Wire harness performance monitoring system based on multi-source data analysis

Abstract

The invention relates to the technical field of vehicle wire harness fault monitoring, in particular to a wire harness performance monitoring system based on multi-source data analysis, and the system comprises a multi-source heterogeneous data sensing module which is used for collecting the electrical parameter analog quantity of a wire harness, environment vibration frequency data and vehicle operation state parameters, and outputting an original multi-source data stream; and the data preprocessing and feature extraction module is used for receiving the original multi-source data stream and carrying out timestamp synchronization and denoising preprocessing on the original multi-source data stream. According to the invention, by establishing a multi-source heterogeneous data fusion mechanism, unified normalization and structured processing of sensor data and vehicle operation data of different types, different dimensions and different time scales are realized, and a hierarchical fusion calculation method of a physical failure model and a deep learning model is combined to construct a health index SOH of a wire harness system. Effective mapping from original monitoring data to health state quantitative representation is realized to a certain extent, and the coordination and rationality of a wiring harness health assessment result are improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle wiring harness fault monitoring technology, and in particular to a wiring harness performance monitoring system based on multi-source data analysis. Background Technology

[0002] With the rapid development of new energy vehicles and intelligent manufacturing equipment, wiring harness systems, as key carriers of electrical energy and signal transmission, have a significant impact on the overall performance of the system due to their operational safety and reliability. Therefore, monitoring and evaluating the health status of wiring harnesses during production and operation is particularly important. To achieve refined management of wiring harness systems, it is necessary to comprehensively utilize heterogeneous data from various sensors and vehicle operating conditions. Through multi-source fusion, status identification, and health index calculation, the electrical, thermal, and mechanical characteristics of the wiring harnesses can be comprehensively analyzed, thereby providing data support and technical foundation for fault early warning, life prediction, and maintenance decisions.

[0003] In existing technologies, the fusion of multi-source heterogeneous data and the assessment of harness health suffer from problems such as inconsistent data types, large differences in dimensions, and mismatched time scales, making it difficult to effectively fuse different sensor and operational data under the same evaluation system. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a wire harness performance monitoring system based on multi-source data analysis, aiming to improve the problems of inconsistent data types, large differences in dimensions, and mismatched time scales in the fusion of multi-source heterogeneous data and wire harness health assessment in the prior art.

[0005] This invention provides the following technical solution: a wire harness performance monitoring system based on multi-source data analysis, comprising:

[0006] The multi-source heterogeneous data sensing module is used to collect analog electrical parameters of the wiring harness, environmental vibration frequency data, and vehicle operating status parameters, and output raw multi-source data streams.

[0007] The data preprocessing and feature extraction module receives the original multi-source data stream, performs timestamp synchronization and noise reduction preprocessing on it, extracts multi-dimensional time-domain and frequency-domain features, and outputs a structured multi-dimensional feature vector.

[0008] The sensor fault diagnosis module receives the structured multidimensional feature vector, performs residual analysis using a time-series prediction model to identify sensor faults, and outputs sensor fault diagnosis results.

[0009] The dynamic data weighting module receives the structured multidimensional feature vector and the sensor fault diagnosis result, calculates the information entropy of each data source in combination with the current vehicle operating status, and generates a dynamic weight matrix based on the diagnosis result, wherein the weight of the sensor identified as faulty is set to a value lower than a first preset threshold.

[0010] The hybrid-driven hierarchical fusion evaluation module receives the structured multidimensional feature vector and dynamic weight matrix, calculates the basic health values ​​of the insulation, connection and mechanical subsystems using the physical failure model, corrects the prediction bias of the physical model using the deep learning model, and outputs the weighted corrected subsystem health index vector.

[0011] The integrated health decision and early warning module receives the weighted and corrected subsystem health index vector, generates an overall health status score and remaining life prediction value of the harness through a decision-level fusion algorithm, and generates a graded early warning instruction when the score is lower than a preset threshold.

[0012] Preferably, the process of acquiring analog electrical parameters of the harness, environmental vibration frequency data, and vehicle operating status parameters, and outputting the raw multi-source data stream specifically includes the following steps:

[0013] Voltage and current sensors deployed at preset monitoring positions on the wire harness are used to continuously monitor analog electrical parameters.

[0014] Accelerometers are used to collect environmental vibration frequency data;

[0015] By monitoring the vehicle's onboard controller local area network bus, the vehicle's operating status parameters can be parsed and obtained.

[0016] The analog electrical parameters, the environmental vibration frequency data, and the vehicle operating status parameters are aggregated to generate the original multi-source data stream.

[0017] Preferably, the steps of performing timestamp synchronization and denoising preprocessing, extracting multi-dimensional time-domain and frequency-domain features, and outputting a structured multi-dimensional feature vector specifically include the following steps:

[0018] For the original multi-source data stream, timestamp alignment is performed on the data from different sources based on a unified master clock;

[0019] Apply a digital filter to the timestamp-aligned data to remove noise;

[0020] Calculate the time-domain statistical characteristics of the filtered data, including mean, variance, kurtosis, and margin factor;

[0021] Fast Fourier Transform is performed on the filtered analog electrical parameters and environmental vibration frequency data to extract frequency domain features;

[0022] The structured multidimensional feature vector is generated by combining the time-domain statistical features and the frequency-domain features.

[0023] Preferably, the step of using a time-series prediction model to perform residual analysis to identify sensor faults and output sensor fault diagnosis results specifically includes the following steps:

[0024] The structured multidimensional feature vector is input into a pre-trained temporal prediction model, which is a long short-term memory network model.

[0025] Using the time-series prediction model, a prediction baseline value is generated for the sensor measurement values ​​in the feature vector;

[0026] Calculate the residual between the actual sensor measurement and the predicted reference value;

[0027] The residual is compared with a preset fault determination threshold;

[0028] When the residual exceeds the fault determination threshold, the corresponding sensor is determined to be faulty.

[0029] Preferably, the dynamic data weighting module calculates the information entropy of each data source in conjunction with the current vehicle operating status, specifically including the following steps:

[0030] The current vehicle operating status is identified based on the structured multidimensional feature vector.

[0031] Based on the current vehicle operating status, the measured values ​​from various data sources in the structured multidimensional feature vector are segmented.

[0032] Statistically analyze the probability distribution of data values ​​within the data segments;

[0033] Based on the probability distribution of the data values, the information entropy of each data source is obtained by summing the product of each probability value and its logarithm.

[0034] Preferably, the dynamic data weighting module generates a dynamic weight matrix based on the sensor fault diagnosis results, specifically including the following steps:

[0035] Iterate through each diagnostic entry in the sensor fault diagnosis results;

[0036] Determine whether the diagnostic item is in a fault state;

[0037] If the diagnostic entry is in a fault state, the weight coefficient of the corresponding data source is set to a value lower than the first preset threshold.

[0038] If the diagnostic entry is a non-fault state, then the corresponding weight coefficient is calculated based on the information entropy of each data source;

[0039] Combine all the weight coefficients to generate the dynamic weight matrix.

[0040] Preferably, the step of calculating the basic health values ​​of the insulation, connection, and mechanical subsystems using a physical failure model, correcting the prediction bias of the physical model using a deep learning model, and outputting a weighted and corrected subsystem health index vector specifically includes the following steps:

[0041] The structured multidimensional feature vector is input into a preset physical failure model, which is based on an electrical stress model, a fatigue cumulative damage model, and a thermal aging model.

[0042] Based on the physical failure model, calculate the basic health values ​​of the insulator system, connection subsystem, and mechanical subsystem respectively;

[0043] The structured multidimensional feature vector and the basic health value are input into a pre-trained deep learning model, which is a multilayer perceptron neural network.

[0044] Using the deep learning model, the prediction bias of the physical failure model is calculated;

[0045] The baseline health value is corrected based on the prediction deviation.

[0046] The modified subsystem health index is weighted and fused with the dynamic weight matrix to generate the weighted modified subsystem health index vector.

[0047] Preferably, the step of generating an overall health status score and remaining life prediction value for the wiring harness through a decision-level fusion algorithm, and generating a graded early warning instruction when the score is lower than a preset threshold, specifically includes the following steps:

[0048] The weighted and corrected subsystem health index vector is input into a preset decision-level fusion algorithm;

[0049] The overall health status score of the harness is generated using the decision-level fusion algorithm.

[0050] Based on the overall health status score of the harness and its historical trend, the remaining lifespan prediction value is predicted and generated.

[0051] The overall health status score of the wiring harness is compared with a preset health score threshold.

[0052] When the overall health status score of the wiring harness is lower than the health score threshold, a warning instruction of the corresponding level is generated based on the numerical range of the score.

[0053] The present invention has the following beneficial effects:

[0054] 1. In this invention, by establishing a multi-source heterogeneous data fusion mechanism, the unified normalization and structured processing of sensor data and vehicle operation data of different types, dimensions, and time scales are realized. By combining the hierarchical fusion calculation method of physical failure model and deep learning model, the health index SOH of the wiring harness system is constructed. To a certain extent, the effective mapping from raw monitoring data to quantitative representation of health status is realized, and the coordination and rationality of wiring harness health assessment results are improved.

[0055] 2. In this invention, by introducing a dynamic data weighting mechanism based on information entropy and a sensor fault diagnosis feedback channel, the weights of different data sources are adaptively allocated. Under the circumstances of changes in the operating environment, sensor noise, or data drift, the influence ratio of various types of data can be appropriately adjusted, making the health assessment process more flexible and robust, thereby improving the stability of the harness health status assessment to a certain extent.

[0056] 3. In this invention, by designing a data missing completion and redundancy fusion strategy, combined with time-series interpolation, Kalman filtering and historical pattern correction methods, key features can be reasonably completed and estimated when some sensors fail or data is abnormal. This keeps the output of the harness health index (SOH) relatively continuous and reliable, and to a certain extent reduces the impact of data missing or faults on the health assessment results, thereby enhancing the adaptability and reliability of the system. Attached Figure Description

[0057] Figure 1 This is a system architecture diagram of a wire harness performance monitoring system based on multi-source data analysis proposed in this invention. Detailed Implementation

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] This invention provides a wire harness performance monitoring system based on multi-source data analysis, such as... Figure 1 As shown, it includes:

[0060] The multi-source heterogeneous data sensing module is used to collect analog electrical parameters of the wiring harness, environmental vibration frequency data, and vehicle operating status parameters, and output raw multi-source data streams.

[0061] Furthermore, the process of collecting analog electrical parameters of the wiring harness, environmental vibration frequency data, and vehicle operating status parameters, and outputting the raw multi-source data stream, specifically includes the following steps:

[0062] Voltage and current sensors deployed at preset monitoring positions on the wire harness are used to continuously monitor analog electrical parameters.

[0063] Accelerometers are used to collect environmental vibration frequency data;

[0064] By monitoring the vehicle's onboard controller local area network bus, the vehicle's operating status parameters can be parsed and obtained.

[0065] The analog electrical parameters, environmental vibration frequency data, and vehicle operating status parameters are aggregated to generate a raw multi-source data stream.

[0066] Specifically, the multi-source heterogeneous data sensing module is used to collect analog electrical parameters of the wiring harness, environmental vibration frequency data, and vehicle operating status parameters, and outputs raw multi-source data streams. By deploying multiple types of sensors at key nodes of the vehicle wiring harness, the module monitors the wiring harness operating status in real time from multiple dimensions. Voltage and current sensors are placed near the main power supply circuit and connectors of the wiring harness to continuously monitor analog electrical parameters such as voltage and current. Accelerometers are installed near the wiring harness fixing clips or support points to collect environmental vibration frequency data. At the same time, by monitoring the vehicle's Controller Area Network (CAN) bus, the module parses the vehicle's operating status parameters, including vehicle speed, RPM, gear, steering angle, braking status, and ambient temperature. The multi-source data is aggregated after being synchronized with timestamps to form a structured raw multi-source data stream, providing an input basis for subsequent data preprocessing, feature extraction, and status assessment.

[0067] Furthermore, analog electrical parameters are acquired by voltage and current sensors via an A / D conversion unit at a fixed sampling frequency to obtain voltage signals. and current signal The two are indexed by time to form an electrical parameter data sequence, while the environmental vibration data is generated by the acceleration signal output from the accelerometer. The vibration frequency components were obtained after processing by Fast Fourier Transform (FFT). ,in Indicates frequency, This spectral component is used to reflect the dynamic excitation characteristics of the environment in which the wiring harness is located. The vehicle operating state parameters are obtained by extracting key fields from the data frame through the CAN bus message parsing module, and standardizing and encoding the state parameters at each moment to obtain the operating state vector. ,in For vehicle speed, Engine speed, For steering angle, This is a braking state quantity. The ambient temperature.

[0068] Finally, the module synchronizes and concatenates the three types of data using a unified time base to generate a multi-source data stream matrix:

[0069] ;

[0070] in, For structured multi-source raw datasets, representing time... The comprehensive status information of the lower wire harness is output as a data matrix to the data preprocessing module for subsequent noise suppression, feature extraction, and health status analysis.

[0071] This multi-source heterogeneous data sensing module can achieve comprehensive monitoring of the electrical, mechanical and operating conditions of wire harnesses without increasing the additional communication burden. The collected data has time synchronization and spatial resolution, which can reflect the real operating characteristics of wire harnesses under complex conditions, providing a highly reliable data foundation for wire harness performance evaluation and fault diagnosis.

[0072] The data preprocessing and feature extraction module receives the original multi-source data stream, performs timestamp synchronization and noise reduction preprocessing on it, extracts multi-dimensional time-domain and frequency-domain features, and outputs a structured multi-dimensional feature vector.

[0073] Further, the process involves timestamp synchronization and denoising preprocessing, and the extraction of multi-dimensional time-domain and frequency-domain features to output a structured multi-dimensional feature vector. This includes the following steps:

[0074] For raw multi-source data streams, timestamp alignment is performed on data from different sources based on a unified master clock;

[0075] Apply a digital filter to the timestamp-aligned data to remove noise;

[0076] Calculate the time-domain statistical characteristics of the filtered data, including mean, variance, kurtosis, and margin factor;

[0077] Fast Fourier Transform is performed on the filtered analog electrical parameters and environmental vibration frequency data to extract frequency domain features;

[0078] By combining time-domain statistical features and frequency-domain features, a structured multidimensional feature vector is generated.

[0079] Specifically, the data preprocessing and feature extraction module receives the raw multi-source data stream output by the multi-source heterogeneous data sensing module, performs timestamp synchronization and noise reduction preprocessing on it, extracts multi-dimensional time-domain and frequency-domain features, and outputs structured multi-dimensional feature vectors. Through unified time reference and signal feature analysis, it realizes the alignment of multi-source heterogeneous signals and structured feature extraction, providing high-quality input for subsequent harness health assessment and fault diagnosis.

[0080] To address the asynchronous issue of different sampling frequencies and times in the original multi-source data stream, we first perform timestamp alignment on the data from different sources based on a unified master clock. Let the original signal sequence be... ,in Indicates the first The output signal of sensors such as voltage, current, or acceleration sensors. For the sampling time sequence of this type of signal, the signals from each sensor are mapped to a unified time axis using a linear interpolation method. The synchronization signal sequence is obtained. The interpolation calculation formula is:

[0081] ;

[0082] in, and For adjacent sampling times, For a moment The sampled values, This is the interpolated synchronization signal value on a unified time axis. This operation ensures that the electrical parameter signal and the vibration signal are consistent in the time dimension.

[0083] The time-synchronized signal often contains sensor noise and environmental interference. Therefore, a digital filtering method is used for noise reduction. The filtering calculation formula is as follows:

[0084] ;

[0085] in, For input signal, For filtering the output signal, These are the forward coefficients of the filter. For feedback coefficients, and These are the forward and feedback orders of the filter, respectively. By properly designing the filter coefficients, high-frequency noise or low-frequency drift can be filtered out, while retaining the main energy components of the signal.

[0086] The time-domain statistical characteristics of the filtered signal are calculated to describe its volatility and stability. The mean value represents the average energy level of the signal, and its calculation formula is as follows:

[0087] ;

[0088] in, The mean of the signal. For the first The signal amplitude at each sampling point This represents the total number of sampling points.

[0089] Variance is used to describe the degree of fluctuation in a signal, and its calculation formula is:

[0090] ;

[0091] in, For signal variance, This represents the signal mean.

[0092] Kurtosis is used to characterize the degree of abrupt changes in signal peaks, and its calculation formula is:

[0093] ;

[0094] in, The kurtosis coefficient, This represents the signal standard deviation. A larger one... The value indicates that there are impact characteristics or abnormal fluctuations in the signal.

[0095] The margin factor reflects the proportional relationship between the signal peak value and the energy distribution, and is calculated using the following formula:

[0096] ;

[0097] in, As a margin factor, The absolute peak amplitude of the signal is given by the denominator, which is the root mean square value of the signal.

[0098] To extract frequency domain features, a Fast Fourier Transform (FFT) was performed on the filtered analog electrical parameters and the environmental vibration signal to obtain the frequency domain spectral amplitude distribution. The calculation formula is as follows:

[0099] ;

[0100] in, For frequency The complex spectrum at that location, For spectral amplitude, For time-domain sampled signals, The number of sampling points. The unit is imaginary. By analyzing the main frequency energy distribution, spectrum center, and frequency energy ratio, the electromagnetic characteristics and mechanical vibration properties of the wire harness under specific operating conditions can be reflected.

[0101] Finally, the calculated time-domain features and frequency-domain features are combined according to feature dimensions to generate a structured multidimensional feature vector:

[0102] ;

[0103] in, The structured feature vector contains both time-domain statistical features and frequency-domain features. Indicates the first The amplitude of each frequency component, The number of frequency features selected corresponds to the time step of this feature vector. Multidimensional operational status information of the lower wiring harness.

[0104] This effectively eliminates time differences between multi-source signals, improves data synchronization and stability, and extracts key feature indicators reflecting the electrical transmission stability, vibration response characteristics and vehicle operating conditions of the wiring harness through filtering and multi-domain feature extraction methods. The processed structured multidimensional feature vector can provide high-dimensional and high-quality input for subsequent sensor fault diagnosis modules and health assessment models, thereby significantly improving the system's state recognition accuracy and reliability.

[0105] The sensor fault diagnosis module receives the structured multidimensional feature vector, performs residual analysis using a time-series prediction model to identify sensor faults, and outputs sensor fault diagnosis results.

[0106] Furthermore, residual analysis using a time-series prediction model to identify sensor faults and output sensor fault diagnosis results specifically includes the following steps:

[0107] The structured multidimensional feature vectors are input into a pre-trained temporal prediction model, which is a long short-term memory network model.

[0108] Using a time-series prediction model, predictive baseline values ​​are generated for sensor measurements in the feature vector;

[0109] Calculate the residual between the actual sensor measurement and the predicted reference value;

[0110] The residual is compared with a preset fault determination threshold;

[0111] When the residual exceeds the fault determination threshold, the corresponding sensor is determined to be faulty.

[0112] Specifically, the sensor fault diagnosis module receives the structured multidimensional feature vector output by the data preprocessing and feature extraction module, uses a time-series prediction model to perform residual analysis to identify sensor faults, and outputs sensor fault diagnosis results. By establishing a prediction model based on time series learning, the feature sequence of the sensor under normal working conditions is modeled, thereby detecting anomalies through prediction errors during system operation, and realizing fault identification and location of sensors such as voltage, current, vibration, and vehicle operating status.

[0113] Structured multidimensional feature vectors The input is fed into a pre-trained time-series prediction model. This model employs a Long Short-Term Memory (LSTM) network structure, which has the ability to capture time-dependent features and effectively model the dynamic changes of sensor features over time. The input to the LSTM model is the time step. Feature vector at time step The output is the prediction result for the next time step or the current feature value. The core computational process of the model includes the state updates of the input gate, forget gate, and output gate, as shown in the following formula:

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] in, For the input feature vector, This is the output vector of the current hidden layer. For memory cell state vectors, These are the activation values ​​for the input gate, forget gate, and output gate, respectively. For the Sigmoid function, It is the hyperbolic tangent function. These are the weight matrices for each gate. For bias terms, This indicates the product operation of corresponding elements.

[0121] After the sensor data stream enters the model, the LSTM network automatically learns the dynamic changes of the sensor based on the feature sequences within the historical time window, thereby outputting the predicted baseline value at each time step. By comparing the actual measured values ​​of the sensor Compared with the predicted baseline value Calculate the residual value The formula is:

[0122] ;

[0123] in, For the first Each sensor at time The residual signal, This is the actual measurement value from the sensor. The residuals serve as the baseline for model predictions and reflect the deviation between the sensor output and the model's expected behavior. When the sensor is working normally, the residuals remain within a low level of random fluctuations, while the residual amplitude increases significantly when drift, inaccuracy, or abnormal noise occurs.

[0124] To achieve automated fault identification, the calculated residual is compared with a preset fault judgment threshold. Let the threshold be... The fault determination condition is:

[0125] | |> => Sensor Malfunction

[0126] in, The absolute value of the residual. The standard threshold, obtained from statistical analysis of historical normal data, can be obtained by setting a confidence interval, such as... ,in and These are the residual mean and standard deviation, respectively. This is an empirical coefficient (usually taken as 2~3).

[0127] When the residual exceeds the threshold, the system immediately determines that the corresponding sensor is abnormal and outputs the sensor fault diagnosis result, including the fault type, sensor number and fault occurrence time. This diagnosis result can be further transmitted to the harness performance evaluation module to eliminate the interference of abnormal signals on subsequent analysis.

[0128] During the training phase, the LSTM model learns the temporal correlation of sensor signals through normal operating data. During the running phase, it uses prediction bias to identify abnormal states in real time. It can identify hidden faults such as chronic drift and intermittent inaccuracy, and has high robustness and generalization ability.

[0129] The dynamic data weighting module receives the structured multidimensional feature vector and the sensor fault diagnosis result, calculates the information entropy of each data source in combination with the current vehicle operating status, and generates a dynamic weight matrix based on the diagnosis result, wherein the weight of the sensor identified as faulty is set to a value lower than a first preset threshold.

[0130] Furthermore, the dynamic data weighting module calculates the information entropy of each data source based on the current vehicle operating status, specifically including the following steps:

[0131] Identify the current vehicle operating status based on structured multidimensional feature vectors;

[0132] Based on the current vehicle operating status, the measured values ​​from various data sources in the structured multidimensional feature vector are segmented.

[0133] The probability distribution of data values ​​within a statistical data segment;

[0134] Based on the probability distribution of data values, the information entropy of each data source is obtained by summing the product of each probability value and its logarithm.

[0135] Furthermore, the dynamic data weighting module generates a dynamic weight matrix based on the sensor fault diagnosis results, specifically including the following steps:

[0136] Iterate through each diagnostic entry in the sensor fault diagnosis results;

[0137] Determine whether the diagnostic item is in a fault state;

[0138] If the diagnostic entry is in a fault state, the weight coefficient of the corresponding data source will be set to a value lower than the first preset threshold.

[0139] If the diagnostic entry is a non-fault state, the corresponding weight coefficient is calculated based on the information entropy of each data source;

[0140] Combine all weight coefficients to generate a dynamic weight matrix.

[0141] Specifically, the dynamic data weighting module receives the structured multidimensional feature vector output from the data preprocessing and feature extraction module, as well as the sensor fault diagnosis results output from the sensor fault diagnosis module. It calculates the information entropy of each data source in combination with the current vehicle operating status, and generates a dynamic weight matrix based on the fault diagnosis results. By calculating the information entropy, it quantifies the uncertainty of each data source and dynamically adjusts the weight of the data source when a sensor fault occurs, so as to achieve precise control of system performance.

[0142] First, based on the structured multidimensional feature vector Identify the current vehicle operating status. The vehicle operating status can be identified through time-series analysis or clustering algorithms of feature vectors, thereby determining the segmentation method of the feature vectors. Assume that at time... The vehicle's operating status is ,in This indicates the current status information. It is based on the vehicle's operating status. The data from each data source in the structured multidimensional feature vector is segmented to obtain the segmented dataset. ,in Indicates the first Segmented datasets from multiple data sources, This represents the number of data points after segmentation.

[0143] Next, we will analyze each data segment. Data value distribution probability within This probability distribution represents the frequency of occurrence of a data value. Assume the discretized values ​​within the data segment are... Its corresponding probability distribution is The formula for calculating the probability distribution is:

[0144] ;

[0145] in, Represents data value The number of times it appears in the segmented dataset. This represents the total number of data points within the data segment.

[0146] Based on the probability distribution of data values, the information entropy of each data source is obtained by summing the products of each probability value and its logarithm. The formula for information entropy is:

[0147] ;

[0148] in, Indicates the first Information entropy of a data source The probability of this data value. This represents the logarithmic function to the base 2. Information entropy quantifies the uncertainty of data distribution; the higher the information entropy, the greater the uncertainty of the data source.

[0149] Subsequently, a dynamic weight matrix is ​​generated by combining the sensor fault diagnosis results. The sensor fault diagnosis results contain information on whether each sensor has malfunctioned. Assuming the sensor fault diagnosis results are... ,in For the first Diagnostic status of each sensor Indicates a fault. This indicates normal. Iterate through each item in the diagnostic results; if the sensor... If the system is in a fault state, the weight coefficient of the corresponding data source will be adjusted. Set below the first preset threshold ,Right now:

[0150] ;

[0151] If the sensor If it is in a normal state, then based on its information entropy Calculate the weighting coefficients. Specifically, the higher the information entropy, the lower the reliability of the data source, and the smaller the weight should be. To achieve an inverse relationship between weight and information entropy, normalization is performed using the reciprocal of information entropy. The weighting coefficients can be calculated using the following formula:

[0152] ;

[0153] in, For the first Information entropy of a data source This assigns a normalized weight to the data source. This ensures that data sources with high information entropy have a lower weighting ratio in the weight allocation, thereby dynamically suppressing data sources with low reliability and improving the stability and accuracy of the overall evaluation results.

[0154] Finally, the weighting coefficients of all sensors are... Combining to generate dynamic weight matrix This matrix is ​​used for subsequent data fusion and decision analysis. Weight matrix The system dynamically adjusts based on vehicle operating status and sensor fault status, enabling it to reflect the reliability of each data source in a specific context in real time, thereby optimizing the overall system performance.

[0155] This allows the system to adjust the weights of each data source in real time based on the current vehicle operating status and sensor fault diagnosis results. This enables the system to automatically reduce the impact of faulty sensors when a sensor failure occurs, while increasing the weight of reliable sensor data, thus ensuring the accuracy and robustness of data fusion and decision analysis.

[0156] The hybrid-driven hierarchical fusion evaluation module receives structured multidimensional feature vectors and dynamic weight matrices, calculates the basic health values ​​of insulation, connection and mechanical subsystems using physical failure models, corrects the prediction bias of the physical model using deep learning models, and outputs a weighted and corrected subsystem health index vector.

[0157] Furthermore, the basic health values ​​of the insulation, connection, and mechanical subsystems are calculated using a physical failure model, and the prediction bias of the physical model is corrected using a deep learning model. The output of the weighted and corrected subsystem health index vector specifically includes the following steps:

[0158] The structured multidimensional feature vector is input into the preset physical failure model, which is based on the electrical stress model, fatigue cumulative damage model and thermal aging model.

[0159] Based on the physical failure model, the basic health values ​​of the insulator system, connection subsystem and mechanical subsystem are calculated respectively.

[0160] The structured multidimensional feature vector and basic health values ​​are input into a pre-trained deep learning model, which is a multilayer perceptron neural network.

[0161] Using a deep learning model, the prediction bias of the physical failure model is calculated;

[0162] Based on the prediction deviation, the baseline health value is corrected;

[0163] The modified subsystem health index is weighted and fused with the dynamic weight matrix to generate a weighted modified subsystem health index vector.

[0164] Specifically, the hybrid-driven hierarchical fusion evaluation module receives structured multidimensional feature vectors and dynamic weight matrices from the dynamic data weighting module. Combining the physical failure model and the deep learning model, it calculates the basic health values ​​of the insulation, connection, and mechanical subsystems and corrects the prediction results of the physical failure model, thereby generating a weighted and corrected subsystem health index vector. By fusing the physical model and the deep learning model and combining the dynamic weighting strategy, the accuracy and reliability of the subsystem health evaluation are improved.

[0165] First, the structured multidimensional feature vectors are... The data is input into a preset physical failure model, which includes an electrical stress model, a fatigue cumulative damage model, and a thermal aging model. These models calculate the health value of each subsystem based on its physical characteristics and operating environment. Specifically, the electrical stress model describes the aging process of the insulator system under electrical stress, the fatigue cumulative damage model calculates the fatigue damage of the connection subsystem under long-term load, and the thermal aging model simulates the aging behavior of the mechanical subsystem under temperature fluctuations. The health value of each subsystem is... Calculated using the following model:

[0166] Insulator system health value :

[0167] ;

[0168] in, This is a function for calculating health values ​​based on an electrical stress model. These are structured feature vectors.

[0169] Connection subsystem health value :

[0170] ;

[0171] in, This is a function for calculating health values ​​based on a fatigue cumulative damage model. These are structured feature vectors.

[0172] Mechanical subsystem health value :

[0173] ;

[0174] in, This is a function for calculating health values ​​based on a thermal aging model. These are structured feature vectors.

[0175] These basic health values This represents the initial subsystem health state calculated under the physical failure model, but these predictions may contain some errors. Therefore, the following section will structure the multidimensional feature vectors. With basic health value The input is fed into a pre-trained deep learning model, which employs a multi-layer perceptron (MLP) neural network. The MLP structure can correct prediction biases in the physical model through nonlinear mapping, as shown in the formula:

[0176] ;

[0177] in, The subsystem health value is corrected through deep learning. For structured feature vectors, The baseline health value is calculated using the physical failure model. MLP is a multilayer perceptron neural network function that, after training, can correct the baseline health value and output the corrected health index.

[0178] After correction of the deep learning model, the resulting corrected health index reflects the difference between the actual subsystem state and the physical model prediction. Subsequently, the corrected health indices of each subsystem are input into the hybrid-driven hierarchical fusion evaluation module, which calculates the overall health index through weighted fusion at the subsystem level. The weight coefficients here correspond to each subsystem itself and are used to characterize the degree of influence of different subsystems on the overall system health status. The weights can be dynamically adjusted based on the reliability, importance, and health volatility of the subsystems. The calculation formula is as follows:

[0179] ;

[0180] in, For the number of subsystems, For the first The health index is corrected by each subsystem. Let its corresponding weight be given, and satisfy the following conditions: .

[0181] The comprehensive health index calculated in this way It can more accurately reflect the overall operating status of the system and achieve adaptive characterization of the health contribution of each subsystem. The final output result represents the reliability of the subsystem under the current condition.

[0182] By combining physical failure models with deep learning models, the limitations of single models are overcome, and the reliability of health indices is improved through dynamic weighted fusion. This approach can better adapt to health assessments in complex dynamic environments and significantly improve the accuracy of subsystem health predictions.

[0183] The integrated health decision and early warning module receives the weighted and corrected subsystem health index vector, generates an overall health status score and remaining life prediction value for the harness through a decision-level fusion algorithm, and generates a graded early warning instruction when the score is lower than a preset threshold.

[0184] Furthermore, the overall health status score and remaining life prediction value of the harness are generated through a decision-level fusion algorithm, and a graded early warning instruction is generated when the score is lower than a preset threshold. This specifically includes the following steps:

[0185] The weighted and corrected subsystem health index vector is input into the preset decision-level fusion algorithm;

[0186] A decision-level fusion algorithm is used to generate an overall health status score for the wiring harness;

[0187] Based on the overall health status score of the wiring harness and its historical trend, predict and generate the remaining lifespan forecast.

[0188] The overall health status score of the wiring harness is compared with the preset health score threshold.

[0189] When the overall health status score of the wiring harness is lower than the health score threshold, a warning instruction of the corresponding level is generated based on the score range.

[0190] Specifically, the integrated health decision-making and early warning module receives the weighted and corrected subsystem health index vector output from the hybrid drive hierarchical fusion evaluation module. It then uses a decision-level fusion algorithm to generate an overall health status score and remaining life prediction value for the wiring harness, and generates a corresponding level of early warning instruction when the health score falls below a preset threshold. Based on the fusion results of the physical failure model and the deep learning model, this module further realizes system-level health assessment and life prediction, providing early warning basis for the operational safety of the vehicle wiring harness system.

[0191] The weighted and corrected subsystem health index vector is represented as:

[0192] ;

[0193] in, , , These are the corrected health indices for the insulation, connection, and mechanical subsystems, respectively. This health index vector is input to a pre-defined decision-level fusion algorithm. The decision-level fusion algorithm comprehensively considers the coupling relationships between subsystems, historical degradation trends, and current operating environment characteristics, outputting an overall harness health status score in a nonlinear mapping form. Decision-level fusion can employ a combination of fuzzy logic reasoning and weighted nonlinear regression, with the following calculation formula:

[0194] ;

[0195] in, The overall health status of the wiring harness is scored, with a value range of [value missing]. , This is the parameter set for the fusion algorithm. This represents the fusion function of the decision layer. The parameters of this function are adaptively optimized during the training phase using a large amount of experimental data. This ensures that the output health score accurately reflects the overall performance degradation level of the wiring harness.

[0196] After the overall health status score of the wire harness is calculated, its future trend is predicted using the time series characteristics of the health score, thereby obtaining the predicted remaining lifespan of the wire harness. It is assumed that over time... Health score Then its trend of change can be described by a regression prediction model:

[0197] ;

[0198] in, Rate the prediction of future moments. The coefficients are obtained by fitting historical health data, and the remaining lifespan can be further estimated based on the rate of decline in health score. (RemainingUsefulLife), its calculation formula is:

[0199] ;

[0200] in, The failure threshold for health scores. The time derivative of the health score represents the rate of health degradation. This metric can be used to predict the time it takes for a harness to progress from its current state to failure.

[0201] Subsequently, the overall health status of the wiring harness was scored. Compared with the preset health score threshold When comparing, When the value is higher than the threshold, the system determines that the harness is operating normally. When the score falls below a threshold, an early warning mechanism is triggered. Based on different score ranges, the system generates tiered early warning instructions, which can be specifically defined as:

[0202] ;

[0203] in, and These are the level 1 and level 2 warning thresholds, determined based on historical data using statistical methods. The warning instructions include information such as harness number, faulty subsystem, health score, and remaining life prediction, and can be sent to the vehicle monitoring system in real time for advance maintenance decisions.

[0204] This enables the transformation of health assessment from the subsystem level to the system level. The decision-level fusion algorithm allows the health status of the wiring harness system to be uniformly measured under multi-source information, providing a time-based degradation trend analysis for remaining life prediction. The hierarchical early warning mechanism enables early response to potential failures and can accurately identify risk signs before the vehicle wiring harness gradually degrades or partially fails.

[0205] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wire harness performance monitoring system based on multi-source data analysis, characterized in that, include: The multi-source heterogeneous data sensing module is used to collect analog electrical parameters of the wiring harness, environmental vibration frequency data, and vehicle operating status parameters, and output raw multi-source data streams. The data preprocessing and feature extraction module receives the original multi-source data stream, performs timestamp synchronization and noise reduction preprocessing on it, extracts multi-dimensional time-domain and frequency-domain features, and outputs a structured multi-dimensional feature vector. The sensor fault diagnosis module receives the structured multidimensional feature vector, performs residual analysis using a time-series prediction model to identify sensor faults, and outputs sensor fault diagnosis results. The dynamic data weighting module receives the structured multidimensional feature vector and the sensor fault diagnosis result, calculates the information entropy of each data source in combination with the current vehicle operating status, and generates a dynamic weight matrix based on the diagnosis result, wherein the weight of the sensor identified as faulty is set to a value lower than a first preset threshold. The hybrid-driven hierarchical fusion evaluation module receives the structured multidimensional feature vector and dynamic weight matrix, calculates the basic health values ​​of the insulation, connection and mechanical subsystems using the physical failure model, corrects the prediction bias of the physical model using the deep learning model, and outputs the weighted corrected subsystem health index vector. The integrated health decision and early warning module receives the weighted and corrected subsystem health index vector, generates an overall health status score and remaining life prediction value of the harness through a decision-level fusion algorithm, and generates a graded early warning instruction when the score is lower than a preset threshold.

2. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The process of acquiring analog electrical parameters of the harness, environmental vibration frequency data, and vehicle operating status parameters, and outputting the raw multi-source data stream specifically includes the following steps: Voltage and current sensors deployed at preset monitoring positions on the wire harness are used to continuously monitor analog electrical parameters. Accelerometers are used to collect environmental vibration frequency data; By monitoring the vehicle's onboard controller local area network bus, the vehicle's operating status parameters can be parsed and obtained. The analog electrical parameters, the environmental vibration frequency data, and the vehicle operating status parameters are aggregated to generate the original multi-source data stream.

3. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The process of performing timestamp synchronization and denoising preprocessing, extracting multi-dimensional time-domain and frequency-domain features, and outputting a structured multi-dimensional feature vector specifically includes the following steps: For the original multi-source data stream, timestamp alignment is performed on the data from different sources based on a unified master clock; Apply a digital filter to the timestamp-aligned data to remove noise; Calculate the time-domain statistical characteristics of the filtered data, including mean, variance, kurtosis, and margin factor; Fast Fourier Transform is performed on the filtered analog electrical parameters and environmental vibration frequency data to extract frequency domain features; The structured multidimensional feature vector is generated by combining the time-domain statistical features and the frequency-domain features.

4. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The process of using a time-series prediction model to perform residual analysis to identify sensor faults and output sensor fault diagnosis results specifically includes the following steps: The structured multidimensional feature vector is input into a pre-trained temporal prediction model, which is a long short-term memory network model. Using the time-series prediction model, a prediction baseline value is generated for the sensor measurement values ​​in the feature vector; Calculate the residual between the actual sensor measurement and the predicted reference value; The residual is compared with a preset fault determination threshold; When the residual exceeds the fault determination threshold, the corresponding sensor is determined to be faulty.

5. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The dynamic data weighting module calculates the information entropy of each data source in conjunction with the current vehicle operating status, specifically including the following steps: The current vehicle operating status is identified based on the structured multidimensional feature vector. Based on the current vehicle operating status, the measured values ​​from various data sources in the structured multidimensional feature vector are segmented. Statistically analyze the probability distribution of data values ​​within the data segments; Based on the probability distribution of the data values, the information entropy of each data source is obtained by summing the product of each probability value and its logarithm.

6. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The dynamic data weighting module generates a dynamic weight matrix based on the sensor fault diagnosis results, specifically including the following steps: Iterate through each diagnostic entry in the sensor fault diagnosis results; Determine whether the diagnostic item is in a fault state; If the diagnostic entry is in a fault state, the weight coefficient of the corresponding data source is set to a value lower than the first preset threshold. If the diagnostic entry is a non-fault state, then the corresponding weight coefficient is calculated based on the information entropy of each data source; Combine all the weight coefficients to generate the dynamic weight matrix.

7. The wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The process of calculating the basic health values ​​of insulation, connection, and mechanical subsystems using a physical failure model, correcting the prediction bias of the physical model using a deep learning model, and outputting a weighted and corrected subsystem health index vector specifically includes the following steps: The structured multidimensional feature vector is input into a preset physical failure model, which is based on an electrical stress model, a fatigue cumulative damage model, and a thermal aging model. Based on the physical failure model, calculate the basic health values ​​of the insulator system, connection subsystem, and mechanical subsystem respectively; The structured multidimensional feature vector and the basic health value are input into a pre-trained deep learning model, which is a multilayer perceptron neural network. Using the deep learning model, the prediction bias of the physical failure model is calculated; The baseline health value is corrected based on the prediction deviation. The modified subsystem health index is weighted and fused with the dynamic weight matrix to generate the weighted modified subsystem health index vector.

8. A wire harness performance monitoring system based on multi-source data analysis according to claim 1, characterized in that, The process of generating an overall health status score and remaining life prediction value for the wiring harness through a decision-level fusion algorithm, and generating a graded early warning instruction when the score is lower than a preset threshold, specifically includes the following steps: The weighted and corrected subsystem health index vector is input into a preset decision-level fusion algorithm; The overall health status score of the harness is generated using the decision-level fusion algorithm. Based on the overall health status score of the harness and its historical trend, the remaining lifespan prediction value is predicted and generated. The overall health status score of the wiring harness is compared with a preset health score threshold. When the overall health status score of the wiring harness is lower than the health score threshold, a warning instruction of the corresponding level is generated based on the numerical range of the score.

Images