input these segments into the hidden layer. The hidden layer is configured to receive the segments output by the input layer, extract similarity information between all possible segments, and calculate output data representing the spatial correlation of the information extracted from the first fault signal. The output layer is configured to receive the output data of the hidden layer, calculate the probability that the first fault signal belongs to each classification, and select the classification with the highest probability as the first classification data. Furthermore, one or more of the segments output by the input layer, the output data of the hidden layer, and the probability that the first fault signal belongs to each classification may be considered intermediate data.

[0082] In some possible implementations, the first signal processing model may include a long short-term memory model, a self-attention model, and a classifier. The long short-term memory model is used to receive the first fault signal and calculate a time series result. The self-attention model is used to receive the time series result and calculate high-dimensional feature data, thereby extracting spatial correlation information from the first fault signal. The classifier is used to receive the high-dimensional feature data, obtain the probability that the first fault signal belongs to each category, and select the category with the highest probability as the first classification data. Furthermore, one or more of the following: the fragments output by the input layer, the time series result, the high-dimensional feature data, and the probability that the first fault signal belongs to each category, are considered intermediate data.

[0083] It should be understood that the aforementioned LSTM model, self-attention model, and classifier can be three different neural network models, or they can be integrated into the same neural network model. When the LSTM model, self-attention model, and classifier are integrated into the same neural network model, the LSTM model, self-attention model, and classifier can be different layers in the neural network model.

[0084] S203: The multi-model microcontroller signal processing device inputs at least one of the intermediate data and the first classification data, and the second fault signal into the second signal processing model in the second signal processing module for signal processing.

[0085] The second signal processing module may be Figure 1 The second signal processing module 107 of the microcontroller signal processing device 100 of the multi-model, the second signal processing model can be Figure 1 The second signal processing model 111 of the multi-model microcontroller signal processing device 100.

[0086] After the first signal processing module calculates the first fault signal to obtain intermediate data and first classification data, the multi-model microcontroller signal processing device inputs at least one of the intermediate data and the first classification data of the first signal processing module and the second fault signal into the second signal processing model in the second signal processing module for signal processing, thereby obtaining second classification data with higher accuracy. The second classification data can be Figure 1 The second classification data 112 of the multi-model microcontroller signal processing device 100 is shown.

[0087] The second signal processing model calculates a self-similarity measure for the second fault signal by exploiting a dependency relationship between data from the first signal processing module and a second fault signal, wherein the data from the first signal processing module includes at least one of intermediate data and first classified data. This measure is found by calculating arbitrary length segments of data formed by combining the data from the first signal processing module and the second fault signal with a high dependency relationship. Since the second signal processing model needs to obtain second classified data with higher accuracy, its input is enriched by representing the correlation between the data from the first signal processing model and the second fault signal. Therefore, the signal processing implemented by the second signal processing model is performed based on the correlation of information extracted from the data from the first signal processing model and the second fault signal.

[0088] In some possible implementations, the second signal processing model can be a neural network model, including an input layer, a hidden layer and an output layer, wherein the input layer is used to receive data formed by combining data from the first signal processing module and the second fault signal with a high dependency relationship, and then cut the combined data into segments of arbitrary length, and input these segments into the hidden layer; the hidden layer is used to receive the segments output by the input layer, extract the similarity information between all possible pairs of segments, and calculate the output data representing the correlation between the data from the first signal processing module and the information extracted from the second fault signal; the output layer is used to receive the output data of the hidden layer, use normalized exponential function, logistic function and other functions to calculate, obtain the probability that the second fault signal belongs to each category, and select the category with the highest probability as the second classification data.

[0089] In some possible implementations, the second signal processing model may include a long short-term memory model, a self-attention model, and a classifier, wherein the long short-term memory model is used to receive data formed by combining data from the first signal processing module and the second fault signal with a high dependency relationship, and calculate the time series results; the self-attention model is used to receive the time series results and calculate the high-dimensional feature data, thereby realizing the correlation of information extracted from the data from the first signal processing module and the second fault signal; the classifier is used to receive the high-dimensional feature data, thereby obtaining the probability that the second fault signal belongs to each category, and select the category with the highest probability as the second classification data.

[0090] It should be understood that the aforementioned LSTM model, self-attention model, and classifier can be three different neural network models, or they can be integrated into the same neural network model. When the LSTM model, self-attention model, and classifier are integrated into the same neural network model, the LSTM model, self-attention model, and classifier can be different layers in the neural network model.

[0091] In some possible implementations, the initial state of the second signal processing model may be an intermediate state, a classified state, or a combination of the intermediate state and the classified state, resulting from the first signal processing model performing signal processing on the first fault signal. For example, the combination result is obtained by multiplying a state matrix representing the intermediate state with a state matrix representing the classified state. The intermediate state represents the state of the first signal processing model during the first signal processing model's signal processing of the first fault signal, and the classified state represents the state of the first signal processing model after the first signal processing model calculates first classified data corresponding to the first fault signal.

[0092] In some possible implementations, when a failure occurs in the first signal processing module or the second signal processing module, a backup module is used to replace the first signal processing module or the second signal processing module.

[0093] It should be understood that in the embodiments of the present invention, "when...", "at...", "if" and "if" all mean that the terminal device or access network device will make corresponding processing under certain objective circumstances, and it is not a time limit, nor does it require the terminal device or access network device to make a judgment action when implementing it, nor does it mean that there are other limitations.

[0094] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of a multi-model microcontroller signal processing device provided by this application. This device can be used to implement the aforementioned multi-model microcontroller signal processing method. Figure 3 As shown, the multi-model microcontroller signal processing device 300 includes: an acquisition module 301 , a first processing module 302 and a second processing module 303 .

[0095] The acquisition module 301 is used to acquire a first fault signal and a second fault signal;

[0096] The first processing module 302 is used to input the first fault signal into the first signal processing model for signal processing to obtain intermediate data and first classification data;

[0097] The second processing module 303 is used to input at least one of the intermediate data and the first classified data, and the second fault signal into a second signal processing model for signal processing.

[0098] Among them, the above-mentioned intermediate data is the data generated in the above-mentioned signal processing process, the above-mentioned first classification data is the result of signal processing performed by the above-mentioned first signal processing model based on the above-mentioned first fault signal, and the dependency relationship between at least one of the above-mentioned intermediate data and the first classification data and the above-mentioned second fault signal is obtained by calculating the correlation measure.

[0099] The acquisition module 301, the first processing module 302 and the second processing module 303 can be stored in Figure 1 In the processor 103 of the microcontroller signal processing device 100 of the multi-model.

[0100] In some possible implementations, the first signal processing model and the second signal processing model each include a neural network model.

[0101] In some possible implementations, the first signal processing model and the second signal processing model both include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

[0102] In some possible implementations, the initial state of the second signal processing model is implemented based on at least one of the intermediate state and the classification state of the first signal processing model.

[0103] Among them, the above-mentioned intermediate state represents the state of the first signal processing model when the above-mentioned first signal processing model performs signal processing on the above-mentioned first fault signal, and the above-mentioned classification state represents the state of the first signal processing model when the above-mentioned first signal processing model calculates the first classification data corresponding to the above-mentioned first fault signal.

[0104] In some possible implementations, the above device further includes a backup module.

[0105] The backup module is used to replace the first processing module 302 or the second processing module 303 when a failure occurs in the first processing module 302 or the second processing module 303 .

[0106] The present application also provides a chip comprising a processor and a memory. The processor may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MCU), or a digital signal processor (DSP). The memory may include any one or more of a volatile memory, a non-volatile memory, or other types of memory. The memory stores executable program code, which the processor executes to implement the functions of the aforementioned acquisition module 301, the first processing module 302, and the second processing module 303, thereby implementing a multi-model microcontroller signal processing method.

[0107] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions for instructing a computing device to execute a multi-model microcontroller signal processing method.

[0108] Applicants hereby independently disclose each individual feature described herein and any combination of two or more such features. As long as such features or combinations can be implemented according to the common sense of a person skilled in the art based on this specification as a whole, regardless of whether such features or combinations of features solve any problem disclosed herein, the scope of the claims is not limited. Applicants indicate that various aspects of the present invention may include any such individual features or combinations of features. In view of the foregoing description, it will be apparent to those skilled in the art that various modifications may be made within the scope of the present invention.

Claims

1. A multi-model microcontroller signal processing device, characterized in that: include: A first pin, used for inputting a first fault signal; The second pin is used to input a second fault signal; The processor includes a first signal processing module and a second signal processing module, wherein the first pin is directly connected to the first signal processing module, and the second pin is directly connected to the second signal processing module. The first signal processing module is used to receive a first fault signal input by the first pin and input the first fault signal into a first signal processing model for signal processing. The second signal processing module is used to receive a second fault signal input by the second pin, input the intermediate data and at least one of the first classification data in the first signal processing model, and the second fault signal into the second signal processing model for signal processing, obtain the probability that the second fault signal belongs to each classification, and select the classification with the highest probability as the second classification data. The intermediate data is the data generated during the signal processing process. , the first classification data is the result of the first signal processing model performing signal processing according to the first fault signal, and the dependency relationship between at least one of the intermediate data and the first classification data and the second fault signal is obtained by calculating the correlation measure; the initial state of the second signal processing model is realized according to at least one of the intermediate state and the classification state of the first signal processing model, wherein the intermediate state represents the state of the first signal processing model during the signal processing of the first fault signal by the first signal processing model, and the classification state represents the state of the first signal processing model when the first signal processing model calculates the first classification data corresponding to the first fault signal; the first fault signal and the second fault signal come from different faults.

2. The multi-model microcontroller signal processing device according to claim 1, characterized in that: Both the first signal processing model and the second signal processing model include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

3. The multi-model microcontroller signal processing device according to claim 1 or 2, characterized in that: The processor further includes a backup module, which is used to replace the first signal processing module and the second signal processing module when any one of the first signal processing module and the second signal processing module fails.

4. A multi-model microcontroller signal processing method, characterized in that: include: The multi-model microcontroller signal processing device obtains a first fault signal through a first pin and obtains a second fault signal through a second pin; The multi-model microcontroller signal processing device inputs the first fault signal into the first signal processing model in the first signal processing module for signal processing to obtain intermediate data and first classified data, wherein the intermediate data is data generated during the signal processing process, and the first classified data is the result of the signal processing performed by the first signal processing model based on the first fault signal; The multi-model microcontroller signal processing device inputs the intermediate data and at least one of the first classification data, and the second fault signal into a second signal processing model in a second signal processing module for signal processing, obtains the probability that the second fault signal belongs to each classification, and selects the classification with the highest probability as the second classification data, and the dependency relationship between at least one of the intermediate data and the first classification data and the second fault signal is obtained by calculating a correlation measure; the initial state of the second signal processing model is achieved based on at least one of the intermediate state and the classification state of the first signal processing model, wherein the intermediate state represents the state of the first signal processing model during the process of the first signal processing model performing signal processing on the first fault signal, and the classification state represents the state of the first signal processing model after the first signal processing model calculates the first classification data corresponding to the first fault signal; The first fault signal and the second fault signal come from different faults.

5. The multi-model microcontroller signal processing method according to claim 4, characterized in that: The first signal processing model and the second signal processing model each include a neural network model.

6. The multi-model microcontroller signal processing method according to claim 5, characterized in that: Both the first signal processing model and the second signal processing model include a self-attention model, the self-attention model of the first signal processing model is used to determine the spatial correlation of information extracted from the first fault signal, and the self-attention model of the second signal processing model is used to determine the correlation of information extracted from at least one of the intermediate data and the first classified data in the first signal processing model, and the second fault signal.

7. A chip, characterized in that: The chip comprises a processor and a memory, wherein the processor is configured to execute instructions stored in the memory, so that the chip executes the method according to any one of claims 4 to 6.

8. A computer-readable storage medium, characterized in that The method comprises computer program instructions, and when the computer program instructions are executed by a computing device, the computing device performs the method according to any one of claims 4 to 6.

Images