Brake pad wear intelligent measurement method and system based on multi-sensor fusion

Abstract

The invention relates to the technical field of automobile active safety, in particular to an intelligent brake pad abrasion measuring method and system based on multi-sensor fusion, and the method comprises the steps that S1, multi-source sensing signals are synchronously collected, and wheel speed pulse signals and brake pressure signals from a vehicle CAN bus are received; s2, multi-modal signal feature extraction: carrying out time domain filtering processing on the original thickness signal to form a thickness change feature vector, carrying out space partition calculation on the temperature distribution signal to form a temperature gradient feature matrix, and carrying out wavelet packet decomposition on the braking vibration signal to form a frequency domain energy feature set; s3, dynamically fusing decisions, outputting a joint feature vector, and generating a wear loss estimated value and a confidence index; and S4, performing Kalman filtering correction on the wear loss estimated value according to the confidence index to form final wear thickness data. According to the brake pad wear intelligent measurement method and system based on multi-sensor fusion, the problem that brake pad wear cannot be accurately monitored and intelligently diagnosed online in real time can be solved.

Description

Technical Field

[0001] The present invention relates to the field of automobile active safety technology, and in particular to a brake pad wear intelligent measurement method and system based on multi-sensor fusion. Background Art

[0002] Wear monitoring technology for new energy vehicle drive brake systems faces new challenges: Traditional brake pad inspection relies on manual visual inspection or mechanical wear indicators, requires wheel hub removal, and can only alarm when extreme wear occurs, failing to meet the real-time monitoring needs of electric vehicle energy recovery braking systems. Among existing electronic monitoring solutions, ultrasonic thickness measurement is susceptible to electromagnetic interference from the motor and brake oil contamination, and visual inspection is not reliable enough under complex lighting conditions and multi-angle installation scenarios. Single temperature or vibration sensor solutions result in significant measurement errors due to high-frequency vibration of the motor and changes in regenerative braking conditions. Existing patented technologies use ultrasonic and temperature fusion monitoring, but do not address the signal distortion problem under severe vibration of the electric brake system; solutions based on image recognition have the disadvantages of high computational complexity and failure in rainy and foggy weather.

[0003] While multi-sensor fusion technology has been applied to tire pressure monitoring in new energy vehicles, brake pad wear monitoring in drive systems involves special operating conditions such as high motor temperatures, electromagnetic interference, and brake dust. Existing technologies lack a dynamic fusion mechanism tailored to the nonlinear degradation characteristics of electric brakes, making it difficult to achieve millimeter-level precision measurements during vehicle operation. With the increasing demand for brake safety in new energy vehicles driven by intelligent driving and brake-by-wire technology, there is an urgent need to develop an online monitoring method that is resistant to electromagnetic interference and adaptive to electric brake operating conditions. Summary of the Invention

[0004] In light of the shortcomings of the aforementioned prior art, the present invention aims to provide an intelligent brake pad wear measurement method and system based on multi-sensor fusion, addressing the challenges of real-time, online, accurate monitoring and intelligent diagnosis of brake pad wear. This method fuses multiple sensors (ranging, temperature, and vibration) with vehicle bus signals to extract multidimensional features such as thickness variation, temperature gradient, vibration frequency domain energy, and braking power. This method then utilizes a cross-modal attention mechanism in a deep neural network to dynamically fuse these features and generate a wear estimate. This method then incorporates a confidence-driven Kalman filter to correct and output the final wear data, addressing the issues of single-sensor susceptibility to interference and offline detection lag.

[0005] The present invention provides an intelligent brake pad wear measurement method based on multi-sensor fusion, comprising: S1: Synchronously collect multi-source sensor signals, generate thickness raw signals through the non-contact distance measuring device installed on the brake caliper assembly, generate temperature distribution signals through the temperature sensor array, generate brake vibration signals through the vibration sensor, and simultaneously receive wheel speed pulse signals and brake pressure signals from the vehicle CAN bus; S2: Extract multimodal signal features, perform time-domain filtering on the original thickness signal to generate a thickness variation feature vector, perform spatial partitioning calculation on the temperature distribution signal to generate a temperature gradient feature matrix, perform wavelet packet decomposition on the brake vibration signal to generate a frequency domain energy feature set, and fuse the wheel speed pulse signal with the brake pressure signal to generate a real-time brake power parameter. S3: Dynamic fusion decision-making: input the thickness change feature vector, temperature gradient feature matrix, frequency domain energy feature set and real-time braking power parameters into the pre-trained deep neural network model, fuse the multi-source features through the cross-modal attention mechanism, output the joint feature vector and generate the wear amount estimation value and confidence index; S4: Perform Kalman filtering correction on the wear amount estimation value according to the confidence index to form the final wear thickness data.

[0006] In one embodiment of the present invention, the non-contact distance measuring device is installed on the metal base inside the dust cover of the brake caliper piston, and its signal transmission direction is facing the brake pad back plate. The trigger condition for collecting the original thickness signal is that the brake pressure signal continuously exceeds the preset pressure threshold and the wheel speed pulse signal indicates that the vehicle is in a non-stationary state. At the same time, the collection of the temperature distribution signal must meet the requirement that the brake disc temperature is in the stable range of the material linear expansion coefficient. The vibration sensor captures the brake vibration signal through a three-axis accelerometer at a sampling rate greater than three times the natural frequency of the braking system.

[0007] In one embodiment of the present invention, the time domain filtering process adopts an adaptive noise cancellation algorithm, and the brake vibration signal is used as the reference noise source to dynamically reduce the noise of the original thickness signal. The construction of the thickness change feature vector includes the normalized sequence of the mean, variance and differential values ​​of adjacent sampling points within the sliding time window. The temperature gradient feature matrix is ​​formed by dividing the temperature sensor array into concentric annular areas, calculating the temperature difference between the center point and the edge point of each area, and associating it with the timestamp of the brake pressure signal to form a two-dimensional feature mapping table.

[0008] In one embodiment of the present invention, the basis function of wavelet packet decomposition selects Morlet wavelet that matches the resonance characteristics of the brake pad material, the frequency domain energy feature set extracts the energy proportion of eight sub-bands covering 50 Hz to 5 kHz, the real-time braking power parameter is calculated by integrating the instantaneous linear velocity and brake pressure signal converted from the wheel speed pulse signal, and the brake disc radius parameter is introduced to generate the unit time friction work curve through integral operation.

[0009] In one embodiment of the present invention, the deep neural network model includes a parallel dual-channel convolution module and a long short-term memory module. The convolution module processes the temperature gradient feature matrix and the frequency domain energy feature set, and the long short-term memory module processes the thickness change feature vector and the real-time braking power parameter. The cross-modal attention mechanism dynamically allocates feature fusion coefficients to generate a joint feature vector by calculating the correlation weight between the convolution channel output and the hidden state of the memory module.

[0010] In one embodiment of the present invention, the confidence index is generated based on the discreteness of the probability distribution of the joint eigenvector in the fully connected layer of the deep neural network. When the discreteness exceeds a preset fluctuation threshold, the Kalman filter correction is activated. The state equation of the Kalman filter introduces the average temperature rise rate of the temperature gradient characteristic matrix as a system noise compensation term, and the observation equation uses the sliding mean of the wear amount estimate as the measurement input.

[0011] In one embodiment of the present invention, a brake pad health status classification step is added after step S4, and the remaining service life percentage is calculated based on the final wear thickness data. At the same time, the high temperature area offset in the temperature gradient feature matrix and the main frequency band migration trend of the frequency domain energy feature set are analyzed. When it is detected that the spatial correlation between the local high temperature area and the main frequency band of the vibration energy exceeds a critical value, a brake pad material thermal degradation warning signal is generated.

[0012] In one embodiment of the present invention, the training data of the deep neural network model includes multi-source signals collected from a laboratory bench accelerated wear test and the true thickness calibrated by a laser displacement sensor. The training process adopts a transfer learning strategy, first pre-training the network encoder with degradation data generated by simulation, then fine-tuning the parameters of the entire network with real vehicle road test data, and adding adversarial samples to enhance the robustness of the model.

[0013] In one embodiment of the present invention, a dynamic signal quality monitoring link is set. When the proportion of effective sensing units in the temperature distribution signal is less than seventy percent or the signal-to-noise ratio of the original thickness signal does not reach the processable threshold for three consecutive sampling cycles, it automatically switches to the degradation mode and outputs the wear amount interval estimation value only relying on the mapping relationship between the frequency domain energy feature set of the vibration sensor and the brake pressure signal.

[0014] The present invention also includes an intelligent brake pad wear measurement system based on multi-sensor fusion, comprising: The acquisition module generates a thickness raw signal through a non-contact distance measuring device installed on the brake caliper assembly, a temperature distribution signal through a temperature sensor array, and a brake vibration signal through a vibration sensor. It also receives wheel speed pulse signals and brake pressure signals from the vehicle's CAN bus. The coordination module performs time-domain filtering on the original thickness signal to form a thickness variation feature vector, performs spatial partitioning calculation on the temperature distribution signal to form a temperature gradient feature matrix, performs wavelet packet decomposition on the brake vibration signal to form a frequency-domain energy feature set, and fuses the wheel speed pulse signal with the brake pressure signal to form a real-time brake power parameter. The comparison module inputs the thickness change feature vector, temperature gradient feature matrix, frequency domain energy feature set, and real-time braking power parameters into a pre-trained deep neural network model. It fuses multi-source features through a cross-modal attention mechanism, outputs a joint feature vector, and generates a wear estimate and confidence index. The analysis module performs Kalman filtering correction on the wear amount estimation value according to the confidence index to form the final wear thickness data.

[0015] The intelligent brake pad wear measurement method and system based on multi-sensor fusion provided by the present invention extracts multi-dimensional features such as thickness change, temperature gradient, vibration frequency domain energy and braking power by fusing multiple sensors (ranging / temperature / vibration) with vehicle bus signals. It uses a cross-modal attention mechanism deep neural network to dynamically fuse features to generate wear estimation, and combines confidence-driven Kalman filtering to correct and output the final wear data, solving the problems of single sensor susceptibility to interference and offline detection lag. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing 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.

[0017] Figure 1 The flowchart of the intelligent measurement method of brake pad wear based on multi-sensor fusion is shown in FIG. Figure 2 Schematic diagram showing steps three and four of the intelligent brake pad wear measurement method based on multi-sensor fusion; Figure 3 This is the system architecture diagram of the brake pad wear intelligent measurement system based on multi-sensor fusion. DETAILED DESCRIPTION

[0018] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0019] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0020] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0021] See Figure 1-Figure 3 , shown is the intelligent measurement method and system of brake pad wear based on multi-sensor fusion of the present invention. The intelligent brake pad wear measurement method based on multi-sensor fusion of the present invention includes the following steps: S1: synchronously collecting multi-source sensor signals, generating a thickness raw signal through a non-contact distance measurement device installed on the brake caliper assembly, generating a temperature distribution signal through a temperature sensor array, generating a brake vibration signal through a vibration sensor, and simultaneously receiving a wheel speed pulse signal and a brake pressure signal from the vehicle's CAN bus; S2: multi-modal signal feature extraction, performing time-domain filtering on the thickness raw signal to form a thickness change feature vector, performing spatial partitioning calculation on the temperature distribution signal to form a temperature gradient feature matrix, performing wavelet packet decomposition on the brake vibration signal to form a frequency domain energy feature set, and fusing the wheel speed pulse signal with the brake pressure signal to form a real-time brake power parameter; S3: dynamic fusion decision-making, inputting the thickness change feature vector, the temperature gradient feature matrix, the frequency domain energy feature set, and the real-time brake power parameter into a pre-trained deep neural network model, fusing multi-source features through a cross-modal attention mechanism, outputting a joint feature vector, and generating a wear amount estimate and a confidence index; and S4: performing Kalman filtering correction on the wear amount estimate based on the confidence index to form final wear thickness data.

[0022] like Figure 1As shown, the core method of the present invention begins with the synchronous acquisition of multi-source sensor signals. By deploying a non-contact distance measuring device at a key position of the brake caliper assembly, an original electrical signal representing the remaining thickness of the brake pad is generated. The device uses millimeter wave or laser distance measurement principles to penetrate oily and dusty environments to obtain thickness data. At the same time, the temperature sensor array integrated on the back of the brake caliper is arranged in a matrix to capture the surface temperature distribution signal of the brake pad. The spacing between its thermocouples or infrared sensor units is no more than three millimeters to ensure thermal field resolution. The vibration sensor is rigidly connected to the brake caliper housing to collect high-frequency vibration signals transmitted by the friction interface during braking. The wheel speed pulse signal and brake pressure signal provided in real time by the vehicle control bus are connected to the processing unit through a hard line. The five types of signals are synchronized at the microsecond level through hardware timestamps and software interrupt mechanisms. After signal acquisition, the multimodal feature extraction stage begins: the original thickness signal is first filtered by an adaptive bandwidth time domain filter to eliminate mechanical vibration coupling noise, and the normalized sequence of mean, variance and difference values ​​within the sliding time window is extracted to form a thickness change feature vector; the temperature distribution signal is divided into concentric annular areas to calculate the radial and circumferential temperature gradients, and the temperature gradient feature matrix is ​​constructed in combination with the brake pressure trigger time point; the brake vibration signal is decomposed into eight sub-bands using Morlet wavelet packets, and the energy proportion of each frequency band from 50 Hz to 5 kHz is extracted to form a frequency domain energy feature set; the wheel speed pulse signal is converted into instantaneous linear velocity through the wheel radius, and integrated with the brake pressure signal according to the friction work model to generate real-time brake power parameters. During the dynamic fusion decision phase, the aforementioned four-dimensional features are fed into a pre-trained deep neural network. This network utilizes a dual-channel architecture: a convolutional channel processes the graphical input of the temperature matrix and vibration frequency domain features, while a long-short-term memory channel analyzes the temporal evolution of the thickness vector and brake power parameters. A cross-modal attention mechanism calculates the correlation matrix between the convolutional feature map and the hidden states of the memory cells, dynamically assigns fusion weights to generate a joint feature vector, and finally outputs a wear estimate and a confidence indicator based on the discreteness of the probability distribution through a fully connected layer. The final correction phase triggers a Kalman filter based on the confidence threshold. When the confidence level falls below 0.3, the average temperature rise rate of the temperature gradient feature matrix is ​​used to compensate for system noise. The ten-time sliding mean of the wear estimate is used as the observation input. The final wear thickness data is then iteratively optimized using a state-space model, achieving a measurement accuracy of ±0.2 mm under high-temperature vibration conditions. This completely addresses the technical drawbacks of traditional methods, which require disassembly for testing and suffer from poor real-time performance.

[0023] Furthermore, the installation position of the non-contact distance measuring device is selected as the titanium alloy base inside the dust cover of the brake caliper piston. The base is laser welded with the brake caliper body to form a thermal isotropic body to avoid temperature difference deformation. The signal emission axis is aligned with the center area of ​​the brake pad back plate in a vertical incidence manner to ensure that the measurement point avoids interference from the back plate rivets and the edge of the wear groove; the acquisition of the original thickness signal must meet three conditions at the same time: the brake pressure signal continuously exceeds the vehicle idle braking threshold (usually greater than five bars), the vehicle speed analyzed by the wheel speed pulse signal is greater than five kilometers per hour and lasts for more than three seconds, and the temperature in the center area of ​​the temperature distribution signal is in the material linear expansion stable range of 80 to 350 degrees Celsius, thereby avoiding measurement distortion caused by cold start thermal deformation and insufficient low-speed creep friction; temperature sensing The array adopts a star-shaped arrangement of sixteen units. The peripheral units cover twenty percent of the outer edge of the brake pad friction radius to capture abnormal edge wear characteristics. All sensor units are encapsulated in an aluminum nitride ceramic substrate to resist high-temperature oxidation. The vibration sensor uses a three-axis accelerometer with a range of ±50 grams. Its mounting surface is grinded to a surface roughness of less than 0.8 microns to ensure rigid coupling. The sampling rate is set to more than three times the third-order natural frequency of the braking system (typical value is greater than twelve kilohertz), and is combined with a fourth-order Butterworth anti-aliasing filter to fully capture the brake pad resonance frequency band. The design ensures that valid original data can be obtained under extreme working conditions such as brake disc thermal warping, metal debris flying and high-frequency mechanical impact through spatial positioning constraints and environmental parameter threshold control, laying a physical foundation for subsequent feature extraction.

[0024] like Figure 2As shown, the raw thickness signal is processed using an improved adaptive noise cancellation algorithm. Using the Z-axis component of the brake vibration signal as the reference noise source, the coefficients of the finite-length unit impulse response filter are dynamically adjusted using the minimum mean square error criterion, eliminating up to 90 percent of mechanical vibration coupling interference. The thickness variation feature vector is constructed from the noise-reduced signal by calculating the signal mean within a 200-millisecond sliding window to reflect the overall wear trend, the variance to characterize the measurement fluctuation caused by brake judder, and the differential values ​​of adjacent sampling points, normalized to reveal the instantaneous wear rate variation. These three features are then concatenated into a feature vector with a weighting ratio of 6:3:1. The temperature gradient feature matrix is ​​generated by combining spatial partitioning and temporal correlation. The temperature sensor array is divided into five concentric annular regions with an inner diameter of 20 mm and an outer diameter of 80 mm. The radial gradient value is calculated as the average temperature difference between the center and edge of each annular region. Temperature values ​​are then simultaneously extracted for each annular region at three time points: the braking pressure rising edge, the holding phase, and the release phase. This creates a two-dimensional feature map containing the radial gradient and its temporal evolution. This map clearly illustrates the temperature field distortion pattern caused by localized hot spots. The frequency-domain energy feature set uses wavelet packet decomposition to deeply extract wear information from vibration signals. The basis function uses a Morlet wavelet with adjustable center frequency, and its waveform parameters are calibrated to a bandwidth factor of 1.5 and a center frequency of 1.75 based on the resonant characteristics of the brake pad material. The decomposition layer is set to four, generating sixteen sub-bands. The eight sub-bands with the highest energy content between 50 Hz and 5 kHz are selected, and their energy values ​​are calculated as a proportion of the total energy in the entire frequency band to form a feature set. Energy mutations in the frequency band around 2,500 Hz are specifically monitored to identify friction material delamination failures. Real-time brake power parameters are integrated with vehicle bus signals and physical models. The wheel speed pulse signal is converted to instantaneous linear velocity using the wheel rolling radius. The instantaneous friction force is calculated using the brake pressure signal with a friction coefficient of 0.35. This is then multiplied by the linear velocity to obtain the instantaneous friction power. Finally, the friction power is integrated over a single braking cycle to generate a friction work per unit time curve. The slope of this curve quantifies the nonlinear effect of braking intensity on wear rate.

[0025] Furthermore, the basis function of wavelet packet decomposition is selected based on the physical matching of the resonance characteristics of the brake pad material, and the Morlet wavelet with adjustable time-frequency resolution is used as the core processing device. Its bandwidth factor is calibrated to 1.5 to match the second-order bending mode of the cast iron brake disc, and the center frequency is set to 1.75 times the fundamental frequency to ensure that the impact response frequency band caused by the shedding of the friction material is covered; the decomposition process executes a four-level tree structure to generate sixteen equal-width sub-bands, and the eight characteristic sub-bands with the highest energy concentration in the range of 50 Hz to 5 kHz are screened by the kurtosis coefficient (typically including the 800 Hz bushing loose characteristic band, the 2 kHz friction pad crack characteristic band and the 4 kHz back plate resonance characteristic band), and the proportion of each sub-band energy to the total energy of the full frequency band is calculated to form an eight-dimensional frequency domain energy feature set, especially monitoring The energy mutation index in the frequency band near 2,500 Hz is used as a warning indicator for material stratification; the generation of real-time braking power parameters integrates multi-source information from the vehicle bus: the wheel speed pulse signal is converted into instantaneous linear velocity through the wheel rolling radius, and the brake pressure signal is mapped to the actual force of the hydraulic cylinder after nonlinear correction of the pressure sensor range. The effective friction radius parameter of the brake disc is introduced to construct an instantaneous friction torque model, and the instantaneous linear velocity and friction torque are multiplied to obtain millisecond-level friction power sampling values. Finally, the power curve from the rising edge of pressure to the end of release in a single braking event is numerically integrated to generate a characteristic curve of friction work per unit time in kilojoules. This curve shows an exponential growth characteristic under continuous emergency braking conditions, and its logarithmic slope is used to quantify the accelerating effect of extreme braking on the wear rate.

[0026] like Figure 2As shown in the figure, the deep neural network adopts a heterogeneous dual-channel architecture to realize multimodal feature processing: the convolution channel contains three levels of residual modules. The first level uses a three-by-three convolution kernel to extract the spatial topological pattern of the temperature gradient feature matrix (such as the circular diffusion or strip-shaped aggregation of hot spots). The second level expands the receptive field through void convolution to capture the temperature field correlation characteristics of the entire brake pad. The third level uses the channel attention mechanism to enhance the feature weights of abnormal temperature zones. The long-term and short-term memory channel is designed as a bidirectional three-layer structure. The input layer processes the time domain trend of the thickness change feature vector, the hidden layer integrates the nonlinear fluctuation characteristics of the real-time braking power parameters, and the output layer generates a hidden state vector that characterizes the wear evolution direction. The cross-modal attention mechanism serves as the fusion hub. First, the end of the convolution channel is converted into a vector. The feature map of the layer is flattened into a spatial feature vector, and the cosine similarity matrix is ​​calculated with the hidden state vector of the long short-term memory channel. The fusion weight coefficient of 0.1 to 0.9 is dynamically assigned according to the similarity value, and the weighted sum is used to generate a joint feature vector; after the dimensionality reduction of the joint vector by the fully connected layer, two key parameters are output: one is the wear amount estimate mapped to a range of 0 to 15 mm through the Sigmoid activation function, and the other is the confidence index calculated based on the probability distribution of the output layer neurons (the Jensen-Shannon divergence is used to quantify the discreteness of the probability distribution); this architecture significantly improves the ability to recognize asymmetric temperature fields and vibration characteristics caused by uneven wear of brake pads through a physically driven feature weight allocation mechanism.

[0027] Specifically, the application of confidence indicators is reflected in the intelligent correction trigger strategy: when the discreteness of the probability distribution output by the neural network exceeds the 0.3 threshold (indicating that the estimation result is highly uncertain), the Kalman filter correction process is activated; the filter state equation is defined as a first-order Markov model of wear thickness, and its system noise term introduces the radial average temperature rise rate of the temperature gradient characteristic matrix as a compensation factor (for every increase of 10 degrees Celsius per second in the temperature rise rate, the system noise variance expands by 0.05 mm squared), effectively offsetting the measurement drift caused by thermal expansion; the observation equation uses the ten sliding mean values ​​of the wear amount estimate as the measurement input (window width Covering three complete braking cycles), the measurement noise variance is dynamically adjusted based on the fluctuation amplitude of the brake power parameters (for every kilowatt increase in the power standard deviation, the noise variance increases by 0.02 mm squared). The filter executes a prediction-update loop: the prediction step extrapolates the current wear thickness prior estimate based on the state at the previous moment, and the update step fuses the current observations to calculate the posterior estimate. The final output is the corrected wear thickness data, whose maximum correction amplitude is limited by the convergence characteristics of the Kalman gain matrix (the steady-state gain is controlled in the range of 0.2 to 0.5), ensuring that the measurement error can still be suppressed to within 0.2 mm when the thermal deformation of the brake disc reaches 300 microns.

[0028] like Figure 3As shown, the present invention also includes an intelligent brake pad wear measurement system based on multi-sensor fusion, including: an acquisition module, which generates a thickness raw signal through a non-contact distance measurement device installed on the brake caliper assembly, generates a temperature distribution signal through a temperature sensor array, generates a brake vibration signal through a vibration sensor, and simultaneously receives a wheel speed pulse signal and a brake pressure signal from the vehicle CAN bus; a coordination module, which performs time-domain filtering on the thickness raw signal to form a thickness change feature vector, performs spatial partitioning calculation on the temperature distribution signal to form a temperature gradient feature matrix, performs wavelet packet decomposition on the brake vibration signal to form a frequency domain energy feature set, and fuses the wheel speed pulse signal with the brake pressure signal to form a real-time brake power parameter; a comparison module, which inputs the thickness change feature vector, the temperature gradient feature matrix, the frequency domain energy feature set, and the real-time brake power parameter into a pre-trained deep neural network model, fuses multi-source features through a cross-modal attention mechanism, outputs a joint feature vector, and generates a wear amount estimate and a confidence index; and an analysis module, which performs Kalman filtering correction on the wear amount estimate according to the confidence index to form final wear thickness data.

[0029] like Figure 3 As shown, after the final wear thickness data is output, a three-level health status diagnosis is performed: the first level converts the remaining thickness into a percentage of the service life based on the material wear model (100% when the thickness of the new plate is 10 mm, and 0% when the wear limit is 2 mm), and divides the health status into five levels according to a 20% gradient; the second level analyzes the spatial abnormality pattern of the temperature gradient feature matrix, calculates the offset of the center of mass of the high-temperature area (more than 50% of the average temperature) relative to the geometric center of the brake pad, and extracts the energy center migration distance of the main frequency band of 2,000 Hz to 3,000 Hz in the frequency domain energy feature concentration (the main frequency band is stable at 2,200 Hz ± 50 Hz during normal wear); the third level establishes a thermal The vibration-coupled fault judgment model identifies a localized thermal degradation fault in the brake pad when the spatial correlation coefficient between the high-temperature zone center of mass offset and the main frequency band migration distance exceeds 0.7 and persists for three consecutive braking cycles. The diagnostic results are integrated into a structured report: the service life percentage is displayed as a progress bar on the vehicle instrument panel, a localized thermal degradation warning triggers an audible and visual alarm, and a fault code is stored. Key parameters (including final wear thickness, high-temperature zone coordinates, and main frequency band migration values) are simultaneously uploaded to the cloud-based diagnostic platform via the vehicle bus. This design achieves a functional leap from basic thickness measurement to material failure warning, particularly addressing the hidden safety risks caused by carbonization of friction materials.

[0030] The intelligent brake pad wear measurement method and system based on multi-sensor fusion of the present invention extracts multi-dimensional features such as thickness change, temperature gradient, vibration frequency domain energy and braking power through the fusion of multi-sensor and vehicle bus signals, uses a cross-modal attention mechanism deep neural network to dynamically fuse features to generate wear estimation, and combines confidence-driven Kalman filtering to correct and output the final wear data, solving the problems of single sensor susceptibility to interference and offline detection lag.

[0031] Therefore, the intelligent brake pad wear measurement method and system based on multi-sensor fusion of the present invention can solve the problem of being unable to achieve real-time online accurate monitoring and intelligent diagnosis of brake pad wear.

[0032] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. The intelligent measurement method of brake pad wear based on multi-sensor fusion is characterized by: include: S1: Synchronously collect multi-source sensor signals, generate thickness raw signals through the non-contact distance measuring device installed on the brake caliper assembly, generate temperature distribution signals through the temperature sensor array, generate brake vibration signals through the vibration sensor, and simultaneously receive wheel speed pulse signals and brake pressure signals from the vehicle CAN bus; S2: Extract multimodal signal features, perform time-domain filtering on the original thickness signal to generate a thickness variation feature vector, perform spatial partitioning calculation on the temperature distribution signal to generate a temperature gradient feature matrix, perform wavelet packet decomposition on the brake vibration signal to generate a frequency domain energy feature set, and fuse the wheel speed pulse signal with the brake pressure signal to generate a real-time brake power parameter. S3: Dynamic fusion decision-making: input the thickness change feature vector, temperature gradient feature matrix, frequency domain energy feature set and real-time braking power parameters into the pre-trained deep neural network model, fuse the multi-source features through the cross-modal attention mechanism, output the joint feature vector and generate the wear amount estimation value and confidence index; S4: Perform Kalman filtering correction on the wear amount estimation value according to the confidence index to form the final wear thickness data.

2. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: The non-contact distance measurement device is installed on the metal base inside the brake caliper piston dust cover, and its signal transmission direction is facing the brake pad backplate. The trigger condition for collecting the original thickness signal is that the brake pressure signal continuously exceeds the preset pressure threshold and the wheel speed pulse signal indicates that the vehicle is in a non-stationary state. At the same time, the collection of the temperature distribution signal must meet the requirement that the brake disc temperature is within the stable range of the material linear expansion coefficient. The vibration sensor captures the brake vibration signal through a three-axis accelerometer at a sampling rate greater than three times the natural frequency of the braking system.

3. The intelligent measurement method for brake pad wear based on multi-sensor fusion according to claim 1 is characterized in that: The time-domain filtering process adopts an adaptive noise cancellation algorithm, using the brake vibration signal as a reference noise source to dynamically reduce the noise of the original thickness signal. The thickness change feature vector is constructed by including a normalized sequence of the mean, variance, and differential values ​​of adjacent sampling points within a sliding time window. The temperature gradient feature matrix is ​​formed by dividing the temperature sensor array into concentric annular areas, calculating the temperature difference between the center point and the edge point of each area, and associating it with the timestamp of the brake pressure signal to form a two-dimensional feature map.

4. The intelligent measurement method for brake pad wear based on multi-sensor fusion according to claim 1 is characterized in that: The basis function of the wavelet packet decomposition selects the Morlet wavelet that matches the resonance characteristics of the brake pad material. The frequency domain energy feature set extracts the energy proportion of eight sub-bands covering 50 Hz to 5 kHz. The calculation of the real-time braking power parameter integrates the instantaneous linear velocity converted from the wheel speed pulse signal and the brake pressure signal, and introduces the brake disc radius parameter to generate the unit time friction work curve through integration operation.

5. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: The deep neural network model includes a parallel dual-channel convolution module and a long short-term memory module. The convolution module processes the temperature gradient feature matrix and the frequency domain energy feature set, and the long short-term memory module processes the thickness change feature vector and the real-time braking power parameter. The cross-modal attention mechanism calculates the correlation weight between the convolution channel output and the hidden state of the memory module, and dynamically allocates feature fusion coefficients to generate a joint feature vector.

6. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: The confidence index is generated based on the discreteness of the probability distribution of the joint eigenvector in the fully connected layer of the deep neural network. When the discreteness exceeds the preset fluctuation threshold, the Kalman filter correction is activated. The state equation of the Kalman filter introduces the average temperature rise rate of the temperature gradient characteristic matrix as a system noise compensation term, and the observation equation uses the sliding mean of the wear amount estimate as the measurement input.

7. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: After step S4, a brake pad health status classification step is added to calculate the remaining service life percentage based on the final wear thickness data. At the same time, the high-temperature area offset in the temperature gradient feature matrix and the main frequency band migration trend of the frequency domain energy feature set are analyzed. When the spatial correlation between the local high-temperature area and the main frequency band of the vibration energy is detected to exceed the critical value, a brake pad material thermal degradation warning signal is generated.

8. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: The training data for the deep neural network model includes multi-source signals collected from laboratory bench accelerated wear tests and the true thickness calibrated by a laser displacement sensor. The training process adopts a transfer learning strategy, first pre-training the network encoder with simulated degradation data, then fine-tuning the full network parameters using real vehicle road test data, and adding adversarial samples to enhance the model's robustness.

9. The intelligent brake pad wear measurement method based on multi-sensor fusion according to claim 1 is characterized in that: A dynamic signal quality monitoring link is set up. When the proportion of effective sensing units in the temperature distribution signal is less than 70% or the signal-to-noise ratio of the original thickness signal does not reach the processable threshold for three consecutive sampling cycles, it automatically switches to the degraded mode and outputs the wear range estimate only based on the mapping relationship between the frequency domain energy feature set of the vibration sensor and the brake pressure signal.

10. A data mining system using the intelligent brake pad wear measurement method based on multi-sensor fusion according to any one of claims 1 to 9, characterized in that: include: an acquisition module, which generates a thickness raw signal through a non-contact distance measuring device installed on the brake caliper assembly, generates a temperature distribution signal through a temperature sensor array, generates a brake vibration signal through a vibration sensor, and simultaneously receives a wheel speed pulse signal and a brake pressure signal from the vehicle's CAN bus; A coordination module that performs time-domain filtering on the original thickness signal to form a thickness variation feature vector, performs spatial partitioning calculation on the temperature distribution signal to form a temperature gradient feature matrix, performs wavelet packet decomposition on the brake vibration signal to form a frequency-domain energy feature set, and fuses the wheel speed pulse signal with the brake pressure signal to form a real-time brake power parameter; a comparison module, which inputs the thickness change feature vector, the temperature gradient feature matrix, the frequency domain energy feature set, and the real-time braking power parameter into a pre-trained deep neural network model, fuses the multi-source features through a cross-modal attention mechanism, outputs a joint feature vector, and generates a wear amount estimate and a confidence index; An analysis module performs Kalman filtering correction on the wear amount estimation value according to the confidence index to form final wear thickness data.

Images