A TCSNN-based degradation prediction system for aircraft engine gas path components

Abstract

A TCSNN-based aircraft engine gas path component degradation prediction system relates to the field of artificial intelligence. The system includes a data input module, a performance prediction module, and a data output module deployed on an FPGA. The system predicts exhaust temperature, a metric reflecting overall engine performance, based on gas path performance parameters of multiple components. The system utilizes a TCSNN network model and improves the TCN to enhance the network model's anti-interference capability and prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a TCSNN-based aero-engine gas path component degradation prediction system. Background Art

[0002] Aircraft engine performance prediction and trend analysis analyzes current and historical data to predict future engine trends and potential failures in key components. Aircraft engine systems are extremely complex, and their primary failure types include airflow failures, mechanical failures, control system failures, and vibration failures. The airflow system regulates and controls airflow to key components such as the compressor, combustor, and turbine, and its stability directly impacts engine efficiency and safety. During cruise mode, key engine airflow performance parameters include thrust, fuel consumption, and exhaust temperature. Trends in these key indicators can reflect engine performance degradation. However, these performance parameters are difficult to obtain or measure in real time. Therefore, predicting engine health and efficiency using directly measurable airflow component performance parameters has become a key research topic. Accurate trend prediction can promptly identify signs of performance degradation, enabling more informed maintenance and repair strategies.

[0003] With the advancement of aviation technology, traditional data analysis methods based on experience and maintenance history are no longer able to meet the complexity and high-precision requirements of modern aircraft engines. Data-driven models and algorithms utilize neural network models to predict aircraft engine performance in real time, avoiding the complexity of traditional physical modeling. The characteristics of time-series data collected by sensors require efficient time-series modeling capabilities for the neural network models used. Among various data-driven methods, the Temporal Convolutional Network (TCN) has been widely used in airflow performance prediction due to its powerful time-series modeling capabilities. Using causal convolution and dilated convolution structures, TCNs can effectively capture long-term dependencies. However, practical applications of TCNs still face some challenges. TCNs require high computational resources, especially when processing large amounts of data, where computational efficiency can become a bottleneck. When dealing with nonlinear dynamics, they also struggle to fully capture complex time-series characteristics, resulting in limited prediction accuracy. Furthermore, traditional TCN models deployed on CPU (Central Processing Unit) / GPU (Graphics Processing Unit) platforms cannot meet the real-time requirements of engine performance prediction.

[0004] By introducing spiking neurons into TCN, computational tasks can be sparser, reducing resource usage while maintaining high sensitivity to dynamic changes in timing signals. TCNs combined with spiking neurons are more suitable for hardware acceleration via FPGAs (Field Programmable Gate Arrays) in terms of resource usage and power consumption. Compared to neural network models deployed on CPU / GPU platforms, they also achieve higher computational parallelism, enabling real-time prediction of engine performance. Therefore, exploring a system for predicting degradation of aircraft engine air path components based on a temporal convolutional spiking neural network (TCSNN) is of practical significance. Summary of the Invention

[0005] This paper proposes a TCSNN-based system for predicting the degradation of aircraft engine gas path components. This system uses time series signals of aircraft engine gas path performance parameters as a basis to predict overall engine performance. This paper improves on the temporal convolutional network (TCN), leveraging causal convolution and residual connections to capture long-term dependencies in the data and employing spiking neurons to encode temporal information, thereby further improving temporal resolution and learning capabilities, and enhancing the accuracy of network model predictions. The TCSNN-based system for predicting the degradation of aircraft engine gas path components is deployed on an FPGA, utilizing hardware acceleration technology to meet the real-time detection requirements of aircraft engines.

[0006] The technical solution of the present invention:

[0007] A TCSNN-based aero-engine gas path component degradation prediction system, comprising a data input module, a performance prediction module, and a data output module deployed on an FPGA;

[0008] The data input module includes a sensor module, a UART serial port receiving module, a data processing module, and a line buffer RAM module. The composition and specific functions of each module are as follows:

[0009] Sensor module: includes various types of sensors, used to collect raw data of the parameters of each gas path component changing over time;

[0010] UART serial port receiving module: used to receive the raw data of the parameters of each gas path component changing over time collected by the sensor module, and generate characteristic data after parsing according to the UART protocol;

[0011] Data processing module: This module processes the characteristic data generated by the UART serial port receiving module according to the UART protocol analysis to obtain a characteristic sequence. The data processing module is deployed on the PS side of the FPGA. The processed characteristic sequence is transmitted from the PS (Processing System) side to the PL (Programmable Logic) side via the AXI bus.

[0012] Row cache RAM module: used to cache the feature sequence processed by the data processing module;

[0013] The performance prediction module includes a weight loading module, a weight ROM module, a time convolution network calculation module, a data flow control module, a pulse neuron calculation module, and a fully connected calculation module. The composition and specific functions of each module are as follows:

[0014] Weight loading module: The temporal convolutional network calculation module sends the weight address index to the weight loading module. The weight address index includes the count of convolution cycles within per layer (C3PL) and the count of convolutional layer cycles (CCLC). The weight loading module generates the address of the weight parameter to be read according to the weight address index.

[0015] Weight ROM module: used to write the weight parameters obtained by TCSNN training into the BRAM (Block RAM) of FPGA in the form of COE file, and output the required weight parameters to the temporal convolutional network calculation module according to the address of the weight parameters generated by the weight loading module;

[0016] Temporal convolutional network computing module: including input data RAM (IDRAM), convolution results RAM (CRRAM), and convolution-residual general calculation (C-RGC) units;

[0017] Input data cache RAM unit: used to cache the characteristic sequence from the row cache RAM module in the data input module;

[0018] Convolution calculation result cache RAM unit: used to cache the intermediate result data after the convolution and residual general calculation units perform convolution calculations and residual calculations;

[0019] Convolution and residual general computing unit: According to the shape of TCSNN and the DSP (Digital Signal Processor) and LUT (Look-Up Table) resources of FPGA, the convolution calculation parallelism is designed. The convolution and residual general computing unit reads the feature sequence in the input data cache RAM unit and the intermediate result data in the convolution calculation result cache RAM unit according to the designed convolution calculation parallelism, and reads the weight parameters in the weight ROM module that matches the feature sequence to perform convolution calculation and residual calculation. During the convolution calculation and residual calculation process, the convolution calculation process of each layer needs to continuously update the convolution cycle count signal in the single layer to refresh the weight parameters. After the convolution calculation of each layer is completed, the overall cycle count signal of the convolution layer is updated, and the intermediate result number of the convolution calculation of the current layer is updated. The data is cached in the convolution calculation result cache RAM unit. When the convolution calculation of the next layer starts, the intermediate result data of the convolution calculation of the previous layer is directly read from the convolution calculation result cache RAM unit to perform the convolution calculation of the next layer. When the convolution calculation of all layers is completed, the convolution and residual general calculation unit performs the residual calculation: read the feature sequence from the input data cache RAM unit, perform convolution calculation on the feature sequence and the refreshed weight parameter, and perform weighted addition on the result of the convolution calculation and the convolution calculation result of the last layer that has been cached in the convolution calculation result cache RAM unit. At this point, all calculations of the entire time convolution network calculation module are completed.

[0020] Pulse neuron calculation module: The pulse neuron calculation module includes Z pulse neurons, according to the time step parameters designed for each pulse neuron Build Each pipeline structure is constructed by a set of LIF (Leaky Integrate and Fired) model operators. The membrane potential of the initial cache in the first-level pipeline structure is set to zero, and the membrane potential is transmitted to the next-level pipeline structure or an output pulse is generated according to the result signal output by the time convolution network calculation module; the pulse neuron calculation module converts the result signal output by the time convolution network calculation module into a pulse signal based on the firing timing of the pulse neuron; the LIF model operator is explained by the following formula:

[0021] (1)

[0022] (2)

[0023] Where: represents the time constant of the membrane, represents the membrane potential of the spiking neuron, represents the resting potential of the spiking neuron, represents the membrane resistance, Indicates the external input current over time The change of; Formula (1) describes the influence of the membrane potential on the leakage effect of the resting potential and the external input current; represents the time when the spiking neuron fires a pulse, Indicates the time it takes for a spike neuron to approach 0 after it emits a spike The membrane potential of the endoplasmic reticulum neuron, represents the reset value of the membrane potential after the spiking neuron emits a pulse; Formula (2) describes the After the pulse is released at the moment, it will approach 0 Resets its membrane potential to ;

[0024] Data flow control module: The pulse signal output by the pulse neuron calculation module needs to be repeatedly input C times into the time convolution network calculation module for convolution calculation and residual calculation, and the pulse neuron calculation module is converted into a pulse signal. After the last calculation of the time convolution network calculation module is completed, it is processed by the pulse neuron calculation module and converted into a pulse signal before being handed over to the fully connected calculation module for calculation; the data flow control module controls the output flow between each module in the performance prediction module based on the convolution cycle count signal within a single layer and the overall convolution layer cycle count signal;

[0025] Fully connected computing module: The fully connected computing module is responsible for mapping the pulse signal obtained by the pulse neuron computing module for the last time to the final output space, and mapping the pulse signal to the predicted value of the exhaust temperature through a two-layer fully connected network.

[0026] The data output module includes a UART serial port sending module and a threshold judgment module;

[0027] UART serial port sending module: used to package the predicted value of exhaust temperature obtained by the performance prediction module according to the UART protocol and send it to the host computer through the serial port to display the predicted value of exhaust temperature and safety alarm;

[0028] Threshold judgment module: used to monitor whether the predicted value of exhaust temperature is higher than the safety threshold. If the predicted value of exhaust temperature is higher than the safety threshold, a safety alarm will be issued.

[0029] Before deploying the TCSNN-based aircraft engine gas path component degradation prediction system on the FPGA, the TCSNN needs to be trained. The process is as follows:

[0030] Step 1: Using sensors to collect raw data on the time-varying parameters of various air path components in an aircraft engine maintained in cruise mode; preprocessing the raw data, including normalization, noise reduction, and data enhancement, to obtain the preprocessed air path performance parameters; performing a correlation analysis between each air path performance parameter and the exhaust temperature, and using each air path performance parameter with a correlation function value greater than a set threshold as a feature sequence input into the TCSNN; and dividing the feature sequence into a training set, a validation set, and a test set in proportion;

[0031] Furthermore, the step 1 specifically includes the following steps:

[0032] Step 1.1: Use sensors to collect raw data on the time-varying low-pressure rotor speed, high-pressure rotor speed, high-pressure compressor outlet pressure, oil pressure, compressor outlet temperature, and lead angle of the aircraft engine at a fixed sampling frequency while the aircraft engine is in cruise mode.

[0033] Step 1.2: Use normalization to scale the raw data to The data sequence between them is processed; the wavelet packet threshold denoising method is used to denoise the normalized data sequence to obtain a denoised data sequence; the sliding overlapping sampling method is used to enhance the denoised data sequence;

[0034] Furthermore, step 1.2 specifically includes the following steps:

[0035] Step 1.2.1: Normalize the original data and scale it to The number of samples is Data series ;

[0036] Data Series Each of the The calculation formula is as follows:

[0037]

[0038] in, is the original data, is the sequence of original data, is the maximum value in the sequence of original data, is the minimum value in the sequence of original data;

[0039] Step 1.2.2, use The wavelet basis is used to perform normalized data sequence Layer wavelet decomposition, we get Layer wavelet coefficients, where the wavelet coefficients obtained by wavelet decomposition of each layer contain approximate coefficients (low-frequency components) and detail coefficients (high-frequency components); The detail coefficients of the layer are screened using a soft threshold function; The approximate coefficients of the layers and the soft threshold function are filtered from the first layer to the The detail coefficients of the layer are inversely transformed by wavelet and reconstructed to obtain the denoised data sequence ;in, for The data after noise reduction;

[0040] Step 1.2.3: Use sliding overlapping sampling method and fixed sample length and overlap length The denoised data sequence is converted into Split into A sample sequence, recorded as:

[0041]

[0042] in, For the denoised data sequence Middle The denoised data of the item;

[0043] Step 1.3: Calculate the mutual information function value between each gas path performance parameter processed in step 1.2 and the exhaust temperature. When the mutual information function value is greater than the set threshold, it is used as the feature sequence input to the TCSNN. When the mutual information function value is less than the set threshold, it is not used as the feature sequence of the TCSNN.

[0044] Mutual information function between each gas path performance parameter and exhaust temperature The expression is:

[0045]

[0046] Where, is the denoised data sequence, Exhaust temperature series , 、 and They are 、 and The entropy of , and:

[0047]

[0048]

[0049]

[0050] Where, express The probability density of express The probability density of for and The joint probability density of

[0051] Step 1.4: Divide the feature sequence obtained in step 1.3 into training set, validation set and test set in proportion;

[0052] Step 2: Build and train the temporal convolutional spike neural network (TCSNN).

[0053] TCSNN adds a spike neuron computing module to the temporal convolutional network (TCN). The TCN contains a C-layer residual network structure, each of which consists of two convolutional layers and one residual layer. The C-layer residual network structure is used to extract features and perform temporal modeling on the input feature sequence. The spike neuron computing module includes X spike neurons, each of which is placed between two adjacent residual network layers to act as a connection. At the same time, based on the firing timing of the spike neurons, the continuous signal output by the residual network structure is converted into a spike signal input and transmitted to the spike neurons, further optimizing the prediction results.

[0054] The training process of TCSNN is as follows: first, the hyperparameters and weights of TCSNN are initialized; then, the training set is input into TCSNN in batches for training until TCSNN converges; and the optimal TCSNN is obtained by adjusting the hyperparameters. The details are as follows:

[0055] Step 2.1: Initialize the hyperparameters and weights of TCSNN;

[0056] Step 2.2: Input the training set into TCSNN in batches and output the predicted value of the exhaust temperature of the aircraft engine. Compare the predicted value of the exhaust temperature with the actual value, calculate the loss value Loss through the Mean Squared Error (MSE) function, and calculate the loss function through the chain rule. Weights for convolutional and residual layers The gradient of is used to back-propagate the gradient descent algorithm to update the weights of each convolutional layer and residual layer of TCSNN;

[0057] Furthermore, the specific formula of the loss value Loss is:

[0058]

[0059] in, is the number of samples, is the actual value, is the predicted value of TCSNN;

[0060] Furthermore, the loss function Weight The gradient formula is:

[0061]

[0062] Step 2.3: Determine the loss value Is it less than the threshold? , when the loss value Less than threshold When , TCSNN is considered to be optimal and training is stopped. Otherwise, the hyperparameters are adjusted and step 2.2 is repeated.

[0063] Step 3: Evaluate the storage and computing resource requirements of TCSNN on FPGA; load the weight parameters obtained from training TCSNN into FPGA in COE file format for storage;

[0064] Step 3.1: Evaluate the memory and logic resource usage of the TCSNN and its deployment based on the data format of the weight parameters used in the FPGA deployment. On-chip DSP resources are used to calculate tensor multiplications in convolution and matrix calculations, and LUTs are used to build the basic units of the TCSNN.

[0065] Step 3.2: Write the weight parameters obtained from TCSNN training into the COE file for reading by the weight ROM module on the FPGA for forward propagation of TCSNN.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] This paper proposes a TCSNN-based prediction method for aero-engine gas path component degradation. The system predicts the exhaust temperature, which reflects the overall engine performance, based on the gas path performance parameters of multiple components. The system uses the TCSNN network model and improves the TCN to enhance the network model's anti-interference ability and prediction accuracy.

[0068] The data preprocessing phase incorporates dual optimization using wavelet packet threshold noise reduction and mutual information feature screening. A fifth-order Daubechies wavelet basis is used to perform multi-scale decomposition of the original signal, and a soft threshold function is used to precisely remove high-frequency noise, improving the signal-to-noise ratio. The mutual information method selects characteristic parameters that are strongly correlated with exhaust temperature (threshold > 0.9), eliminating redundant interference and improving the model's predictive stability under complex operating conditions.

[0069] The network model used in the present invention uses causal convolution and unidirectional convolution kernel design to ensure the causality of time series signal processing and avoid information leakage problems. Residual connections are used to solve the gradient vanishing problem in deep networks through cross-layer feature fusion, thereby efficiently extracting time series features and capturing long-term dependencies in data. Spiking neurons are modeled based on the dynamic changes of time series signals and use their biologically inspired dynamic threshold mechanism to trigger pulse signals only in high-information areas, significantly reducing redundant calculations and further improving temporal resolution and learning ability. At the same time, sparse activation is used to reduce computational complexity.

[0070] The system is deployed on an FPGA to achieve low-latency computing. When performing convolution calculations, a ring RAM storage structure is designed so that all intermediate calculation results share the same on-chip RAM resources, which can greatly reuse and save on-chip RAM resources and reduce the resource overhead caused by frequent data transfer. The spiking neuron computing module adopts a cascade pipeline design, which parallelizes the membrane potential updates at multiple time steps to avoid the timing blocking problems in traditional serial computing.

[0071] Combined with the parallel computing capabilities of FPGA, the system can complete the synchronous processing of multi-channel data within a single clock cycle, with an overall prediction delay of less than 5ms, meeting the real-time monitoring needs of aircraft engines. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a framework diagram of the TCSNN-based aero-engine gas path component degradation prediction system of the present invention;

[0073] Figure 2 Schematic diagram of the network model structure of TCSNN of the present invention;

[0074] Figure 3 The deployment diagram of the TCSNN network model of the present invention on the FPGA development board, wherein (a) is a schematic diagram of the deployment of the performance prediction module, and (b) is a schematic diagram of the deployment of the temporal convolutional network module;

[0075] Figure 4 A schematic diagram of the ring RAM storage structure design of the present invention;

[0076] Figure 5 This is a comparison chart of the true value and the model value of the present invention. DETAILED DESCRIPTION

[0077] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.

[0078] The present invention proposes a TCSNN-based aero-engine gas path component degradation prediction system. Figure 1As shown, the TCSNN-based aircraft engine gas path component degradation prediction system includes a data input module, a performance prediction module and a data output module deployed on the FPGA;

[0079] The data input module includes a sensor module, a UART serial port receiving module, a data processing module, and a line buffer RAM module. The composition and specific functions of each module are as follows:

[0080] Sensor module: includes various types of sensors, used to collect raw data of the parameters of each gas path component changing over time;

[0081] UART serial port receiving module: used to receive the raw data of the parameters of each gas path component changing over time collected by the sensor module, and generate characteristic data after parsing according to the UART protocol;

[0082] Data processing module: This module processes the characteristic data generated by the UART serial port receiving module according to the UART protocol analysis to obtain a characteristic sequence. The data processing module is deployed on the PS side of the FPGA. The processed characteristic sequence is transmitted from the PS (Processing System) side to the PL (Programmable Logic) side via the AXI bus.

[0083] Row cache RAM module: used to cache the feature sequence processed by the data processing module;

[0084] The performance prediction module: Figure 3 As shown in (a), it includes a weight loading module, a weight ROM module, a temporal convolutional network calculation module, a data flow control module, a pulse neuron calculation module, and a fully connected calculation module. The composition and specific functions of each module are as follows:

[0085] Weight loading module: The temporal convolutional network calculation module sends the weight address index to the weight loading module. The weight address index includes the count of convolution cycles within per layer (C3PL) and the count of convolutional layer cycles (CCLC). The weight loading module generates the address of the weight parameter to be read according to the weight address index.

[0086] Weight ROM module: used to write the weight parameters obtained by TCSNN training into the BRAM (Block RAM) of FPGA in the form of COE file, and output the required weight parameters to the temporal convolutional network calculation module according to the address of the weight parameters generated by the weight loading module;

[0087] Time convolution network computing module: such as Figure 3 As shown in (b), it includes the input data cache RAM unit (InputData RAM, IDRAM), the convolution calculation result cache RAM unit (Convolution Results RAM, CRRAM) and the convolution and residual general calculation (Convolution-Residual General Calculation, C-RGC) unit;

[0088] Input data cache RAM unit: used to cache the characteristic sequence from the row cache RAM module in the data input module;

[0089] Convolution calculation result cache RAM unit: used to cache the intermediate result data after the convolution and residual general calculation units perform convolution calculations and residual calculations;

[0090] Convolution and residual general computing unit: According to the shape of TCSNN and the DSP (Digital Signal Processor) and LUT (Look-Up Table) resources of FPGA, the convolution calculation parallelism is designed. The convolution and residual general computing unit reads the feature sequence in the input data cache RAM unit and the intermediate result data in the convolution calculation result cache RAM unit according to the designed convolution calculation parallelism, and reads the weight parameters in the weight ROM module that matches the feature sequence to perform convolution calculation and residual calculation. During the convolution calculation and residual calculation process, the convolution calculation process of each layer needs to continuously update the convolution cycle count signal in the single layer to refresh the weight parameters. After the convolution calculation of each layer is completed, the overall cycle count signal of the convolution layer is updated, and the intermediate result number of the convolution calculation of the current layer is updated. The data is cached in the convolution calculation result cache RAM unit. When the convolution calculation of the next layer starts, the intermediate result data of the convolution calculation of the previous layer is directly read from the convolution calculation result cache RAM unit to perform the convolution calculation of the next layer. When the convolution calculation of all layers is completed, the convolution and residual general calculation unit performs the residual calculation: read the feature sequence from the input data cache RAM unit, perform convolution calculation on the feature sequence and the refreshed weight parameter, and perform weighted addition on the result of the convolution calculation and the convolution calculation result of the last layer that has been cached in the convolution calculation result cache RAM unit. At this point, all calculations of the entire time convolution network calculation module are completed.

[0091] Pulse neuron calculation module: The pulse neuron calculation module includes Z pulse neurons, according to the time step parameters designed for each pulse neuron Build Each pipeline structure is constructed by a set of LIF (Leaky Integrate and Fired) model operators. The membrane potential of the initial cache in the first-level pipeline structure is set to zero, and the membrane potential is transmitted to the next-level pipeline structure or an output pulse is generated according to the result signal output by the time convolution network calculation module; the pulse neuron calculation module converts the result signal output by the time convolution network calculation module into a pulse signal based on the firing timing of the pulse neuron; the LIF model operator is explained by the following formula:

[0092] (1)

[0093] (2)

[0094] Where: represents the time constant of the membrane, represents the membrane potential of the spiking neuron, represents the resting potential of the spiking neuron, represents the membrane resistance, Indicates the external input current over time The change of; Formula (1) describes the influence of the membrane potential on the leakage effect of the resting potential and the external input current; represents the time when the spiking neuron fires a pulse, Indicates the time it takes for a spike neuron to approach 0 after it emits a spike The membrane potential of the endoplasmic reticulum neuron, represents the reset value of the membrane potential after the spiking neuron emits a pulse; Formula (2) describes the After the pulse is released at the moment, it will approach 0 Resets its membrane potential to ;

[0095] Data flow control module: The pulse signal output by the pulse neuron calculation module needs to be repeatedly input C times into the time convolution network calculation module for convolution calculation and residual calculation, and the pulse neuron calculation module is converted into a pulse signal. After the last calculation of the time convolution network calculation module is completed, it is processed by the pulse neuron calculation module and converted into a pulse signal before being handed over to the fully connected calculation module for calculation; the data flow control module controls the output flow between each module in the performance prediction module based on the convolution cycle count signal within a single layer and the overall convolution layer cycle count signal;

[0096] Fully connected computing module: The fully connected computing module is responsible for mapping the pulse signal obtained by the pulse neuron computing module for the last time to the final output space, and mapping the pulse signal to the predicted value of the exhaust temperature through a two-layer fully connected network.

[0097] The data output module includes a UART serial port sending module and a threshold judgment module;

[0098] UART serial port sending module: used to package the predicted value of exhaust temperature obtained by the performance prediction module according to the UART protocol and send it to the host computer through the serial port to display the predicted value of exhaust temperature and safety alarm;

[0099] Threshold judgment module: used to monitor whether the predicted value of exhaust temperature is higher than the safety threshold. If the predicted value of exhaust temperature is higher than the safety threshold, a safety alarm will be issued.

[0100] Before deploying the TCSNN-based aircraft engine gas path component degradation prediction system on the FPGA, the TCSNN needs to be trained. The process is as follows:

[0101] Step 1: Using sensors to collect raw data on the time-varying parameters of various air path components in an aircraft engine maintained in cruise mode; preprocessing the raw data, including normalization, noise reduction, and data enhancement, to obtain the preprocessed air path performance parameters; performing a correlation analysis between each air path performance parameter and the exhaust temperature, and using each air path performance parameter with a correlation function value greater than a set threshold as a feature sequence input into the TCSNN; and dividing the feature sequence into a training set, a validation set, and a test set in proportion;

[0102] Furthermore, the step 1 specifically includes the following steps:

[0103] Step 1.1: Set the sampling frequency to 32768 Hz and use sensors to collect raw data on the low-pressure rotor speed, high-pressure rotor speed, high-pressure compressor outlet pressure, oil pressure, compressor outlet temperature, and lead angle changes over time when the aircraft engine is in cruise mode.

[0104] Step 1.2: Use normalization to scale the raw data to The data sequence between them is processed; the wavelet packet threshold denoising method is used to denoise the normalized data sequence to obtain a denoised data sequence; the sliding overlapping sampling method is used to enhance the denoised data sequence, the amount of data collected per second is used as the window length of the overlapping sampling, and the step size is set to 8192;

[0105] Furthermore, step 1.2 specifically includes the following steps:

[0106] Step 1.2.1: Normalize the original data and scale it to The number of samples is Data series ;

[0107] Data Series Each of the The calculation formula is as follows:

[0108]

[0109] in, is the original data, is the sequence of original data, is the maximum value in the sequence of original data, is the minimum value in the sequence of original data;

[0110] Step 1.2.2, use The wavelet basis is used to perform normalized data sequence Layer wavelet decomposition, we get Layer wavelet coefficients, where the wavelet coefficients obtained by wavelet decomposition of each layer contain approximate coefficients (low-frequency components) and detail coefficients (high-frequency components); The detail coefficients of the layer are screened using a soft threshold function; The approximate coefficients of the layers and the soft threshold function are filtered from the first layer to the The detail coefficients of the layer are inversely transformed by wavelet and reconstructed to obtain the denoised data sequence ;in, for The data after noise reduction;

[0111] Step 1.2.3: Use sliding overlapping sampling method and fixed sample length and overlap length The denoised data sequence is converted into Split into A sample sequence, recorded as:

[0112]

[0113] in, For the denoised data sequence Middle The denoised data of the item;

[0114] Step 1.3: Calculate the mutual information function value between each gas path performance parameter processed in step 1.2 and the exhaust temperature. When the mutual information function value is greater than the set threshold, it is used as the feature sequence input to the TCSNN. When the mutual information function value is less than the set threshold, it is not used as the feature sequence of the TCSNN.

[0115] Mutual information function between each gas path performance parameter and exhaust temperature The expression is:

[0116]

[0117] Where, is the denoised data sequence, Exhaust temperature series , 、 and They are 、 and The entropy of , and:

[0118]

[0119]

[0120]

[0121] Where, express The probability density of express The probability density of for and The joint probability density of

[0122] Table 1 shows the exhaust gas temperature EGT and low pressure rotor speed , high pressure rotor speed , high pressure compressor outlet pressure , oil pressure , compressor outlet temperature , chamfer angle The mutual information function value of :

[0123] Table 1 Mutual information function values

[0124]

[0125] 0.9 is used as the correlation analysis threshold. 、 、 The mutual information function value of is greater than the correlation analysis threshold and is used as an effective parameter for predicting exhaust temperature;

[0126] Step 1.4: Divide the feature sequence obtained in step 1.3 into a training set, a validation set, and a test set in a ratio of 7:1:2;

[0127] Step 2: Build and train the temporal convolutional spike neural network (TCSNN).

[0128] like Figure 2As shown in the figure, TCSNN adds a pulse neuron calculation module to the temporal convolutional network (TCN). The temporal convolutional network (TCN) contains a C-layer residual network structure. Each residual network structure consists of two convolutional layers and one residual layer. The C-layer residual network structure is used to extract features and perform temporal modeling on the input feature sequence. The pulse neuron calculation module includes X pulse neurons. Each pulse neuron is placed between two adjacent residual network structures to play a connecting role. At the same time, based on the firing timing of the pulse neurons, the continuous signal output by the residual network structure is converted into a pulse signal input and transmitted to the pulse neuron, further optimizing the prediction results.

[0129] The training process of TCSNN is as follows: first, the hyperparameters and weights of TCSNN are initialized; then, the training set is input into TCSNN in batches for training until TCSNN converges; and the optimal TCSNN is obtained by adjusting the hyperparameters. The details are as follows:

[0130] Step 2.1: Set the number of training batches to 64, the learning rate to 0.001, the number of iterations to 300, and randomly initialize the neural weights of each layer of the TCSNN network model;

[0131] Step 2.2: Input the training set into TCSNN in batches and output the predicted value of the exhaust temperature of the aircraft engine. Compare the predicted value of the exhaust temperature with the actual value, calculate the loss value Loss through the Mean Squared Error (MSE) function, and calculate the loss function through the chain rule. Weights for convolutional and residual layers The gradient of is used to back-propagate the gradient descent algorithm to update the weights of each convolutional layer and residual layer of TCSNN;

[0132] Furthermore, the specific formula of the loss value Loss is:

[0133]

[0134] in, is the number of samples, is the actual value, is the predicted value of TCSNN;

[0135] Furthermore, the loss function Weight The gradient formula is:

[0136]

[0137] Step 2.3: Determine the loss value Is it less than the threshold? , when the loss value Less than threshold When , TCSNN is considered to be optimal and training is stopped. Otherwise, the hyperparameters are adjusted and step 2.2 is repeated.

[0138] Step 3: Evaluate the storage and computing resource requirements of TCSNN on FPGA; load the weight parameters obtained from training TCSNN into FPGA in COE file format for storage;

[0139] Step 3.1: Select Xilinx UltraScale+ series FPGA and configure 16-bit fixed-point numbers (Q2.13 format). The temporal convolution network calculation module adopts a 4-way parallel pipeline design, with 4 DSP48E2 units allocated to each way, to achieve single-cycle completion of 4×3×3 convolution kernel calculation. The ring RAM storage structure is as follows: Figure 4 As shown, the depth is set to 2048, and dual-port BRAM is used to achieve parallel reading and writing; the weight ROM adopts a segmented storage strategy, mapping the shallow weights with high frequency access to Block RAM, and the deep weights are stored in distributed RAM;

[0140] Step 3.2: Write the weight parameters obtained from TCSNN training into the COE file for reading by the weight ROM module on the FPGA for forward propagation of TCSNN.

[0141] Step 4: The sensor module collects the parameters of each gas path component of the aircraft engine in real time and inputs them into the FPGA. The true value is compared with the TCSNN predicted value to verify the accuracy of the network. A time series comparison diagram of the true value and the TCSNN predicted value is drawn, and a comparison diagram of the true value and the TCSNN predicted value is drawn. Figure 5 As shown, the X-axis is the flight cycle, and the Y-axis is the absolute error between the true value and the TCSNN predicted value; the lower right corner is a comparison chart of the true value and the TCSNN predicted value, with the X-axis being the TCSNN predicted value and the Y-axis being the true value of the parameter. When X=Y, it means that there is no error between the TCSNN inference value and the true value.

Claims

1. A TCSNN-based aircraft engine gas path component degradation prediction system, characterized by: The TCSNN-based aircraft engine gas path component degradation prediction system includes a data input module, a performance prediction module, and a data output module deployed on the FPGA; The data input module includes a sensor module, a UART serial port receiving module, a data processing module, and a line buffer RAM module; the composition and specific functions of each module are as follows: Sensor module: includes various types of sensors, used to collect raw data of the parameters of each gas path component changing over time; UART serial port receiving module: used to receive the raw data of the parameters of each gas path component changing over time collected by the sensor module, and generate characteristic data after parsing according to the UART protocol; Data processing module: used to process the characteristic data generated by the UART serial port receiving module according to the UART protocol analysis to obtain a characteristic sequence; The data processing module is deployed on the PS side of the FPGA, and the processed feature sequence is transmitted from the PS side to the PL side via the AXI bus; Row cache RAM module: used to cache the feature sequence processed by the data processing module; The performance prediction module includes a weight loading module, a weight ROM module, a time convolution network calculation module, a data flow control module, a pulse neuron calculation module, and a fully connected calculation module. The composition and specific functions of each module are as follows: Weight loading module: The temporal convolutional network calculation module sends the weight address index to the weight loading module. The weight address index includes the convolution cycle count signal within a single layer and the overall convolution layer cycle count signal. The weight loading module generates the address of the weight parameter to be read according to the weight address index; Weight ROM module: used to write the weight parameters obtained by TCSNN training into the BRAM of FPGA in the form of COE file, and output the required weight parameters to the temporal convolutional network calculation module according to the address of the weight parameters generated by the weight loading module; Temporal convolutional network computing module: including input data cache RAM unit, convolution calculation result cache RAM unit and convolution and residual general computing unit; Input data cache RAM unit: used to cache the characteristic sequence from the row cache RAM module in the data input module; Convolution calculation result cache RAM unit: used to cache the intermediate result data after the convolution and residual general calculation units perform convolution calculations and residual calculations; Convolution and residual general computing unit: The convolution calculation parallelism is designed according to the shape of TCSNN and the DSP and LUT resources of FPGA. The convolution and residual general computing unit reads the feature sequence in the input data cache RAM unit and the intermediate result data in the convolution calculation result cache RAM unit according to the designed convolution calculation parallelism, and reads the weight parameters in the weight ROM module that matches the feature sequence to perform convolution calculation and residual calculation; During the convolution calculation and residual calculation process, the convolution calculation process of each layer needs to continuously update the convolution cycle count signal in the single layer to refresh the weight parameters; After the convolution calculation of each layer is completed, the overall cycle count signal of the convolution layer is updated, and The intermediate result data of the convolution calculation of the current layer is cached in the convolution calculation result cache RAM unit. When the convolution calculation of the next layer starts, the intermediate result data of the convolution calculation of the previous layer is directly read from the convolution calculation result cache RAM unit to perform the convolution calculation of the next layer. When the convolution calculation of all layers is completed, the convolution and residual general calculation unit performs the residual calculation: read the feature sequence from the input data cache RAM unit, and perform convolution calculation on the feature sequence and the refreshed weight parameter. The result of the convolution calculation is weightedly added with the convolution calculation result of the last layer that has been cached in the convolution calculation result cache RAM unit. At this point, all calculations of the entire time convolution network calculation module are completed. Pulse neuron calculation module: The pulse neuron calculation module includes Z pulse neurons, according to the time step parameters designed for each pulse neuron Build Each pipeline structure is constructed by a set of LIF model operators. The membrane potential of the initial cache in the first pipeline structure is set to zero, and the membrane potential is transmitted to the next pipeline structure or an output pulse is generated according to the result signal output by the time convolution network calculation module. The pulse neuron calculation module converts the result signal output by the time convolution network calculation module into a pulse signal based on the firing sequence of the pulse neurons; the LIF model operator is explained by the following formula: （1） （2） Where: represents the time constant of the membrane, represents the membrane potential of the spiking neuron, represents the resting potential of the spiking neuron, represents the membrane resistance, Indicates the external input current over time The change of; Formula (1) describes the influence of the membrane potential on the leakage effect of the resting potential and the external input current; represents the time when the spiking neuron fires a pulse, Indicates the time it takes for a spike neuron to approach 0 after it emits a spike The membrane potential of the endoplasmic reticulum neuron, represents the reset value of the membrane potential after the spiking neuron emits a pulse; Formula (2) describes the After the pulse is released at the moment, it will approach 0 Resets its membrane potential to ; Data flow control module: The pulse signal output by the pulse neuron calculation module needs to be repeatedly input C times into the time convolution network calculation module for convolution calculation and residual calculation, and the pulse neuron calculation module is converted into a pulse signal. After the last time convolution network calculation module is completed, it is processed by the pulse neuron calculation module and converted into a pulse signal, and then handed over to the fully connected calculation module for calculation; The data flow control module controls the output flow between each module in the performance prediction module according to the convolution cycle count signal within a single layer and the overall convolution layer cycle count signal; Fully connected computing module: The fully connected computing module is responsible for mapping the pulse signal obtained by the pulse neuron computing module to the final output space. It maps the pulse signal to the predicted value of the exhaust temperature through a two-layer fully connected network. The data output module includes a UART serial port sending module and a threshold judgment module; UART serial port sending module: used to package the predicted value of exhaust temperature obtained by the performance prediction module according to the UART protocol and send it to the host computer through the serial port to display the predicted value of exhaust temperature and safety alarm; Threshold judgment module: used to monitor whether the predicted value of exhaust temperature is higher than the safety threshold. If the predicted value of exhaust temperature is higher than the safety threshold, a safety alarm will be issued.

2. The TCSNN-based aircraft engine gas path component degradation prediction system according to claim 1, characterized in that: Before deploying the TCSNN-based aircraft engine gas path component degradation prediction system on the FPGA, the TCSNN needs to be trained. The process is as follows: Step 1: Using sensors to collect raw data on the time-varying parameters of various air path components in an aircraft engine maintained in cruise mode; preprocessing the raw data, including normalization, noise reduction, and data enhancement, to obtain the preprocessed air path performance parameters; performing a correlation analysis on each air path performance parameter with the exhaust temperature, and using each air path performance parameter with a correlation function value greater than a set threshold as a feature sequence input into the TCSNN; Divide the feature sequence into training set, validation set and test set in proportion; The step 1 specifically includes the following steps: Step 1.1: Use sensors to collect raw data on the time-varying low-pressure rotor speed, high-pressure rotor speed, high-pressure compressor outlet pressure, oil pressure, compressor outlet temperature, and lead angle of the aircraft engine at a fixed sampling frequency while the aircraft engine is in cruise mode. Step 1.2: Use normalization to scale the raw data to The data sequence between them is processed; the wavelet packet threshold denoising method is used to denoise the normalized data sequence to obtain a denoised data sequence; the sliding overlapping sampling method is used to enhance the denoised data sequence; Step 1.2 specifically includes the following steps: Step 1.2.1: Normalize the original data and scale it to The number of samples is Data series ; Data Series Each of The calculation formula is as follows: in, is the original data, is the sequence of original data, is the maximum value in the sequence of original data, is the minimum value in the sequence of original data; Step 1.2.2, use The wavelet basis is used to perform normalized data sequence Layer wavelet decomposition, we get Layer wavelet coefficients, where the wavelet coefficients obtained by wavelet decomposition of each layer include approximate coefficients and detail coefficients; The detail coefficients of the layer are screened using a soft threshold function; The approximate coefficients of the layers and the soft threshold function are filtered from the first layer to the The detail coefficients of the layer are inversely transformed by wavelet and reconstructed to obtain the denoised data sequence ;in, for The data after noise reduction; Step 1.2.3: Use sliding overlapping sampling method and fixed sample length and overlap length The denoised data sequence is converted into Split into A sample sequence, recorded as: in, For the denoised data sequence Middle The denoised data of the item; Step 1.3: Calculate the mutual information function value between each gas path performance parameter processed in step 1.2 and the exhaust temperature. When the mutual information function value is greater than the set threshold, it is used as the feature sequence input to the TCSNN. When the mutual information function value is less than the set threshold, it is not used as the feature sequence of the TCSNN. Mutual information function between each gas path performance parameter and exhaust temperature The expression is: Where, is the denoised data sequence, Exhaust temperature series , 、 and They are 、 and The entropy of , and: Where, express The probability density of express The probability density of for and The joint probability density of Step 1.4: Divide the feature sequence obtained in step 1.3 into training set, validation set and test set in proportion; Step 2: Build and train the temporal convolutional spike neural network (TCSNN). TCSNN adds a spike neuron computing module to the temporal convolutional network (TCN). The TCN contains a C-layer residual network structure, each of which consists of two convolutional layers and one residual layer. The C-layer residual network structure is used to extract features and perform temporal modeling on the input feature sequence. The spike neuron computing module includes X spike neurons, each of which is placed between two adjacent residual network layers to act as a connection. At the same time, based on the firing timing of the spike neurons, the continuous signal output by the residual network structure is converted into a spike signal input and transmitted to the spike neurons, further optimizing the prediction results. The training process of TCSNN is as follows: first, the hyperparameters and weights of TCSNN are initialized; then, the training set is input into TCSNN in batches for training until TCSNN converges; and the optimal TCSNN is obtained by adjusting the hyperparameters. The details are as follows: Step 2.1: Initialize the hyperparameters and weights of TCSNN; Step 2.2: Input the training set into TCSNN in batches and output the predicted value of the exhaust temperature of the aircraft engine. Compare the predicted value of the exhaust temperature with the actual value, calculate the loss value Loss through the mean square error function, and calculate the loss function through the chain rule. Weights for convolutional and residual layers The gradient of is used to back-propagate the gradient descent algorithm to update the weights of each convolutional layer and residual layer of TCSNN; The specific formula for loss value Loss is: in, is the number of samples, is the actual value, is the predicted value of TCSNN; Loss Function Weight The gradient formula is: Step 2.3: Determine the loss value Is it less than the threshold? , when the loss value Less than threshold When , TCSNN is considered to be optimal and training is stopped. Otherwise, the hyperparameters are adjusted and step 2.2 is repeated. Step 3: Evaluate the storage and computing resource requirements of TCSNN on FPGA; load the weight parameters obtained from training TCSNN into FPGA in COE file format for storage; Step 3.1: Evaluate the memory and logic resource usage of the TCSNN and its deployment based on the data format of the weight parameters used in the FPGA deployment. On-chip DSP resources are used to calculate tensor multiplications in convolution and matrix calculations, and LUTs are used to build the basic units of the TCSNN. Step 3.2: Write the weight parameters obtained from TCSNN training into the COE file for reading by the weight ROM module on the FPGA for forward propagation of TCSNN.

Images