Method and system for detecting stability of camshaft sensor

Abstract

The invention discloses a camshaft sensor stability detection method and system, and relates to the technical field of intelligent sensors, and the method comprises the steps: carrying out the multi-source collection of a camshaft sensor, obtaining a multi-dimensional sensing data set, carrying out the conjoint analysis of the multi-dimensional sensing data set, and extracting a dynamic stability characteristic parameter set; constructing a sensor operation state matrix based on the dynamic stability characteristic parameter set in combination with an environment parameter set, and generating a sensor stability diagnosis result; and matching an optimization strategy library according to the sensor stability diagnosis result, and generating a stability detection report. The technical problems that in the complex operation environment of an engine, a traditional camshaft sensor stability detection means has defects and cannot meet the requirements of accurate evaluation and effective management and control are solved, the stability of the camshaft sensor is accurately detected in the complex operation environment of the engine, accurate data are comprehensively obtained, and the reliability of the camshaft sensor is improved. And the technical effects of accurate evaluation of the stability and effective control of the engine performance are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent sensors, and in particular to a camshaft sensor stability detection method and system. Background Art

[0002] Camshaft sensors play a critical role in engine operation, and their stability directly impacts engine performance. Existing technologies primarily use simple static parameter detection methods to test camshaft sensor stability. While these methods are somewhat effective in assessing the basic performance of these intelligent sensors under stable operating conditions, the advancement of automotive engine technology and the increasing complexity of engine operating conditions have exposed numerous limitations to traditional detection techniques. Summary of the Invention

[0003] This application solves the technical problem that traditional camshaft sensor stability detection methods have defects and cannot meet the needs of accurate evaluation and effective management in complex engine operating environments. This application collects multi-source data from the camshaft sensor to obtain multi-dimensional sensor data such as vibration, temperature, and electromagnetic. After joint analysis, it extracts dynamic stability characteristic parameters, combines environmental parameters to construct an operating state matrix to identify abnormal fluctuation patterns, and then matches the optimization strategy library to accurately evaluate the stability of the camshaft sensor, making the camshaft sensor stability detection results more accurate and reliable, meeting the needs of accurate evaluation and effective management of engine performance.

[0004] In response to the above technical problems, the present application proposes a technical solution for a camshaft sensor stability detection method and system.

[0005] In the first aspect, the present application provides a camshaft sensor stability detection method, wherein the method includes: performing multi-source acquisition on the camshaft sensor to obtain a multidimensional sensor data set, performing joint analysis on the multidimensional sensor data set, and extracting a dynamic stability characteristic parameter set; constructing a sensor operating state matrix based on the dynamic stability characteristic parameter set in combination with an environmental parameter set, identifying abnormal fluctuation patterns according to the sensor operating state matrix, and generating a sensor stability diagnosis result; matching an optimization strategy library according to the sensor stability diagnosis result to generate a stability detection report.

[0006] In the second aspect, the present application provides a camshaft sensor stability detection system, wherein the system includes: a parameter set extraction module, which is used to perform multi-source acquisition of the camshaft sensor to obtain a multidimensional sensor data set, perform joint analysis on the multidimensional sensor data set, and extract a dynamic stability characteristic parameter set; a diagnosis result generation module, which is used to construct a sensor operation state matrix based on the dynamic stability characteristic parameter set combined with the environmental parameter set, identify abnormal fluctuation patterns according to the sensor operation state matrix, and generate a sensor stability diagnosis result; a detection report generation module, which is used to match the optimization strategy library according to the sensor stability diagnosis result to generate a stability detection report.

[0007] This application proposes one or more technical solutions, which have at least the following technical effects: This application acquires multi-dimensional sensor data such as vibration, temperature, and electromagnetic through multi-source acquisition of the camshaft sensor, and clarifies the data sources required for detection. Then, these multi-dimensional sensor data are jointly analyzed, and wavelet packet decomposition, regional grid segmentation and other technologies are used to extract the vibration stability feature vector, temperature stability feature matrix and electromagnetic interference feature set, and then multi-modal fusion is performed to generate a dynamic stability feature parameter set. Then, the sensor operation state matrix is ​​constructed in combination with the environmental parameters, and the spatiotemporal features are extracted through the deep residual network, and the abnormal fluctuation pattern is identified to generate the diagnosis result. Finally, the optimization strategy library is matched according to the diagnosis results, and a stability detection report is generated to achieve a comprehensive detection of the camshaft sensor stability, achieving the technical effect of accurately detecting the stability of the camshaft sensor in the complex operating environment of the engine, comprehensively obtaining accurate data, and meeting the requirements of accurate evaluation of its stability and effective control of engine performance.

[0008] The above content summarizes the present application's solution to a camshaft sensor stability detection method and system. The present application will describe the steps of the technical solution in detail in the following specific implementation methods to facilitate a clear and complete understanding of the present application by technical personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 This is a flow chart of a camshaft sensor stability detection method provided in an embodiment of the present application.

[0011] Figure 2 It is a structural schematic diagram of a camshaft sensor stability detection system provided in an embodiment of the present application.

[0012] Description of the accompanying drawings: parameter set extraction module 1, diagnosis result generation module 2, detection report generation module 3. DETAILED DESCRIPTION

[0013] This application obtains multi-dimensional sensor data sets such as vibration signal sequences, temperature gradient maps, and amplitude-frequency characteristic parameters by performing multi-source acquisition of camshaft sensors. These data are jointly analyzed, and wavelet packet decomposition, regional grid segmentation and other technologies are used to extract vibration stability feature vectors, temperature stability feature matrices, and electromagnetic interference feature sets, and multi-modal fusion is performed to obtain a dynamic stability feature parameter set. The sensor operating state matrix is ​​constructed in combination with environmental parameters, and abnormal fluctuation patterns are identified to generate diagnostic results. The matching optimization strategy library is used to generate a stability detection report containing maintenance recommendations and other content, thereby achieving comprehensive detection and maintenance planning of the camshaft sensor stability, achieving accurate detection of the camshaft sensor stability in a complex engine operating environment, and comprehensively obtaining accurate data, and meeting the technical effects of accurate stability evaluation and effective control of engine performance.

[0014] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0015] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0016] Example 1, as Figure 1 As shown, a camshaft sensor stability detection method, wherein the method includes: Step A100: performing multi-source acquisition on the camshaft sensor to obtain a multi-dimensional sensor data set, performing joint analysis on the multi-dimensional sensor data set, and extracting a dynamic stability characteristic parameter set.

[0017] In this embodiment, the camshaft sensor is a key intelligent sensor installed on the engine, and its stability is crucial to engine control accuracy. Dynamic stability characteristic parameters are a comprehensive set of characteristic parameters generated from a multidimensional sensor dataset through wavelet packet decomposition, regional gridding, harmonic distortion analysis, and multimodal fusion.

[0018] Specifically, a three-axis vibration sensor is first set on its installation base to synchronously trigger and collect vibration acceleration signals to obtain a vibration signal sequence, and an infrared thermal imager is used to collect surface temperature according to a preset sampling period distribution to generate a temperature gradient map. An electromagnetic induction probe is used to synchronously capture the sensor in the working state to determine the electromagnetic signal waveform data and extract the amplitude-frequency characteristic parameters. Then, the three timestamps are aligned to construct a multidimensional sensing data set. The specific steps are described in detail in A110-A140; when jointly analyzing the multidimensional sensing data set to extract the dynamic stability characteristic parameter set, the vibration signal sequence is traversed to perform wavelet packet decomposition and feature extraction to generate a vibration stability characteristic vector, the sensor area is gridded according to the temperature gradient map to construct a temperature stability characteristic matrix, and harmonic distortion analysis is performed based on the electromagnetic signal waveform data to generate an electromagnetic interference feature set. Finally, the three are multimodally fused to generate a dynamic stability characteristic parameter set. The specific steps are described in detail in A150-A180.

[0019] Step A200: constructing a sensor operation state matrix based on the dynamic stability characteristic parameter set in combination with the environmental parameter set, identifying abnormal fluctuation patterns according to the sensor operation state matrix, and generating a sensor stability diagnosis result.

[0020] In the embodiment of the present application, the environmental parameter set is a parameter set used to reflect the external operating conditions of the camshaft sensor. The abnormal fluctuation pattern is a characteristic change pattern that deviates from the normal operating state of the sensor, identified by similarity matching with the normal operating condition record log.

[0021] Optionally, first construct an environmental parameter set and map it to the dynamic stability characteristic parameter set to obtain a parameter mapping relationship, analyze the environmental impact based on this to obtain the environmental sensitivity coefficient, then time-series segment the dynamic stability characteristic parameter set to determine the time window feature subset, and finally weightedly splice it with the environmental parameter set according to the environmental sensitivity coefficient to construct the sensor operation state matrix. The specific steps are described in detail in A210-A240; when identifying abnormal fluctuation patterns according to the sensor operation state matrix to generate sensor stability diagnosis results, first normalize the matrix to generate a normalized state vector sequence, map it to a deep residual network to extract spatiotemporal features to obtain a state feature vector, retrieve the normal operating condition record log and match it with its similarity to identify abnormal fluctuation patterns, generate a confidence score through confidence evaluation, and combine the pattern to locate the fault to obtain abnormal distribution coordinates, which are added to the sensor stability diagnosis results. The specific steps are described in detail in A250-A290.

[0022] Step A300: Matching the optimization strategy library according to the sensor stability diagnosis result to generate a stability detection report.

[0023] In one embodiment of the present application, the diagnostic results are first disassembled to determine multiple diagnostic metadata including sensor health and fault heat map; then, based on these diagnostic metadata, the optimization strategy library is traversed to match and obtain the target maintenance strategy; then, the maintenance history data of the camshaft sensor is retrieved, and prediction is performed based on the data to determine the maintenance prediction cycle; then, the target maintenance strategy is adjusted according to the maintenance prediction cycle to determine the maintenance priority sequence, and a maintenance risk analysis is performed in combination with the sensor health to obtain the maintenance risk warning level; finally, a stability detection report of the camshaft sensor is constructed based on the fault heat map and the maintenance risk warning level. The specific steps are described in detail in A310-A350.

[0024] Furthermore, step A100 in the method provided in the embodiment of the present application includes: A110: A three-axis vibration sensor is set on the mounting base of the camshaft sensor, and the vibration acceleration signal is collected by synchronously triggering the three-axis vibration sensor to obtain a vibration signal sequence.

[0025] A120: The surface temperature distribution of the camshaft sensor is collected by an infrared thermal imager according to a preset sampling period to generate a temperature gradient map.

[0026] A130: An electromagnetic induction probe is used to synchronously capture the camshaft sensor in the working state, determine the electromagnetic signal waveform data, and extract the amplitude-frequency characteristic parameters based on the electromagnetic signal waveform data.

[0027] A140: Align the timestamps of the vibration signal sequence, the temperature gradient map, and the amplitude-frequency characteristic parameters to construct the multidimensional sensing data set.

[0028] In this embodiment, the vibration signal sequence is a sequence of vibration acceleration signals in the X, Y, and Z directions, acquired through synchronous triggering of a three-axis vibration sensor mounted on the camshaft sensor mounting base. The amplitude-frequency characteristic parameters are derived by extracting the frequency domain characteristic parameters using algorithms such as the Fast Fourier Transform (FFT) after synchronously capturing the electromagnetic signal waveform data from the operating camshaft sensor using an electromagnetic induction probe.

[0029] Specifically, first, for the collection of vibration signals, a three-axis vibration sensor is set on the mounting base of the camshaft sensor. The three-axis vibration sensor uses its synchronous trigger mechanism (such as hardware clock synchronization, time deviation <1μs) to collect vibration acceleration signals in the X, Y, and Z directions in real time. The vibration signal sequence is obtained at a sampling frequency of 10kHz (covering the main vibration frequency band of the engine from 0-5000Hz). This sequence can accurately reflect the dynamic response of the sensor in a mechanical vibration environment.

[0030] Next, to collect temperature signals, an infrared thermal imager is used to collect the distributed temperature of the camshaft sensor surface at a preset sampling period of 25Hz (to meet the needs of capturing rapid changes in engine temperature). The thermal imager has a resolution of 640×480 pixels and a temperature detection accuracy of ±1°C. The generated temperature gradient map can be gridded into 50×50 temperature units (each unit corresponds to a 0.5mm×0.5mm sensor surface area), enabling temperature anomaly monitoring of chip-level heating areas (such as solder joints and integrated circuit modules).

[0031] To acquire electromagnetic characteristic signals, an electromagnetic induction probe is used with a synchronous trigger mechanism (with an error of less than 50ns with the vibration and temperature acquisition trigger signals) to capture the electromagnetic signal waveform data (frequency range 100kHz-1GHz) from the camshaft sensor in operation. After anti-aliasing filtering, the time domain signal is sampled at equal intervals (sampling rate ≥ 2GHz, satisfying the Nyquist sampling theorem) and segmented into N-point data frames (e.g., N=4096). A Hanning window is applied to each frame to suppress spectral leakage. The time domain signal is converted to the frequency domain using the fast Fourier transform (FFT) algorithm. The complex amplitude (the square root of the sum of the squares of the real and imaginary parts) is calculated at each frequency point to obtain the amplitude-frequency characteristic distribution with a resolution of sampling rate / N. Parameters such as the fundamental wave amplitude (the amplitude corresponding to the main frequency of the signal), harmonic distortion (THD, the ratio of the square root of the sum of the squares of the amplitudes of each harmonic to the fundamental wave amplitude, with a detection accuracy of 0.1%), and signal-to-noise ratio (SNR, the ratio of the fundamental wave amplitude to the average amplitude of the noise frequency band) are extracted from it to form an amplitude-frequency characteristic parameter set that characterizes the electromagnetic compatibility of the sensor, and accurately identify the impact of weak electromagnetic interference such as PWM (pulse width modulation) modulation interference and radio frequency noise on the frequency characteristics of the sensor signal.

[0032] Finally, the vibration signal sequence (time resolution 100μs), temperature gradient map (time stamp accuracy 1ms), and amplitude-frequency characteristic parameters (sampling interval 1ms) are time-stamp aligned through a high-precision clock synchronization system to ensure that the deviation of multi-source data in the time dimension is less than 200ns, thereby constructing a multidimensional sensing dataset containing time-domain vibration characteristics, spatial-domain temperature distribution, and frequency-domain electromagnetic characteristics.

[0033] Through the above steps, a multidimensional sensing data set covering mechanical vibration, temperature distribution, and electromagnetic characteristics was constructed, providing a high-precision, high-dimensional data foundation for the subsequent joint analysis of the dynamic stability of sensors under complex working conditions.

[0034] Furthermore, step A100 in the method provided in the embodiment of the present application includes: A150: traverse the vibration signal sequence to perform wavelet packet decomposition, perform feature extraction based on the decomposition result, and generate a vibration stability feature vector.

[0035] A160: Perform regional grid segmentation on the camshaft sensor according to the temperature gradient map to construct a temperature stability characteristic matrix.

[0036] A170: Perform harmonic distortion analysis based on the electromagnetic signal waveform data to generate an electromagnetic interference feature set.

[0037] A180: Perform multimodal fusion on the vibration stability feature vector, the temperature stability feature matrix, and the electromagnetic interference feature set to generate the dynamic stability feature parameter set.

[0038] In the embodiments of this application, wavelet packet decomposition is a multi-resolution time-frequency analysis method that recursively decomposes the original signal into different frequency bands while preserving the time domain information. Regional grid segmentation is a technique for extracting spatial features from temperature gradient maps. Harmonic distortion analysis is a method for extracting frequency domain features from electromagnetic signal waveform data.

[0039] Optionally, first, for the vibration signal sequence, use the wavelet packet decomposition algorithm to perform time-frequency domain analysis: decompose the vibration signal (10kHz sampling rate) into 8 frequency bands (0-1.25kHz, 1.25-2.5kHz to 7.5-10kHz, each 1.25kHz is a frequency band) with a 3-layer decomposition depth, calculate the energy of the signal in each frequency band (that is, the sum of the squares of the amplitudes of each sampling point of the signal in the frequency band), divide the energy of each frequency band by the total energy of all frequency bands, and obtain the energy proportion of the 8 frequency bands ; Then calculate the energy entropy characteristics based on the information entropy formula Through this process, a one-dimensional vibration stability feature vector containing the energy entropy values ​​of 8 frequency bands is generated. The larger the entropy value, the more dispersed the vibration energy is distributed in different frequency bands, and the worse the dynamic stability of the sensor.

[0040] Next, we process the temperature gradient map. We divide the 640×480 pixel temperature image into 50×50 grid cells (each cell corresponds to an actual size of 0.5mm×0.5mm) based on the sensor's physical structure. We calculate two key metrics for each grid cell: The transient temperature gradient (the absolute value of the temperature difference between adjacent grid cells) traverses the upper, lower, left, and right adjacent grids of the unit (the boundary cells are calculated based on the actual adjacent cells), calculates the absolute value of the temperature difference between the unit and each adjacent unit, and takes the average of all adjacent differences (or directly retains the differences in each direction) as the transient temperature gradient of the unit, reflecting the degree of local thermal stress concentration.

[0041] Cyclic fluctuation variance (standard deviation of unit temperature within 100 sampling periods): collect the temperature data of the unit within 100 sampling periods, calculate the standard deviation of this set of data (that is, take the average of the sum of the squares of the differences between each temperature value and the average value and then take the square root). This is used as the cyclic fluctuation variance to characterize the stability of the unit temperature in the time dimension. A larger variance indicates a more severe temperature fluctuation.

[0042] The above two indicators are arranged according to grid positions to construct a 50×50 temperature stability characteristic matrix with 2500 matrix elements. Compared with traditional single-point temperature mean statistics, it can locate local overheating areas with an accuracy of 0.25mm² and quantify the degree of dynamic fluctuation.

[0043] Then, for the electromagnetic signal waveform data, frequency domain sensitive features are extracted through harmonic distortion analysis: first, the fundamental wave amplitude attenuation rate is calculated (the percentage of the difference between the current fundamental wave amplitude and the fundamental wave amplitude in the sensor calibration state, reflecting the degree of signal strength degradation). Secondly, the proportion of high-frequency harmonics such as the 3rd, 5th, and 7th harmonics is extracted (the ratio of the sum of the amplitudes of each harmonic to the fundamental wave amplitude, reflecting the signal distortion introduced by electromagnetic interference). An electromagnetic interference feature set containing 10 harmonic characteristic parameters is generated. Compared with the traditional method of only detecting the signal-to-noise ratio, this method can more accurately identify specific types of electromagnetic interference such as PWM signal interference and radio frequency noise.

[0044] Finally, multimodal fusion is performed to normalize the vibration stability feature vector to generate a standard vector, analyze and reduce the dimension of the temperature stability feature matrix to determine the temperature feature vector, and dynamically weight the electromagnetic interference feature set in combination with the engine speed data to generate electromagnetic feature parameters; attention analysis is performed based on the above three types of features to determine the contribution weight coefficient, and the dynamic stability feature parameter set is generated by fusion according to the weight coefficient. The specific steps are detailed in A181-A185.

[0045] Through the above steps, high-discrimination, low-redundancy core feature inputs are provided for the subsequent construction of the sensor operation status matrix and abnormal fluctuation pattern recognition, realizing the key conversion from raw multi-source data to usable features for intelligent diagnosis.

[0046] Furthermore, step A180 in the method provided in the embodiment of the present application includes: A181: Standardize the vibration stability characteristic vector to generate a vibration stability characteristic standard vector.

[0047] A182: Analyze and reduce the dimension of the temperature stability characteristic matrix to determine the temperature characteristic vector.

[0048] A183: Introducing engine speed data based on a camshaft sensor, dynamically weighting the electromagnetic interference feature set and the engine speed data to generate electromagnetic feature parameters.

[0049] A184: Perform attention analysis based on the vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameters to determine the contribution weight coefficients of multiple modal characteristics.

[0050] A185: The vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameter are integrated according to the contribution weight coefficient to generate the dynamic stability characteristic parameter set.

[0051] In the embodiment of the present application, analytical dimensionality reduction is a high-dimensional data processing technology for the temperature stability feature matrix. Attention analysis is a multimodal feature weight allocation mechanism based on the adaptability requirements of the intelligent sensor working conditions.

[0052] Specifically, the vibration stability feature vector is first Z-score normalized. Statistical parameters are calculated for each of the eight dimensions of the vector (corresponding to the energy entropy values ​​of the eight frequency bands after wavelet packet decomposition). By traversing 1,000 sets of historical sample data, the mean μ (assuming μ = 0.32 for the third frequency band) and the standard deviation σ (assuming σ = 0.15 for the third frequency band) of each dimension are calculated. Then, a normalization transformation is performed on the original 8-dimensional feature vector X = [x1, x2, …, x8] acquired in real time. For each dimension xᵢ (i = 1, 2, …, 8), the formula xᵢ' = (xᵢ - μᵢ) / σᵢ is applied. For example, when the energy entropy value of the third frequency band x3 = 0.45, the normalized value is x3' = (0.45 - 0.32) / 0.15 ≈ 0.867. This process maps the original eigenvector into a standard vector X'=[x1',x2',…,x8'] with mean 0 and variance 1, so that the energy entropy features of different frequency bands have the same dimension, eliminating the fusion deviation caused by the difference in feature scale, and then obtaining the vibration stability feature standard vector.

[0053] For the temperature stability feature matrix (50×50=2500 elements), principal component analysis (PCA) is used for dimensionality reduction: Step a: Expand the temperature matrix of each sample into a 2500-dimensional feature vector by row, and construct a dataset X containing N samples (N ≥ 1000, meeting the statistical significance requirement).

[0054] Step b: Calculate the covariance matrix C (2500 × 2500 dimensions) of the dataset, which represents the linear correlations between the features of each grid cell. By decomposing the covariance matrix using eigenvalues, we obtain a set of eigenvectors arranged in descending order of eigenvalue. For example, the cumulative variance contribution corresponding to the first three largest eigenvalues, λ1, λ2, and λ3, reaches 90%, i.e., (λ1+λ2+λ3) / Σλᵢ=90%, indicating that these three principal components capture the main variations in the original temperature distribution data.

[0055] Step c: Project each 2500-dimensional original feature vector onto the subspace formed by the three principal components. For any original vector x, its 3D temperature feature vector y is calculated as y = [x・e1, x・e2, x・e3], where e1, e2, and e3 are the unit eigenvectors corresponding to the first three principal components. This projection process compresses the temperature features of each sample from 2500 dimensions to 3 dimensions, reducing the computational complexity from O(N × 2500²) to O(N × 2500 × 3). The reduced temperature feature vector retains the core spatial distribution of temperature gradients and fluctuations on the sensor surface, effectively reducing the interference of redundant features on model training and mitigating fitting risk.

[0056] In the electromagnetic feature processing phase, dynamic weighting and model building are performed to make the electromagnetic interference feature set fit the actual operating environment. First, a speed-electromagnetic interference mapping model is established: Step d: Collect a large amount of engine speed data (sampling frequency 50 Hz, range 800-6000 rpm) and the corresponding electromagnetic interference signature set (10-dimensional parameters). Simultaneously record the corresponding speed data (sampling frequency 50 Hz, ensuring synchronization with the electromagnetic signal acquisition frequency), forming a dataset containing more than 2000 samples. The data is then preprocessed to convert the speed to dimensionless values ​​(e.g., normalize to the range [0, 1]). Z-score normalization is performed on the electromagnetic parameters to eliminate dimensionality effects.

[0057] Step e: Considering that electromagnetic interference (such as THD, SNR) may have a nonlinear relationship with engine speed (the increase in the switching device operation frequency at high speed is likely to cause high-frequency harmonics), a second-order polynomial regression model is selected as the basic framework, assuming that the model form is f(speed)=a +b⋅speed+c, where a, b, and c are the coefficients to be fitted. The objective function (mean square error (MSE)) is optimized using the least squares method to calculate the coefficient combination that minimizes the difference between the predicted electromagnetic parameters and the actual measured values.

[0058] Step f: To verify the model's generalization capability, 10-fold cross-validation was used to assess fitting accuracy, ensuring that the MSE for both the training and test sets was less than 0.8% and the R² coefficient was greater than 0.95, indicating that the model effectively captures the nonlinear mapping relationship between speed and electromagnetic parameters. The resulting mapping model dynamically adjusts the weighting of the electromagnetic interference feature set based on the real-time speed: at low speeds (e.g., <1500 rpm), the fundamental amplitude decay rate has a more significant impact on sensor stability, so it is assigned a weight of 0.4; at high speeds (e.g., >4000 rpm), the weighting of high-frequency harmonics is increased to 0.6. This ultimately generates electromagnetic feature parameters that vary with engine operating conditions, enhancing the multimodal fusion feature's ability to characterize the actual operating environment.

[0059] Next, attention analysis was performed to construct a three-layer neural network (12-dimensional input layer, 8-dimensional hidden layer, 3-dimensional output layer). Using operating parameters such as engine load and coolant temperature as input, the network was trained to learn the contribution weight coefficients of each modal feature: under high-speed and heavy-load conditions, the vibration feature weight reached 0.55, the temperature feature weight 0.25, and the electromagnetic feature weight 0.2. Under idle conditions, the weight distribution was adjusted to 0.3, 0.4, and 0.3. This process was iteratively optimized using 1,000 sets of calibration data, and the model converged when the prediction error was less than 3%.

[0060] Finally, the three types of features are fused according to the attention weight, and the standardized vibration feature vector (8 dimensions), the temperature feature vector (3 dimensions) after dimensionality reduction, and the dynamic electromagnetic feature parameters (1 dimension) are linearly combined according to the weight coefficient (for example, the fusion formula under high-speed conditions is: dynamic parameters = 0.55×vibration + 0.25×temperature + 0.2×electromagnetic), generating a dynamic stability feature parameter set containing 12 feature parameters.

[0061] By standardizing the elimination of feature scale differences, performing PCA dimensionality reduction on high-dimensional data, introducing dynamic weighting of operating condition parameters, and adaptively allocating weights through the attention mechanism, a dynamic stability feature parameter set that is deeply associated with the engine operating state is constructed, providing a more discriminative and robust feature basis for subsequent abnormal fluctuation pattern recognition.

[0062] Furthermore, step A200 in the method provided in the embodiment of the present application includes: A210: Construct an environmental parameter set, and map the dynamic stability characteristic parameter set with the environmental parameter set to obtain a parameter mapping relationship.

[0063] A220: Perform an environmental impact analysis on the dynamic stability characteristic parameter set according to the parameter mapping relationship to obtain an environmental sensitivity coefficient.

[0064] A230: Traverse the dynamic stability characteristic parameter set to perform time series segmentation and determine a time window characteristic subset.

[0065] A240: Perform weighted concatenation on the time window feature subset and the environmental parameter set according to the environmental sensitivity coefficient to construct the sensor operation state matrix.

[0066] In the embodiment of the present application, environmental impact analysis is performed by establishing a quantitative mapping relationship between environmental parameters and sensor features. Time series segmentation is a process of dividing the dynamic stability feature parameter set into time dimensions and enhancing features using a sliding window mechanism.

[0067] Specifically, we first construct an environmental parameter set. A distributed sensor network (including temperature, humidity, and electromagnetic interference intensity sensors) deployed within the engine compartment and near the sensors collects ambient temperature, relative humidity, and electromagnetic interference intensity in real time. The sampling frequency is synchronized with the dynamic stability characteristic parameter set (50Hz). Parameter mapping is then performed, and a random forest regression algorithm is used to establish a correlation model between the environmental parameters and the dynamic stability characteristic parameters: 500 decision trees were generated using bootstrap sampling with replacement from 1,000 sets of historical data (80% of the data was used as the training set, and the remaining 20% ​​as the validation set). At each node split, each tree randomly selected one-third of the features (approximately 14 dimensions) from the environmental parameters (temperature, humidity, and electromagnetic interference intensity) and dynamic stability characteristic parameters (38 dimensions, derived from comprehensive characterization of the multi-physical domain dynamic characteristics of the camshaft sensor and feature engineering optimization). The optimal splitting attribute was selected using the minimum mean squared error (MSE) criterion, and growth was terminated when the number of leaf node samples fell below 10 or the MSE change rate was less than 1%. Feature importance scores were calculated by integrating the regression results of all decision trees (for example, taking the mean of the predicted temperature stability feature vector). For each feature, the average reduction in MSE caused by its split across all trees was calculated and normalized to yield scores for ambient temperature, humidity, and electromagnetic interference intensity. Finally, a parameter mapping table containing 42 mapping relationships was generated (each environmental parameter corresponds to the influence coefficient of 14 dynamic characteristics). After 10-fold cross-validation, the model prediction error was stabilized within 5%, effectively capturing the nonlinear correlation between environmental parameters and sensor characteristics.

[0068] Next, an environmental impact analysis is performed based on the parameter mapping relationship, and the environmental sensitivity coefficient is calculated: for each dynamic stability characteristic parameter, its fluctuation amplitude under different environmental conditions is analyzed, and the gradient of the influence of the environmental parameter on the characteristic parameter is calculated through partial derivatives to generate a 38×3 environmental sensitivity coefficient matrix (38 is the number of dynamic characteristic parameters, and 3 is the number of environmental parameters).

[0069] Then, a sliding window mechanism was used to segment the time series, with a window size of 50 sampling points (corresponding to 1-second data length) and a step size of 10 points. This partitioned the dynamic stability feature parameter set into overlapping time window feature subsets, each containing a 50×38 time series feature matrix. The features within each window were subjected to time-domain statistics (such as mean, standard deviation, and kurtosis) and frequency-domain transformation (using short-time Fourier transform to extract time-frequency features), expanding the original time series features to 150 dimensions and tripling the feature representation capability.

[0070] Finally, weighted splicing is performed to fuse each time window feature subset (150 dimensions) with the current environmental parameter set (3 dimensions) according to the environmental sensitivity coefficient matrix. The environmental parameters are expanded (for example, the temperature value is mapped to the thermal stress influencing factor) and spliced ​​with the time window feature subset by column to form a 153-dimensional vector. The vector is then weighted using the sensitivity coefficient matrix to construct a 50×153 sensor operation status matrix. The matrix elements contain the time series features corrected for environmental impacts and the fusion information of environmental parameters.

[0071] Through real-time collection of environmental parameters, random forest mapping modeling, sliding window time series segmentation and weighted splicing of sensitivity coefficients, more comprehensive and environmentally adaptable input data is provided for the subsequent spatiotemporal feature extraction of the deep residual network, thereby improving the accuracy and robustness of camshaft sensor stability detection under complex working conditions.

[0072] Furthermore, step A200 in the method provided in the embodiment of the present application includes: A250: Perform normalization processing based on the sensor operation state matrix to generate a normalized state vector sequence.

[0073] A260: Map the normalized state vector sequence to a deep residual network for spatiotemporal feature extraction to obtain a state feature vector.

[0074] A270: Retrieve the normal operating condition record log of the camshaft sensor, perform similarity matching with the normal operating condition record log according to the state feature vector, and identify the abnormal fluctuation pattern.

[0075] A280: Perform confidence assessment based on the abnormal fluctuation pattern to generate a confidence score, perform fault location according to the confidence score combined with the abnormal fluctuation pattern, and obtain abnormal distribution coordinates.

[0076] A290: Add the abnormal distribution coordinates to the sensor stability diagnosis results.

[0077] In the embodiment of the present application, the deep residual network is an improved convolutional neural network architecture. Confidence assessment is the process of quantifying the credibility of the identified abnormal fluctuation pattern through an integrated learning model.

[0078] Specifically, first, the sensor operating state matrix is ​​normalized, and the Min-Max normalization method is used to map the 153-dimensional matrix elements to the [-1, 1] interval (for example, the ambient temperature from -40°C to 120°C is mapped to -1 to 1), generating a normalized state vector sequence of equal scale to ensure the numerical stability of the deep residual network input data.

[0079] Then, the normalized state vector sequence (single sample dimension is 50×153) is input into the deep residual network for spatiotemporal feature extraction: the input layer is first compressed from 153 dimensions to 128 dimensions through a fully connected layer to reduce computational complexity; the first residual block contains three temporal convolution layers with a convolution kernel size of 5×1 (sliding along the time axis with a step size of 1), which extracts short-term dependency features of 5 consecutive time steps (such as the high-frequency jitter pattern of the vibration signal). Each layer outputs a 64-dimensional feature map, and the residual connection is used to avoid gradient vanishing; the second residual block uses a bidirectional LSTM layer (128 hidden units) to capture long-term contextual information up to 10 seconds (500 time steps) (such as the cumulative impact of slow temperature drift on sensor signals), and calculates the importance weight of the features of each time step through the attention mechanism layer (for example, the weight of the corresponding time step of the vibration feature under high-speed conditions is automatically increased to 0.7); the output layer generates a 256-dimensional state feature vector through global average pooling.

[0080] Next, a feature library is built based on the normal operating condition record log, and the feature similarity score is retrieved and calculated using the state feature vector as the index. When the score is lower than expected, anomaly detection is triggered. The abnormal fluctuation pattern is determined by identifying feature outlier data points and analyzing their distribution correlation. The specific steps are described in detail in A271-A274.

[0081] For confidence assessment, the ensemble learning model combines a random forest and a support vector machine (SVM). Using six evaluation metrics (such as feature deviation and anomaly duration) as input, it first trains on historical anomaly samples (consisting of 2,000 sets of valid anomalies and 3,000 sets of normal samples). The random forest constructs 500 decision trees to capture nonlinear correlations between metrics, while the SVM constructs an optimal classification hyperplane in the high-dimensional feature space. During training, 5-fold cross-validation is used to optimize model parameters. Within each fold, the decision threshold is adjusted to find the critical value that optimizes model performance (such as F1 score and balanced accuracy). During the fusion phase, the probability values ​​(0-1) output by the two models are weighted averaged (weights determined through grid search to be 0.6 for the random forest and 0.4 for the SVM) to generate a composite confidence score. For example, when an abnormal sample has a random forest score of 0.78 and an SVM score of 0.72, the final score is 0.78 × 0.6 + 0.72 × 0.4 = 0.756. This exceeds the 0.75 threshold and is considered a valid anomaly. Subsequently, using predefined physical mapping relationships, such as the correspondence between row i and column j of the temperature feature matrix and the sensor chip location (i × 0.5 mm, j × 0.5 mm), combined with the timestamp T of the anomaly occurrence, the anomaly distribution coordinates (X, Y, T) are generated with an accuracy of 0.5 mm × 0.5 mm × 100 ms, enabling precise location of sensor anomalies.

[0082] Finally, the anomaly distribution coordinates are added to the diagnosis results, including information such as the anomaly type (vibration / temperature / electromagnetic anomaly), location and time of occurrence.

[0083] By normalizing the data scale, fusing temporal convolution with a deep residual network and bidirectional LSTM to extract spatiotemporal features, and combining similarity matching with confidence assessment to locate anomalies, the conversion from the sensor operating state matrix to accurate anomaly diagnosis results is achieved, providing an efficient and reliable technical solution for the stability detection of camshaft sensors under complex working conditions.

[0084] Furthermore, step A270 in the method provided in the embodiment of the present application includes: A271: Perform feature analysis based on the normal operating condition record log and build a normal operating condition feature library.

[0085] A272: Using the state feature vector as an index, searching the normal operating condition feature library, performing similarity calculation based on the search results, and obtaining a feature similarity score.

[0086] A273: When the feature similarity score is lower than the expected score value, trigger an anomaly detection instruction, execute the anomaly detection instruction to identify the state feature vector, and determine the feature outlier data point.

[0087] A274: Perform distribution correlation analysis based on the characteristic outlier data points to determine the abnormal fluctuation pattern.

[0088] In one embodiment, when constructing a normal operating condition feature library, a normal operating condition record log containing 100,000 groups of samples (covering typical operating conditions such as idling, high speed, and load mutation) is preprocessed, and a 256-dimensional state feature vector is extracted through a deep residual network. The K-means clustering algorithm is used: first, for the 100,000 groups of normal operating condition samples consisting of the 256-dimensional state feature vectors, 8 cluster centers are preset (corresponding to 8 typical operating conditions such as idling and high speed). By calculating the Euclidean distance between the sample and each center, each sample is assigned to the cluster with the closest distance; then, the mean vector of the samples in each cluster is calculated, which is used as the new cluster center and iteratively updated; the process of assigning samples and updating centers is repeated until the cluster center no longer changes significantly or the preset number of iterations is reached. Finally, the samples are divided into 8 categories, and the mean vector of each group is used as the typical feature template of the operating condition. A normal operating condition feature library containing 2,000 feature templates (250 sample means per group) is formed, and the average cosine similarity between templates is greater than 0.95, achieving effective clustering division of the normal state of all sensor operating conditions.

[0089] Then, using the real-time generated 256-dimensional state feature vector as an index, a nearest neighbor search is performed on the feature library (using cosine similarity to calculate distance). The feature similarity score is calculated for each operating condition category (for example, similarity under idle conditions = vector dot product / module length product), and the maximum value of all category scores is taken as the comprehensive similarity score. The expected score threshold is determined through cross-validation. When the real-time score falls below this threshold, the anomaly detection instruction is triggered: The Isolation Forest algorithm (Isolation Forest) is used to detect outliers in each dimension of the state feature vector. First, real-time data points consisting of a 256-dimensional state feature vector are input into an ensemble model consisting of 100 isolation trees. Each tree randomly selects one feature (such as the energy entropy of the fifth dimension of the vibration feature) and a random partition value for that feature (such as the normal range mean ±3 times the standard deviation). The data space is recursively partitioned until a single sample is isolated or the preset maximum tree depth (20 layers) is reached. The average path length of the sample across all trees is calculated and converted into an outlier score ranging from 0 to 1 (higher scores indicate a greater likelihood of outlier).

[0090] During detection, all 256 dimensions of each state feature vector are jointly analyzed, rather than independently judging each dimension. When the outlier scores of a sample in three or more dimensions simultaneously exceed the expected score threshold (assuming 0.75), it is determined to be a valid outlier data point. For example, if the fifth dimension of the vibration feature (score 0.82), the 12th dimension of the temperature gradient (score 0.85), and the electromagnetic interference THD parameter (score 0.78) all exceed the standard and meet the conditions of all three dimensions, an anomaly flag is triggered. If only a single dimension scores > 0.8 (such as the vibration feature alone is abnormal), it is considered a noise fluctuation and ignored. This mechanism reduces the false positive rate of a single dimension through multi-dimensional joint detection, ensuring the reliability of anomaly identification.

[0091] Finally, a distribution correlation analysis is performed based on characteristic outlier data points (original data points with outliers in one or more dimensions of the state feature vector): In the temporal dimension, the frequency of outliers in consecutive time windows is counted (e.g., outliers appearing in three or more adjacent windows are considered a persistent anomaly); in the feature dimension, principal component analysis is used to determine the primary contributing dimensions of the outliers (e.g., a temperature gradient feature accounting for 40% is classified as a thermal stability anomaly); and in the spatial dimension, the sensor's physical structure mapping is combined (e.g., an outlier temperature feature corresponds to the grid cell in row i and column j, located in the 0.5mm x 0.5mm area in the upper left corner of the chip). Finally, the temporal and spatial distribution characteristics of the outliers are matched against a historical anomaly pattern database (a database that stores valid anomaly data and its associated features confirmed by confidence assessment during historical testing, constructed by accumulating valid anomaly data with a score greater than 0.75 over a long period of testing) to identify specific abnormal fluctuation patterns (e.g., persistent thermal stress concentration in a high-temperature zone).

[0092] Through the above steps, a complete detection process from feature similarity evaluation to multi-dimensional abnormal pattern positioning is realized, providing reliable technical support for the accurate stability diagnosis of camshaft sensors.

[0093] Furthermore, step A300 in the method provided in the embodiment of the present application includes: A310: Deconstructing the sensor stability diagnosis result to determine a plurality of diagnostic metadata, where the plurality of diagnostic metadata includes a sensor health and a fault heat map of the camshaft sensor.

[0094] A320: Traverse the optimization strategy library based on the multiple diagnostic metadata to perform matching and obtain a target maintenance strategy.

[0095] A330: retrieves the maintenance history data of the camshaft sensor, performs a prediction based on the maintenance history data, and determines a maintenance prediction cycle.

[0096] A340: Adjust the target maintenance strategy according to the maintenance prediction cycle, determine a maintenance priority sequence, perform maintenance risk analysis based on the maintenance priority sequence and the sensor health of the camshaft sensor, and obtain a maintenance risk warning level.

[0097] A350: Constructing the stability detection report of the camshaft sensor based on the fault thermogram and the maintenance risk warning level.

[0098] Optionally, first, decompose the sensor stability diagnosis results and extract the sensor health metric from the anomaly distribution coordinates (X, Y, T). Gaussian kernel density estimation is used to calculate the frequency of anomalies within a unit area (0.5 mm × 0.5 mm). (For example, if an area experiences five anomalies within 30 minutes, the health metric is deducted by 20%). Combined with the confidence score (valid anomalies weighed 0.6, invalid anomalies weighed 0.2), the health metric is quantified on a scale of 0-100 (normal status ≥ 85). A fault heatmap is also generated: using a 100×100 grid of the sensor chip as the base, colors are filled according to the anomaly distribution density (red represents areas with high anomaly frequency), forming visual diagnostic metadata that includes temporal and spatial distribution characteristics.

[0099] Next, based on the diagnostic metadata, the system traverses the optimization strategy library (which contains maintenance strategies for different anomaly patterns, such as high-frequency vibration anomalies → checking sensor mounting bolt torque, thermal stress concentration → optimizing chip heat dissipation coatings, etc.). A two-way matching algorithm is used: First, the corresponding hardware maintenance strategy is retrieved based on the physical area located by the fault heat map (e.g., the grid in row i and column j). Second, the maintenance level is matched based on the sensor health level (e.g., a score below 60 triggers a deep maintenance strategy). For example, if a persistent thermal stress anomaly is detected in the high-temperature area in the upper left corner of the chip (health level 55, with the red area on the heat map accounting for 15%), the system automatically retrieves the corresponding numbered strategy from the strategy library: heat dissipation and inspection of the heat sink module + replacement of the thermal grease.

[0100] Next, the system retrieved 5,000 records of historical maintenance data (covering maintenance time, fault types, and repair outcomes over the past three years) and used the Cox proportional hazards model from survival analysis to develop a maintenance cycle prediction model. Input parameters included the current sensor health, historical maintenance intervals, and fault type recurrence rate, and the model output a maintenance cycle prediction. Based on the target maintenance strategy (e.g., hardware replacement requiring an 8-hour downtime), the system prioritized the predicted cycle: abnormalities with a health score of less than 50 and a predicted cycle of less than 100 hours were set to the emergency level (priority 1), while abnormalities with a health score of 60-80 and a predicted cycle of more than 300 hours were set to the advisory level (priority 3), forming a maintenance sequence with three priority levels.

[0101] When conducting maintenance risk analysis, a three-dimensional risk matrix is ​​constructed: the X-axis represents the impact of the fault (classified as high, medium, or low based on whether the abnormal distribution area involves core sensing components), the Y-axis represents the maintenance difficulty (chip replacement requires a senior technician, software calibration is considered elementary), and the Z-axis represents the health degradation rate (a weekly health degradation of >10 points indicates high risk). Using the Analytic Hierarchy Process (AHP) method, weights are calculated (impact 0.5, maintenance difficulty 0.3, and degradation rate 0.2) to generate a risk warning level (red, yellow, or blue). For example, if a core chip area is abnormal (high impact), maintenance requires a senior technician (medium difficulty), and the health degradation is 15 points per week (high rate), the overall risk level is red, indicating that maintenance is required within 48 hours.

[0102] Finally, the stability test report integrates the fault heat map (visually displaying three red abnormal areas), maintenance recommendations (three strategies sorted by priority, such as emergency: replace the heat dissipation module in the upper left corner), maintenance prediction cycle (re-inspection recommended after 320 hours) and risk warning level (two red risk areas require immediate attention), forming a structured document containing fault location information and maintenance recommendations.

[0103] Through the above steps, a closed loop from fault location to precise maintenance strategy generation is achieved, providing data-driven decision support for the full life cycle management of camshaft sensors, and effectively improving maintenance efficiency and system reliability under complex working conditions.

[0104] Furthermore, step A400 in the method provided in the embodiment of the present application includes: A410: Construct a multi-source policy original data set, perform relationship extraction and semantic annotation on the multi-source policy original data set, and generate a structured policy knowledge unit.

[0105] A420: Conduct impact analysis based on the abnormal fluctuation patterns and construct a three-dimensional strategy classification system.

[0106] A430: Performing real-time diagnosis on the structured strategy knowledge unit according to the three-dimensional strategy classification system to determine a strategy reasoning optimization flowchart.

[0107] A440: Perform simulation verification based on the strategy reasoning optimization flowchart, perform iterative updates according to the verification results, and build the optimization strategy library.

[0108] In the embodiment of the present application, the structured strategy knowledge unit is a standardized and computable knowledge carrier formed by processing multi-source heterogeneous data.

[0109] In one embodiment, first, a multi-source strategy original data set is constructed: the maintenance work order database of the equipment management system is accessed (a core component for storing and managing structured data related to equipment maintenance, mainly carrying various types of record information generated during historical maintenance processes), and structured records containing fault codes, maintenance measures, and repair effects (repair time, recurrence rate) from the past five years are extracted; technical personnel in this field collect the full life cycle operation data of multiple camshaft sensors through the industrial Internet of Things platform (sampling frequency 50Hz), and construct an operating parameter sequence (speed, temperature, vibration acceleration, etc.) aligned with the fault timestamp; natural language processing technology is used to parse maintenance manuals and interview records of experts in this field, and implicit experience knowledge such as thermal stress concentration → applying thermal grease is extracted through named entity recognition (NER) to generate an unstructured text strategy set. The above content is integrated to obtain a multi-source strategy original data set.

[0110] Next, we perform relationship extraction and semantic annotation on the original multi-source policy dataset: we use the BERT model to semantically encode unstructured text, and use the conditional random field (CRF) to identify triplets of fault type, maintenance action, and effect indicator (e.g., electromagnetic interference anomaly → shielding layer grounding treatment → 30% reduction in THD value). Combined with the fault code mapping in the structured work order data, we ultimately generate a dataset of structured policy knowledge units. Each unit contains fields such as fault mode (38-dimensional feature outlier combination), maintenance action (12 categories, such as hardware replacement and software calibration), and effect quantification indicator (health improvement value, abnormality recurrence period).

[0111] Then, a three-dimensional strategy classification system was constructed based on abnormal fluctuation patterns: the X-axis represents the fault type (such as abnormal vibration and thermal stability), the Y-axis represents the maintenance action (divided into three levels: emergency treatment, intermediate adjustment, and long-term optimization), and the Z-axis represents the performance indicators (health recovery rate, maintenance cost, and downtime). A decision tree algorithm was used to calculate the correlation weights of each dimension, forming a dynamically scalable classification framework.

[0112] Finally, the structured knowledge units are mapped to a three-dimensional system for policy reasoning. Using a rule engine combined with case-based reasoning (CBR) technology, when a real-time diagnosis of an abnormal fluctuation pattern (e.g., thermal stress concentration in the upper left corner of a chip) is input, the system first matches the X-axis fault type (abnormal thermal stability), retrieves all associated maintenance actions along the Y-axis (e.g., replacing the heat sink module, optimizing the circuit board layout), and then generates a reasoning flowchart containing three candidate policies based on the Z-axis performance metric (prioritizing policies with a health improvement of >30% and a cost of <500 yuan). The policy execution process is simulated and verified using Petri nets. For example, the execution steps of the heat sink module replacement policy (disassembly → inspection → installation → calibration) are simulated to evaluate its impact on the temperature distribution of the sensor chip. Strategies that pass verification are incorporated into the optimized strategy library, while those that fail are returned to the knowledge unit layer for parameter correction. This forms a closed-loop iterative mechanism: data collection – knowledge structuring – classification reasoning – simulation verification. The policy library is updated annually to ensure adaptability to new abnormal patterns.

[0113] Through multi-source data fusion collection, NLP semantic structured processing, three-dimensional classification system construction and simulation verification iteration, accurate mapping and dynamic optimization of maintenance strategies and abnormal patterns are achieved, providing efficient and scalable strategic support for predictive maintenance of camshaft sensors, and improving the scientific nature and reliability of maintenance decisions under complex working conditions.

[0114] In summary, the camshaft sensor stability detection method provided by the embodiment of the present application has the following technical effects: This application builds a data exchange channel between the equipment management system and the optimization strategy library, uses natural language processing technology to parse unstructured text, and obtains structured strategy knowledge units through operations such as relationship extraction and semantic annotation, and stores them in a classified manner in the strategy library. Through the construction of a three-dimensional strategy classification system and strategy reasoning optimization, combined with simulation verification and iterative update mechanisms, strategy matching and dynamic adjustment are carried out based on historical maintenance data and real-time diagnostic results to ensure the accuracy and timeliness of maintenance strategies. This achieves the technical effect of accurately detecting the stability of the camshaft sensor in the complex operating environment of the engine, comprehensively obtaining accurate data, and meeting the requirements of accurately evaluating its stability and effectively controlling engine performance.

[0115] Example 2, as Figure 2As shown, based on the same inventive concept as the aforementioned embodiment 1, the embodiment of the present application provides a camshaft sensor stability detection system, the system comprising: The parameter set extraction module 1 is used to perform multi-source acquisition on the camshaft sensor to obtain a multi-dimensional sensor data set, perform joint analysis on the multi-dimensional sensor data set, and extract a dynamic stability characteristic parameter set.

[0116] The diagnostic result generating module 2 is used to construct a sensor operation state matrix based on the dynamic stability characteristic parameter set combined with the environmental parameter set, identify abnormal fluctuation patterns according to the sensor operation state matrix, and generate a sensor stability diagnostic result.

[0117] The detection report generating module 3 is used to match the optimization strategy library according to the sensor stability diagnosis result to generate a stability detection report.

[0118] Furthermore, the parameter set extraction module 1 is used to perform the following steps: A three-axis vibration sensor is set on the mounting base of the camshaft sensor, and the vibration acceleration signal is synchronously triggered and collected by the three-axis vibration sensor to obtain a vibration signal sequence; the surface temperature distribution of the camshaft sensor is collected by an infrared thermal imager according to a preset sampling period to generate a temperature gradient map; an electromagnetic induction probe is used to synchronously capture the camshaft sensor in a working state to determine the electromagnetic signal waveform data, and the amplitude-frequency characteristic parameters are extracted based on the electromagnetic signal waveform data; the vibration signal sequence, the temperature gradient map, and the amplitude-frequency characteristic parameters are timestamp-aligned to construct the multidimensional sensing data set.

[0119] Furthermore, the parameter set extraction module 1 is used to perform the following steps: The vibration signal sequence is traversed to perform wavelet packet decomposition, and features are extracted based on the decomposition results to generate a vibration stability feature vector; the camshaft sensor is regionally gridded according to the temperature gradient map to construct a temperature stability feature matrix; harmonic distortion analysis is performed based on the electromagnetic signal waveform data to generate an electromagnetic interference feature set; the vibration stability feature vector, the temperature stability feature matrix, and the electromagnetic interference feature set are multimodally fused to generate the dynamic stability feature parameter set.

[0120] Furthermore, the parameter set extraction module 1 is used to perform the following steps: The vibration stability characteristic vector is standardized to generate a vibration stability characteristic standard vector; the temperature stability characteristic matrix is ​​analyzed and reduced in dimension to determine a temperature characteristic vector; the engine speed data is introduced according to the camshaft sensor, and the electromagnetic interference feature set and the engine speed data are dynamically weighted to generate electromagnetic characteristic parameters; attention analysis is performed based on the vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameters to determine the contribution weight coefficients of multiple modal features; the vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameters are fused according to the contribution weight coefficients to generate the dynamic stability characteristic parameter set.

[0121] Furthermore, the diagnosis result generating module 2 is configured to perform the following steps: Construct an environmental parameter set, map the dynamic stability characteristic parameter set with the environmental parameter set to obtain a parameter mapping relationship; perform an environmental impact analysis on the dynamic stability characteristic parameter set according to the parameter mapping relationship to obtain an environmental sensitivity coefficient; traverse the dynamic stability characteristic parameter set to perform time series segmentation to determine a time window feature subset; weightedly splice the time window feature subset and the environmental parameter set according to the environmental sensitivity coefficient to construct the sensor operation state matrix.

[0122] Furthermore, the diagnosis result generating module 2 is configured to perform the following steps: Normalization processing is performed based on the sensor operating state matrix to generate a normalized state vector sequence; the normalized state vector sequence is mapped to a deep residual network for spatiotemporal feature extraction to obtain a state feature vector; the normal operating condition record log of the camshaft sensor is retrieved, and similarity matching is performed between the state feature vector and the normal operating condition record log to identify the abnormal fluctuation pattern; confidence assessment is performed based on the abnormal fluctuation pattern to generate a confidence score, and fault location is performed according to the confidence score combined with the abnormal fluctuation pattern to obtain abnormal distribution coordinates; the abnormal distribution coordinates are added to the sensor stability diagnosis result.

[0123] Furthermore, the diagnosis result generating module 2 is configured to perform the following steps: Based on the normal operating condition record log, feature analysis is performed to construct a normal operating condition feature library; the normal operating condition feature library is searched using the state feature vector as an index, and similarity calculation is performed based on the search result to obtain a feature similarity score; when the feature similarity score is lower than the expected score value, an anomaly detection instruction is triggered, and the anomaly detection instruction is executed to identify the state feature vector and determine feature outlier data points; based on the feature outlier data points, distribution correlation analysis is performed to determine the abnormal fluctuation pattern.

[0124] Furthermore, the test report generating module 3 is used to perform the following steps: Based on the sensor stability diagnosis result, the system is disassembled to determine multiple diagnostic metadata, which include the sensor health and fault heat map of the camshaft sensor; based on the multiple diagnostic metadata, the system traverses the optimization strategy library for matching to obtain a target maintenance strategy; the maintenance history record data of the camshaft sensor is retrieved, and prediction is performed based on the maintenance history record data to determine a maintenance prediction cycle; the target maintenance strategy is adjusted according to the maintenance prediction cycle to determine a maintenance priority sequence, and a maintenance risk analysis is performed based on the maintenance priority sequence in combination with the sensor health of the camshaft sensor to obtain a maintenance risk warning level; and the stability detection report of the camshaft sensor is constructed based on the fault heat map in combination with the maintenance risk warning level.

[0125] Furthermore, the test report generating module 3 is used to perform the following steps: Construct a multi-source strategy original data set, extract relationships and perform semantic annotation on the multi-source strategy original data set to generate structured strategy knowledge units; conduct impact analysis based on the abnormal fluctuation pattern to construct a three-dimensional strategy classification system; perform real-time diagnosis on the structured strategy knowledge units according to the three-dimensional strategy classification system to determine a strategy reasoning optimization flowchart; perform simulation verification based on the strategy reasoning optimization flowchart, perform iterative updates based on the verification results, and construct the optimization strategy library.

[0126] A camshaft sensor stability detection system provided by an embodiment of the present invention can execute a camshaft sensor stability detection method provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects of the execution method.

[0127] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0128] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and can still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A camshaft sensor stability detection method, characterized in that: The method comprises: Performing multi-source acquisition on camshaft sensors to obtain a multi-dimensional sensor data set, performing joint analysis on the multi-dimensional sensor data set, and extracting a dynamic stability characteristic parameter set; constructing a sensor operation state matrix based on the dynamic stability characteristic parameter set and the environmental parameter set, identifying abnormal fluctuation patterns according to the sensor operation state matrix, and generating a sensor stability diagnosis result; An optimization strategy library is matched according to the sensor stability diagnosis result to generate a stability detection report.

2. A camshaft sensor stability detection method according to claim 1, characterized in that: Multi-source acquisition of camshaft sensors to obtain a multi-dimensional sensor data set includes the following methods: A triaxial vibration sensor is arranged on a mounting base of the camshaft sensor, and a vibration acceleration signal is collected by synchronously triggering the triaxial vibration sensor to obtain a vibration signal sequence; The surface temperature distribution of the camshaft sensor is collected by an infrared thermal imager according to a preset sampling period to generate a temperature gradient map; Using an electromagnetic induction probe to synchronously capture the camshaft sensor in a working state, determining electromagnetic signal waveform data, and extracting amplitude-frequency characteristic parameters based on the electromagnetic signal waveform data; The vibration signal sequence, the temperature gradient map, and the amplitude-frequency characteristic parameters are time stamp-aligned to construct the multidimensional sensing data set.

3. A camshaft sensor stability detection method according to claim 2, characterized in that: Performing a joint analysis on the multidimensional sensor data set to extract a dynamic stability characteristic parameter set, the method comprising: Traversing the vibration signal sequence to perform wavelet packet decomposition, performing feature extraction based on the decomposition result, and generating a vibration stability feature vector; Performing regional grid segmentation on the camshaft sensor according to the temperature gradient map to construct a temperature stability characteristic matrix; Performing harmonic distortion analysis based on the electromagnetic signal waveform data to generate an electromagnetic interference feature set; The vibration stability characteristic vector, the temperature stability characteristic matrix, and the electromagnetic interference characteristic set are multimodally fused to generate the dynamic stability characteristic parameter set.

4. A camshaft sensor stability detection method according to claim 3, characterized in that: Performing multimodal fusion on the vibration stability characteristic vector, the temperature stability characteristic matrix, and the electromagnetic interference characteristic set to generate the dynamic stability characteristic parameter set, the method comprising: performing standardization processing on the vibration stability characteristic vector to generate a vibration stability characteristic standard vector; Analyzing and reducing the dimension of the temperature stability characteristic matrix to determine the temperature characteristic vector; Introducing engine speed data based on a camshaft sensor, dynamically weighting the electromagnetic interference feature set and the engine speed data to generate electromagnetic feature parameters; Performing attention analysis based on the vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameter to determine contribution weight coefficients of multiple modal features; The vibration stability characteristic standard vector, the temperature characteristic vector, and the electromagnetic characteristic parameter are fused according to the contribution weight coefficient to generate the dynamic stability characteristic parameter set.

5. The camshaft sensor stability detection method according to claim 1, wherein: The sensor operation state matrix is ​​constructed based on the dynamic stability characteristic parameter set and the environmental parameter set, and the method includes: Constructing an environmental parameter set, and mapping the dynamic stability characteristic parameter set with the environmental parameter set to obtain a parameter mapping relationship; Performing an environmental impact analysis on the dynamic stability characteristic parameter set according to the parameter mapping relationship to obtain an environmental sensitivity coefficient; Traversing the dynamic stability characteristic parameter set to perform time series segmentation and determine a time window characteristic subset; The time window feature subset and the environmental parameter set are weightedly spliced ​​according to the environmental sensitivity coefficient to construct the sensor operation state matrix.

6. A camshaft sensor stability detection method according to claim 1, characterized in that: Identifying abnormal fluctuation patterns according to the sensor operating state matrix and generating sensor stability diagnosis results, the method includes: Performing normalization processing based on the sensor operating state matrix to generate a normalized state vector sequence; Mapping the normalized state vector sequence to a deep residual network for spatiotemporal feature extraction to obtain a state feature vector; Retrieving a normal operating condition record log of a camshaft sensor, performing similarity matching between the state feature vector and the normal operating condition record log, and identifying the abnormal fluctuation pattern; Performing a confidence assessment based on the abnormal fluctuation pattern to generate a confidence score, performing fault location according to the confidence score combined with the abnormal fluctuation pattern to obtain abnormal distribution coordinates; The abnormal distribution coordinates are added to the sensor stability diagnosis result.

7. A camshaft sensor stability detection method according to claim 6, characterized in that: Retrieving a normal operating condition log of a camshaft sensor, performing similarity matching between the state feature vector and the normal operating condition log, and identifying the abnormal fluctuation pattern, the method includes: Perform feature analysis based on the normal operating condition record log to build a normal operating condition feature library; Using the state feature vector as an index, searching the normal operating condition feature library, performing similarity calculation based on the search results, and obtaining a feature similarity score; When the feature similarity score is lower than the expected score value, triggering an anomaly detection instruction, executing the anomaly detection instruction to identify the state feature vector and determine the feature outlier data point; A distribution association analysis is performed based on the characteristic outlier data points to determine the abnormal fluctuation pattern.

8. The camshaft sensor stability detection method according to claim 1, wherein: According to the sensor stability diagnosis result, the optimization strategy library is matched and a stability detection report is generated, the method comprising: Deconstructing the sensor stability diagnosis result to determine a plurality of diagnostic metadata, the plurality of diagnostic metadata including a sensor health and a fault heat map of the camshaft sensor; Traversing the optimization strategy library based on the multiple diagnostic metadata to perform matching and obtain a target maintenance strategy; Retrieving maintenance history data of a camshaft sensor, performing a prediction based on the maintenance history data, and determining a maintenance prediction cycle; adjusting the target maintenance strategy according to the maintenance prediction cycle to determine a maintenance priority sequence, performing a maintenance risk analysis based on the maintenance priority sequence in combination with the sensor health of the camshaft sensor to obtain a maintenance risk warning level; The stability detection report of the camshaft sensor is constructed based on the fault thermogram and the maintenance risk warning level.

9. The camshaft sensor stability detection method according to claim 1, wherein: The process of constructing the optimization strategy library includes: Constructing a multi-source strategy original data set, performing relationship extraction and semantic annotation on the multi-source strategy original data set, and generating a structured strategy knowledge unit; Conduct impact analysis based on the abnormal fluctuation patterns to construct a three-dimensional strategy classification system; Performing real-time diagnosis on the structured strategy knowledge unit according to the three-dimensional strategy classification system to determine a strategy reasoning optimization flow chart; Simulation verification is performed based on the strategy reasoning optimization flowchart, and iterative updates are performed according to the verification results to construct the optimization strategy library.

10. A camshaft sensor stability detection system, characterized in that: A system for implementing a camshaft sensor stability detection method according to any one of claims 1 to 9, comprising: a parameter set extraction module, configured to perform multi-source acquisition of camshaft sensors to obtain a multi-dimensional sensor data set, perform joint analysis on the multi-dimensional sensor data set, and extract a dynamic stability characteristic parameter set; a diagnosis result generating module, configured to construct a sensor operation state matrix based on the dynamic stability characteristic parameter set in combination with the environmental parameter set, identify abnormal fluctuation patterns according to the sensor operation state matrix, and generate a sensor stability diagnosis result; The detection report generation module is used to match the optimization strategy library according to the sensor stability diagnosis result to generate a stability detection report.

Images