Brake life prediction method and system

Abstract

The application provides a brake life prediction method and system, and relates to the technical field of deep learning. Firstly, a wear deformation atlas of a target brake under continuous braking conditions is constructed. Then, surface morphology reconstruction processing is performed on the wear deformation atlas to obtain a surface morphology reconstruction model. Then, the surface morphology reconstruction model is input into a preset stress field evolution network to generate an internal stress field inversion graph. Then, material fatigue positioning processing is performed according to the internal stress field inversion graph to obtain a material fatigue positioning distribution graph. Finally, residual life prediction processing is performed based on the material fatigue positioning distribution graph to generate a brake life prediction result. The application can comprehensively and accurately record the wear deformation process, finely describe the surface characteristics, reveal the connection between the surface and the internal stress, accurately locate the fatigue damage, and comprehensively and accurately predict the residual life.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and more specifically, to a method and system for predicting brake lifespan. Background Technology

[0002] During the use of brakes, life prediction is crucial for ensuring safe equipment operation, rationally scheduling maintenance plans, and reducing operating costs. Under continuous braking conditions, the friction pair surfaces of brakes undergo complex wear and deformation processes. This wear and deformation not only affects the braking performance of the brake but also leads to changes in the internal stress field, resulting in cumulative material fatigue damage and ultimately impacting the service life of the brake.

[0003] Currently, traditional brake life prediction methods have many limitations. Some methods are mainly based on empirical formulas or simple physical models, which often ignore the dynamic characteristics of wear deformation during braking and the complex relationship between internal stress fields and material fatigue, leading to significant deviations between prediction results and actual conditions. Other methods, while employing advanced detection technologies such as optical and ultrasonic testing to obtain brake surface information, typically only acquire localized information, failing to comprehensively and accurately reflect the wear deformation and internal stress field changes throughout the brake's entire service life. Furthermore, existing methods lack effective analysis and modeling tools when processing complex wear deformation and stress field data, failing to delve into the underlying physical laws, thus limiting the accuracy and reliability of brake life prediction. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a brake life prediction method, the method comprising:

[0005] A wear deformation atlas of the target brake under continuous braking conditions is constructed. The wear deformation atlas is a set of graph structures that reflect the continuous deformation process of wear morphology after mapping the wear morphology changes of the friction pair surface under continuous braking action in the form of a three-dimensional grid structure according to the braking action sequence.

[0006] The wear deformation atlas is subjected to surface morphology reconstruction processing to obtain a surface morphology reconstruction model. The surface morphology reconstruction model is a three-dimensional mesh model composed of multiple reconstructed triangular facets. Each reconstructed triangular facet is associated with and stored as a surface curvature value as a morphology feature attribute label.

[0007] The surface morphology reconstruction model is input into a preset stress field evolution network for stress field inversion processing to generate an internal stress field inversion map. The preset stress field evolution network is a deep learning network model that includes a wear morphology feature extraction layer, a morphology spatial encoding layer, a graph convolution processing layer, a stress field evolution simulation layer, and an inversion mapping layer. It is used to map surface morphology information into an internal stress field distribution. The internal stress field inversion map is a three-dimensional spatial distribution map containing the correspondence between surface morphology and internal stress field generated by spatially superimposing the isosurface of the stress field inversion value with the surface morphology reconstruction model.

[0008] Based on the internal stress field inversion diagram, material fatigue location processing is performed to obtain a material fatigue location distribution map. The material fatigue location distribution map is a three-dimensional distribution map containing the fatigue cumulative damage degree parameter of each spatial location point after filling in the fatigue cumulative damage degree parameter of each spatial location point according to the spatial coordinate position.

[0009] Based on the material fatigue location distribution map, the remaining life of the target brake is predicted, generating a brake life prediction result that includes the minimum remaining effective thickness parameter and the remaining life cycle parameter.

[0010] Furthermore, embodiments of the present invention also provide a brake life prediction system, comprising:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described brake life prediction method by executing the machine-executable instructions.

[0012] Based on the above, a wear deformation atlas of the target brake under continuous braking conditions was constructed. This atlas, in the form of a three-dimensional mesh structure, recorded the continuous deformation process of the friction pair surface wear morphology. Surface topography reconstruction processing was performed on the wear deformation atlas to obtain a surface topography reconstruction model containing surface curvature value topography feature attribute labels. This model can describe the geometric features of the brake surface. The surface topography reconstruction model was input into a preset stress field evolution network for stress field inversion processing, generating an internal stress field inversion map. Through a multi-layer deep learning structure, this map effectively maps surface topography information to internal stress field distribution, revealing the intrinsic relationship between surface wear and internal stress. Material fatigue localization processing was performed based on the internal stress field inversion map to obtain a material fatigue localization distribution map. This map accurately locates the degree of fatigue cumulative damage at various spatial points inside the brake. Finally, based on the material fatigue localization distribution map, the remaining life of the target brake was predicted. The generated brake life prediction result includes the minimum remaining effective thickness parameter and the remaining life cycle parameter, comprehensively and accurately reflecting the remaining service life of the brake and effectively improving the safety of brake use. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the execution flow of the brake life prediction method provided in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of exemplary hardware and software components of the brake life prediction system provided in an embodiment of the present invention. Detailed Implementation

[0015] Figure 1 This is a flowchart illustrating a brake life prediction method according to an embodiment of the present invention, which will be described in detail below.

[0016] Step S110: Construct a wear deformation atlas of the target brake under continuous braking conditions. The wear deformation atlas is a set of graph structures that reflect the continuous deformation process of wear morphology after mapping the wear morphology changes of the friction pair surface under continuous braking action in the form of a three-dimensional grid structure according to the braking action sequence.

[0017] In this embodiment, a disc brake of a certain type of vehicle is used as the target brake, whose friction pair consists of a brake disc and brake pads. To construct a wear deformation atlas of this brake under continuous braking conditions, it is first necessary to obtain its surface morphology change data during continuous braking. In the brake bench test, a series of continuous braking actions are set up, each braking action including the complete process of brake pressure application, frictional contact, and brake release. After each braking action is completed, a high-precision three-dimensional optical profilometer is used to scan the friction surface of the brake disc to obtain the surface wear morphology image after that braking action.

[0018] Step S111: Obtain the surface wear morphology image corresponding to each braking action of the target brake under continuous braking conditions, and arrange the surface wear morphology images corresponding to each braking action according to the execution time sequence of the braking action to form a wear morphology image time sequence sequence with braking action time sequence identifier.

[0019] In the continuous braking test, N braking actions were performed, labeled as braking action T1, braking action T2, braking action T3, and so on up to braking action TN. After each braking action, the three-dimensional topographic data of the central annular area of ​​the brake disc surface, i.e., the main area in contact with the brake pads, was acquired using an optical profilometer, generating corresponding surface wear topographic images, denoted as surface wear topographic image I1, surface wear topographic image I2, surface wear topographic image I3, and so on up to surface wear topographic image IN. The resolution of the above surface wear topographic images is M rows by M columns, and each pixel stores the height value h(x, y) at that position. Subsequently, surface wear topographic images I1, I2, I3, and so on up to surface wear topographic image IN were arranged according to the execution sequence of the braking actions, i.e., braking action T1, braking action T2, braking action T3, and so on up to braking action TN, forming a wear topographic image time sequence with braking action time sequence identifiers.

[0020] Step S112: Perform image pixel grayscale value difference calculation processing on the wear morphology images corresponding to adjacent braking actions in the time sequence of the wear morphology images to obtain a grayscale difference distribution map of the wear morphology images between adjacent braking actions. Based on the grayscale difference distribution map, extract the pixel areas where the grayscale change gradient exceeds the preset grayscale change gradient threshold and mark them as areas with significant wear morphology changes.

[0021] Adjacent wear morphology images are sequentially extracted from the aforementioned time-series of wear morphology images, such as surface wear morphology image I1 and surface wear morphology image I2. For surface wear morphology images I1 and I2, the position (x, y) of each pixel is traversed, and the height difference Δh(x, y) = h2(x, y) - h1(x, y) at that position in the two surface wear morphology images is calculated. For each pixel, a square neighborhood window with a side length of 3 is constructed with that pixel as the center. The arithmetic mean μ_Δh of the height difference Δh of all pixels within the neighborhood window is calculated, and the standard deviation σ_Δh of the height difference Δh of all pixels within the neighborhood window is calculated. Then, the gray-level gradient G(x, y) of the pixel is G(x, y) = (Δh(x, y) - μ_Δh) / σ_Δh. After traversing all pixels, a gray-level difference distribution map is obtained. Set a grayscale gradient threshold G_th, extract the region where G(x,y)>G_th, and mark it as a region with significant wear morphology changes.

[0022] Step S113: Perform spatial coordinate calibration on the pixels in the area of ​​significant wear morphology change to obtain the set of boundary contour points of each area of ​​significant wear morphology change in the coordinate system of the friction pair surface, and input the set of boundary contour points into a preset wear deformation map generator to perform triangulation on the set of boundary contour points to generate a mesh structure of the area of ​​significant wear morphology change composed of multiple triangular facets.

[0023] Spatial coordinate calibration is performed on pixels within each region showing significant wear morphology changes. The calibration coefficient is set to the pixel spacing d, then the spatial coordinates (x, y) of the pixel located in the i-th row and j-th column are x=j d, y=i d, where the height value h(x, y) of this pixel is the z-coordinate. This yields the set of boundary contour points for each region showing significant wear morphology changes. This set of boundary contour points is then input into a preset wear deformation map generator. This generator uses the Delaunay triangulation algorithm to process the boundary contour point set, constructing a mesh composed of triangular facets. Each triangle's circumcircle does not contain any other points from the boundary contour point set, generating a mesh structure for regions showing significant wear changes composed of multiple triangular facets.

[0024] Step S114: Perform normal vector calculation on each triangular facet in the mesh structure of the area with significant wear changes to obtain the normal vector direction parameter of each triangular facet, and use the normal vector direction parameter of each triangular facet as the deformation attribute label of the corresponding triangular facet.

[0025] For any triangular facet in a mesh structure representing a region of significant wear variation, let the coordinates of its three vertices be A(x1, y1, z1), B(x2, y2, z2), and C(x3, y3, z3). Calculate the vector AB = (x2-x1, y2-y1, z2-z1) from vertex A to vertex B, and calculate the vector AC = (x3-x1, y3-y1, z3-z1) from vertex A to vertex C. Calculate the cross product of vectors AB and AC to obtain the normal vector N, where the x-component of N is (y2-y1). (z3-z1)-(z2-z1) (y3-y1), the y-component of N is (z2-z1). (x3-x1)-(x2-x1) (z3-z1), the z-component of N is (x2-x1). (y3-y1)-(y2-y1) (x3-x1). Normalize the normal vector N to obtain the normal vector direction parameter, which serves as the deformation attribute label for the corresponding triangular facet.

[0026] Step S115: Perform region segmentation processing on the mesh structure of the area with significant wear change according to the deformation attribute label of each triangular facet, and aggregate and classify triangular facets with the same deformation attribute label and spatially adjacent into the same wear change sub-region.

[0027] A region growing algorithm is used for region segmentation. All triangular faces in the mesh structure of regions with significant wear changes are traversed, and an unclassified triangular face is selected as a seed face. Adjacent triangular faces sharing edges with the seed face are examined, and the angle θ between the normal direction parameters of the adjacent triangular face and the normal direction parameter of the seed face is calculated, where cosθ is equal to the dot product of the two normal direction parameters. If θ is less than a preset angle threshold θ_th, the adjacent triangular face is assigned to the current region. The newly assigned triangular face is used as a new seed face to continue expanding until no adjacent triangular faces satisfy the condition. This process is repeated until all triangular faces are assigned to a certain wear-change sub-region.

[0028] Step S116: Perform vector synthesis processing on the normal vector direction parameters of all triangular facets in the same wear change sub-region, calculate the synthesized vector direction of the normal vector direction parameters of all triangular facets in the same wear change sub-region as the regional deformation direction parameter of the wear change sub-region, and associate and store it with the braking action timing identifier corresponding to the wear change sub-region to generate a wear change sub-region deformation direction record corresponding to each braking action.

[0029] For each segmented wear-change sub-region, let K be the number of triangular facets contained within the sub-region, and let the normal vector direction parameter of each triangular facet be a unit vector n1, n2, ..., nK. Add the x-components of each unit vector to obtain the composite vector: x-component S_x = ∑_{i=1}^{K}n_i_x, y-components S_y = ∑_{i=1}^{K}n_i_y, and z-components S_z = ∑_{i=1}^{K}n_i_z, resulting in the composite vector S = (S_x, S_y, S_z). Normalize this composite vector S to obtain the region deformation direction parameter. Associate this region deformation direction parameter with the braking action timing identifier corresponding to the current wear-change sub-region, generating a wear-change sub-region deformation direction record.

[0030] Step S117: Based on the deformation direction record of the wear change sub-region corresponding to each braking action, the wear change sub-regions corresponding to adjacent braking actions are associated and mapped according to the spatial position matching relationship to construct a wear deformation atlas that reflects the wear pattern of the friction pair surface under the continuous deformation process of the structure under continuous braking action.

[0031] For two adjacent braking actions Tj and Tj+1, obtain the deformation direction records of their corresponding wear change sub-regions. For each wear change sub-region R_j_i in braking action Tj, calculate the geometric center coordinates C_j_i of the sub-region, where the x-coordinate is equal to the average of the x-coordinates of all triangular facet vertices within the sub-region, and the y and z coordinates are calculated similarly. In the wear change sub-region deformation direction records of braking action Tj+1, find the wear change sub-region R_{j+1}_k whose geometric center coordinates are spatially closest to C_j_i, where the spatial distance d = √((x_j_i - x_{j+1}_k)^2 + (y_j_i - y_{j+1}_k)^2 + (z_j_i - z_{j+1}_k)^2), and establish an association mapping between the two to form a directed edge. Traverse all adjacent braking actions to construct a wear deformation atlas.

[0032] Step S120: Perform surface topography reconstruction processing on the wear deformation map to obtain a surface topography reconstruction model. The surface topography reconstruction model is a three-dimensional mesh model composed of multiple reconstructed triangular facets. Each reconstructed triangular facet is associated with and stored as a surface curvature value as a topography feature attribute label.

[0033] Step S121: Extract the wear deformation map corresponding to each braking action from the wear deformation map set, and sort the wear deformation map corresponding to each braking action according to the timing of the braking action to form a wear deformation map sequence with timing identifier.

[0034] The wear deformation map corresponding to each braking action is extracted from the wear deformation map set, and sorted according to the time sequence of the braking actions, namely braking action T1, braking action T2, braking action T3, and so on up to braking action TN, to form a wear deformation map sequence with time sequence identifier.

[0035] Step S122: Extract the grid vertex coordinates of each wear deformation map in the wear deformation map sequence to obtain the set of vertex spatial coordinates of all triangular facets in each wear deformation map. Based on the set of vertex spatial coordinates, calculate the coordinates of the geometric center point of the vertex spatial coordinates of all triangular facets in each wear deformation map to obtain the coordinate parameters of the geometric center point of each wear deformation map.

[0036] For each wear deformation map in the wear deformation map sequence, traverse all triangular patches within all wear deformation sub-regions, extract the coordinates of the three vertices of each triangular patch, and form a set of vertex spatial coordinates. Let this set contain P vertices, and the coordinates of each vertex be (x_p, y_p, z_p). Calculate the coordinates of the geometric center point: x_c = (∑_{p=1}^{P}x_p) / P, y_c = (∑_{p=1}^{P}y_p) / P, z_c = (∑_{p=1}^{P}z_p) / P.

[0037] Step S123: Using the geometric center point coordinates of each wear deformation map as a reference point, perform relative coordinate transformation on the spatial coordinates of the vertices of all triangular facets in each wear deformation map to generate a set of relative coordinates of all triangular facet vertices in each wear deformation map. Input the set of relative coordinates of the set of relative coordinates of all triangular facet vertices in each wear deformation map into the feature encoding layer of the preset surface topography reconstruction network, and perform spatial position feature extraction on the set of relative coordinates of all triangular facet vertices in each wear deformation map to obtain the spatial position feature encoding vector corresponding to each wear deformation map.

[0038] Using the geometric center coordinates (x_c, y_c, z_c) as a reference point, a relative coordinate transformation is performed on the coordinates (x, y, z) of each vertex in the vertex spatial coordinate set to obtain relative coordinates (x_r, y_r, z_r) = (x-x_c, y-y_c, z-z_c), forming a relative coordinate set. This relative coordinate set is then input into the feature encoding layer of a predefined surface topography reconstruction network. This feature encoding layer consists of multiple multilayer perceptron modules, which map the three-dimensional coordinates of each point into a high-dimensional feature vector. A max-pooling operation is used to obtain global features, which are then concatenated with the features of each point and passed through the multilayer perceptron to obtain the high-dimensional encoded features of each point. The aggregated features yield a spatial location feature encoding vector.

[0039] Step S124: Input the spatial position feature encoding vector corresponding to each wear deformation map into the feature fusion layer of the surface topography reconstruction network, perform feature splicing processing on the spatial position feature encoding vectors corresponding to adjacent braking actions in temporal order to generate a temporal feature fusion vector reflecting the continuous evolution process of wear morphology, input the temporal feature fusion vector into the topography reconstruction decoding layer of the surface topography reconstruction network, perform inverse feature mapping processing on the temporal feature fusion vector to generate a reconstructed surface three-dimensional point cloud data set.

[0040] The spatial location feature encoding vector corresponding to each wear deformation map is input into the feature fusion layer. The spatial location feature encoding vectors V_j and V_{j+1} corresponding to adjacent braking actions are extracted in temporal order and concatenated along the feature dimension to obtain the temporal feature fusion vector. This temporal feature fusion vector sequence is then input into the topography reconstruction decoding layer, which consists of an inverse multilayer perceptron module and an upsampling module. This layer progressively reduces the feature dimension and increases the number of points, outputting a reconstructed 3D point cloud dataset of the surface.

[0041] Step S125: Perform point cloud meshing processing on the reconstructed surface 3D point cloud data set. By connecting the point clouds in the reconstructed surface 3D point cloud data set according to spatial proximity, a surface 3D mesh model composed of multiple reconstructed triangular facets is generated. Perform surface curvature calculation processing on each reconstructed triangular facet in the surface 3D mesh model to obtain the surface curvature value of each reconstructed triangular facet. Use the surface curvature value of each reconstructed triangular facet as the shape feature attribute label of the corresponding reconstructed triangular facet.

[0042] A Poisson surface reconstruction algorithm is used to mesh the reconstructed 3D point cloud dataset. The normal vector of each point is calculated, an indicator function is constructed, and an approximate solution to the indicator function is obtained by solving the Poisson equation. Isosurfaces are extracted, generating a 3D surface mesh model composed of multiple reconstructed triangular facets. For each reconstructed triangular facet, the principal curvatures at its three vertices are calculated and averaged to obtain the surface curvature value. This surface curvature value is then stored as a morphological feature attribute label.

[0043] Step S126: Associate and store the shape feature attribute labels of each reconstructed triangular facet with the surface three-dimensional mesh model to generate a surface shape reconstruction model containing the shape feature attribute labels of each reconstructed triangular facet.

[0044] The topographic feature attribute label, i.e. the surface curvature value, of each reconstructed triangular facet is stored in the data structure of the corresponding reconstructed triangular facet, generating a surface topographic reconstruction model containing the topographic feature attribute label of each reconstructed triangular facet.

[0045] Step S130: Input the surface morphology reconstruction model into a preset stress field evolution network for stress field inversion processing to generate an internal stress field inversion map. The preset stress field evolution network is a deep learning network model that includes a wear morphology feature extraction layer, a morphology spatial encoding layer, a graph convolution processing layer, a stress field evolution simulation layer, and an inversion mapping layer. It is used to map surface morphology information into an internal stress field distribution. The internal stress field inversion map is a three-dimensional spatial distribution map containing the correspondence between surface morphology and internal stress field generated by spatially superimposing the isosurface of the stress field inversion value with the surface morphology reconstruction model.

[0046] Step S131: Input the surface topography reconstruction model into the topography feature extraction layer of the stress field evolution network, and perform feature vectorization processing on the topography feature attribute labels of each reconstructed triangular facet in the surface topography reconstruction model to obtain the topography feature vector of each reconstructed triangular facet.

[0047] The shape feature extraction layer traverses all reconstructed triangular facets. For each reconstructed triangular facet, its shape feature attribute label, namely the surface curvature value, as well as the area, three interior angles, and normal vector direction parameters of the reconstructed triangular facet, are concatenated to form the original feature vector. This original feature vector is then mapped to a high-dimensional feature space through a fully connected layer to obtain the shape feature vector.

[0048] Step S132: Input the topographic feature vector of each reconstructed triangular facet into the topographic space encoding layer of the stress field evolution network, map the topographic feature vector of each reconstructed triangular facet to a preset three-dimensional spatial coordinate system, generate the spatial encoding position coordinates corresponding to each reconstructed triangular facet, and construct a topographic space graph structure with each reconstructed triangular facet as a node and the spatial connection relationship between adjacent reconstructed triangular facets as edges based on the spatial encoding position coordinates corresponding to each reconstructed triangular facet.

[0049] The shape space encoding layer generates spatial encoded position coordinates for each reconstructed triangular facet, and transforms them using a position encoding function based on their geometric center coordinates. Based on the spatial encoded position coordinates of all reconstructed triangular facets, a shape space topology graph is constructed, with each reconstructed triangular facet as a node and the spatial connections between adjacent reconstructed triangular facets as edges.

[0050] Step S133: Input the topological graph structure of the topology space into the graph convolution processing layer of the stress field evolution network, and perform neighborhood information aggregation processing on the topological feature vector of each node in the topological graph structure to obtain the neighborhood aggregation feature vector corresponding to each node.

[0051] The graph convolutional processing layer employs a graph convolutional network architecture, updating node features through multiple graph convolutional layers. In each graph convolutional layer, for each node, the Euclidean distance d_ij between it and its neighboring nodes is calculated, and the aggregation weight w_ij = 1 / (d_ij^2). The neighborhood aggregation feature vector of node i is equal to the sum of the shape feature vectors of all neighboring nodes multiplied by w_ij, plus the shape feature vector of node i itself multiplied by the self-loop weight. This process of stacking multiple layers yields the neighborhood aggregation feature vector for each node.

[0052] Step S134: Through the stress field simulation layer of the stress field evolution network, the neighborhood aggregation feature vector corresponding to each node is input into the preset stress field differential equation solver, and the neighborhood aggregation feature vector of each node is iteratively solved to obtain the intermediate state vector of the stress field corresponding to each node.

[0053] The stress field simulation layer includes a pre-defined solver for the stress field differential equations, implemented using the finite difference method. The neighborhood aggregated feature vector of each node is used as the initial condition, and the spatial topological relationship of the topological graph is used as the solution domain. The linear elasticity equilibrium equations are discretized and solved iteratively to obtain the displacement vector of each node. Then, the strain tensor and stress tensor are calculated, and the six independent components of the stress tensor are used as the intermediate state vector of the stress field.

[0054] Step S135: Input the intermediate state vector of the stress field corresponding to each node into the inversion mapping layer of the stress field evolution network, perform inverse mapping transformation on the intermediate state vector of the stress field of each node to obtain the inversion value of the stress field corresponding to each node, and construct a three-dimensional stress field inversion point cloud data set with the spatial coding position coordinates of each node as the horizontal and vertical coordinates and the stress field inversion value of each node as the vertical coordinate.

[0055] The inversion mapping layer consists of a fully connected network. The intermediate state vector of the stress field of each node is input into the fully connected network, and the output scalar value is the inverted stress field value. The Mises equivalent stress is taken as the inverted stress field value. Based on the spatial encoded location coordinates of each node and the corresponding inverted stress field value, a three-dimensional stress field inversion point cloud data set is constructed.

[0056] Step S136: Perform spatial interpolation processing on the three-dimensional stress field inversion point cloud data set. By interpolating the point clouds in the three-dimensional stress field inversion point cloud data set according to spatial distance weights, a continuously distributed spatial distribution field of stress field inversion values ​​is generated. Perform isosurface extraction processing on the spatial distribution field of stress field inversion values, extract the spatial location points in the spatial distribution field of stress field inversion values ​​that are equal to the preset stress field inversion threshold, and connect the spatial location points to form an isosurface surface.

[0057] An inverse distance weighted interpolation algorithm is employed. For any point to be interpolated in 3D space, the spatial distance between it and all points in the 3D stress field inversion point cloud dataset is calculated. The closest points are selected as reference points. The interpolation weight of each reference point is equal to the square of the reciprocal of the spatial distance between that reference point and the point to be interpolated, divided by the sum of the squares of the reciprocals of the spatial distances between all reference points and the point to be interpolated. The stress field inversion value of the point to be interpolated is equal to the sum of the stress field inversion values ​​of all reference points multiplied by their corresponding interpolation weights, generating a continuously distributed spatial distribution field of stress field inversion values. A preset stress field inversion threshold is set, and the moving cube algorithm is used to extract the isosurface.

[0058] Step S137: Spatially superimpose the isosurface and the surface morphology reconstruction model to generate an internal stress field inversion map containing the correspondence between the surface morphology and the internal stress field.

[0059] The isosurface and the surface topography reconstruction model are placed in the same three-dimensional coordinate system and spatially superimposed using graphics rendering techniques. The surface topography reconstruction model is presented in a semi-transparent manner, while the isosurface is presented as colored contour lines, generating an inversion map of the internal stress field.

[0060] Step S140: Perform material fatigue location processing based on the internal stress field inversion map to obtain a material fatigue location distribution map. The material fatigue location distribution map is a three-dimensional distribution map containing the fatigue cumulative damage degree parameter of each spatial location point after filling in the fatigue cumulative damage degree parameter of each spatial location point according to the spatial coordinate position.

[0061] Step S141: Extract the stress field inversion value corresponding to each spatial location point from the internal stress field inversion diagram, and mark the spatial location points whose stress field inversion value is greater than the preset material yield stress threshold as the set of plastic deformation starting points.

[0062] The stress field inversion value of each spatial location point is extracted from the internal stress field inversion map. A preset material yield stress threshold σ_y is set, and spatial location points with stress field inversion values ​​greater than σ_y are marked as plastic deformation initiation points, thus forming a set of plastic deformation initiation points.

[0063] Step S142: Perform neighborhood expansion processing on each plastic deformation starting point in the set of plastic deformation starting points. Search for surrounding spatial locations with each plastic deformation starting point as the center point. Spatial locations whose stress field inversion value difference from the center point is less than a preset stress field inversion value difference threshold are assigned to the plastic deformation region corresponding to the center point, thus obtaining multiple plastic deformation regions.

[0064] For each point where plastic deformation begins, using that point as the center, search for adjacent spatial locations. Adjacency is defined as a spatial distance less than one voxel. For each adjacent point, calculate the absolute value of the difference between its inverted stress field value and the inverted stress field value at the center point. If this absolute value is less than a preset stress field inversion value difference threshold Δσ_th, then the adjacent point is included in the current plastic deformation region. The search continues with the newly included point as the center until no adjacent points satisfying the condition are found, resulting in multiple plastic deformation regions.

[0065] Step S143: Perform area calculation processing on each of the multiple plastic deformation regions to obtain the area parameter of each plastic deformation region, and select the plastic deformation regions whose area parameter is greater than the preset area parameter threshold as the main plastic deformation regions.

[0066] For each plastic deformation region, the moving cube algorithm is used to extract its outer surface mesh, and the sum of the areas of all triangular faces is calculated as the region area parameter. A preset region area parameter threshold A_th is set, and plastic deformation regions with a region area parameter greater than A_th are selected as the main plastic deformation regions.

[0067] Step S144: Perform region boundary curvature calculation processing on each of the main plastic deformation regions to obtain a region boundary curvature distribution map of each main plastic deformation region, and identify the boundary location points whose region boundary curvature values ​​exceed the preset region boundary curvature threshold as crack initiation candidate points based on the region boundary curvature distribution map.

[0068] Extract the boundary contour of each major plastic deformation region. For each boundary point on the boundary contour, take that point and its two adjacent boundary points; these three points define a local arc. The reciprocal of the radius of this local arc is the curvature value of that point. Traverse all boundary points to obtain the region boundary curvature distribution map. Set a preset region boundary curvature threshold κ_th, and mark boundary points with curvature values ​​greater than κ_th ​​as candidate crack initiation points.

[0069] Step S145: Perform stress cycle accumulation processing on each crack initiation candidate point to obtain the stress cycle accumulation parameter corresponding to each crack initiation candidate point, and match the stress cycle accumulation parameter corresponding to each crack initiation candidate point with the preset material fatigue life curve to obtain the fatigue damage degree parameter corresponding to each crack initiation candidate point.

[0070] For each candidate crack initiation point, the number of times the stress field inversion value exceeds the material fatigue limit σ_f during all braking actions is counted, yielding the stress cycle accumulation parameter n. A preset material fatigue life curve is plotted with equivalent stress σ on the x-axis and failure cycle count N_f on the y-axis. For the equivalent stress σ_e at a given point, the corresponding failure cycle count N_f is found on the curve, and the fatigue damage degree parameter D = n / N_f at that point is then calculated.

[0071] Step S146: Based on the fatigue damage degree parameter corresponding to each crack initiation candidate point, associate and store the spatial coordinates of each crack initiation candidate point with the corresponding fatigue damage degree parameter to generate a fatigue damage degree record of the crack initiation candidate point.

[0072] The spatial coordinates (x, y, z) of each crack initiation candidate point are associated and stored with the corresponding fatigue damage degree parameter D to form a record of the fatigue damage degree of the crack initiation candidate point.

[0073] Step S147: Sort all crack initiation candidate points in the fatigue damage degree record according to the fatigue damage degree parameter from largest to smallest, and extract a set number of crack initiation candidate points with the highest fatigue damage degree parameter in the sorting results as key fatigue damage points.

[0074] All crack initiation candidate points are sorted from largest to smallest according to the fatigue damage degree parameter D, and the top 10% of crack initiation candidate points are extracted as key fatigue damage points.

[0075] Step S148: Using the spatial coordinates of the key fatigue damage point as the center, perform fatigue damage degree diffusion processing on the surrounding spatial locations. By distributing the fatigue damage degree parameter of the key fatigue damage point to the surrounding spatial locations according to the spatial distance attenuation weight, fatigue damage degree parameters for each spatial location are generated.

[0076] For each critical fatigue damage point, a three-dimensional Gaussian kernel is constructed centered on its spatial coordinates. The variance σ_g of the Gaussian kernel is determined based on the damage influence range of the material. For any point in three-dimensional space, its fatigue damage degree parameter D_total is equal to the weighted sum of the contributions of all critical fatigue damage points at that point. The contribution value of each critical fatigue damage point is its fatigue damage degree parameter multiplied by the Gaussian kernel function value, which is e^{-(d^2) / (2σ_g^2)}, where d is the spatial distance between that point and the critical fatigue damage point.

[0077] Step S149: Fill the fatigue damage degree parameter of each spatial location point according to the spatial coordinate position to generate a material fatigue location distribution map containing the fatigue damage degree parameter of each spatial location point.

[0078] The fatigue damage parameter D_total of each spatial location point is filled into a three-dimensional array according to its spatial coordinates to generate a material fatigue location distribution map.

[0079] Step S150: Based on the material fatigue location distribution map, perform remaining life prediction processing on the target brake to generate brake life prediction results including minimum remaining effective thickness parameters and remaining life cycle parameters.

[0080] Step S151: Extract the fatigue damage degree parameter corresponding to each spatial location point from the material fatigue location distribution map, and mark the spatial location points whose fatigue damage degree parameter is greater than the preset fatigue damage failure threshold as the failure starting point set.

[0081] Extract the fatigue damage degree parameter D_total for each spatial location point from the material fatigue location distribution map, set a preset fatigue damage failure threshold D_f, and mark the spatial location points where D_total>D_f as failure start points to form a set of failure start points.

[0082] Step S152: Perform spatial connectivity analysis on each failure starting point in the set of failure starting points, aggregate and classify spatially adjacent failure starting points into the same failure connectivity region, and obtain multiple failure connectivity regions.

[0083] A connected component labeling algorithm is used to traverse all failure starting points. For each unlabeled failure starting point, a breadth-first search is used to group all failure starting points that are spatially adjacent to that point (i.e., adjacent in 26 directions) into the same connected region, resulting in multiple failed connected regions.

[0084] Step S153: Perform penetration analysis on the thickness direction of each failed connected region, calculate the maximum penetration depth parameter of each failed connected region in the direction perpendicular to the friction pair surface, obtain the penetration depth parameter of each failed connected region, and perform difference calculation on the penetration depth parameter of each failed connected region and the preset total thickness parameter of the friction pair to obtain the remaining effective thickness parameter corresponding to each failed connected region.

[0085] Determine the normal direction of the friction pair surface. For each spatial location point within the failed connected region, calculate the distance (depth) from that point to the friction pair surface along the normal direction. Take the maximum depth of all points as the maximum penetration depth parameter h_max. The preset total thickness parameter of the friction pair is H_total. Then, the remaining effective thickness parameter h_rem of the failed connected region is h_rem = H_total - h_max.

[0086] Step S154: Perform minimum value filtering on the remaining effective thickness parameter corresponding to each failed connected region, extract the minimum value among all the remaining effective thickness parameters corresponding to the failed connected regions as the minimum remaining effective thickness parameter, and calculate the ratio between the minimum remaining effective thickness parameter and the preset thickness decay rate parameter to obtain the initial remaining lifetime cycle parameter.

[0087] The minimum value among the remaining effective thickness parameters h_rem of all failed connected regions is taken as the minimum remaining effective thickness parameter h_min. The preset thickness decay rate parameter is r, then the initial remaining lifetime period parameter N_initial = h_min / r.

[0088] Step S155: Perform operating condition correction processing on the initial remaining life cycle parameters. Adjust the initial remaining life cycle parameters according to the actual braking frequency parameters and actual braking intensity parameters of the target brake to obtain the corrected remaining life cycle parameters.

[0089] By introducing the actual braking frequency parameter f_actual and the actual braking intensity parameter s_actual, and setting the reference braking frequency parameter f_base and the reference braking intensity parameter s_base, the corrected remaining lifespan parameter N_corrected = N_initial. (f_base / f_actual) (s_base / s_actual).

[0090] Step S156: Perform confidence interval estimation processing on the corrected remaining life cycle parameters, calculate the upper and lower fluctuation range of the predicted cycle based on the distribution dispersion of the fatigue damage degree parameters in the material fatigue location distribution map, and obtain the confidence interval parameters of the remaining life cycle.

[0091] Calculate the standard deviation σ_D of the fatigue damage degree parameter in all connected regions of failure in the material fatigue location distribution map. Based on a preset 95% confidence level, determine the confidence factor k=1.96. Then, the upper limit of the confidence interval for the remaining life cycle parameter is N_upper=N_corrected. (1+k √σ_D), the lower limit parameter of the confidence interval N_lower=N_corrected (1-k √σ_D).

[0092] Step S157: Associate and store the corrected remaining lifetime cycle parameter with the confidence interval parameter of the remaining lifetime cycle to generate an intermediate remaining lifetime prediction result containing the remaining lifetime cycle parameter and its confidence interval parameter.

[0093] The corrected remaining lifetime period parameter N_corrected is associated with the upper limit of the confidence interval parameter N_upper and the lower limit of the confidence interval parameter N_lower and stored to form the intermediate result of the remaining lifetime prediction.

[0094] Step S158: The intermediate result of the remaining life prediction is fused with the minimum remaining effective thickness parameter to generate a brake life prediction result containing the minimum remaining effective thickness parameter and the remaining life cycle parameter.

[0095] The intermediate results of the remaining life prediction are fused with the minimum remaining effective thickness parameter h_min to generate a brake life prediction result that includes the minimum remaining effective thickness parameter, the remaining life cycle parameter and their confidence interval parameters.

[0096] Step S210: Obtain the real-time surface vibration signal sequence generated by the target brake under the current braking action, perform vibration signal spectral feature extraction processing on the real-time surface vibration signal sequence to obtain a real-time vibration signal spectral feature map, and perform spectral energy distribution difference calculation processing on the real-time vibration signal spectral feature map and the preset healthy state vibration spectral reference map to obtain a real-time vibration signal spectral difference distribution map.

[0097] An acceleration sensor is installed on the target brake to collect vibration signals from the brake disc surface during the current braking action, forming a real-time surface vibration signal sequence. A fast Fourier transform is performed on this sequence to obtain a real-time vibration signal spectrum feature map, with frequency f as the abscissa and energy amplitude A(f) as the ordinate. A preset healthy state vibration spectrum baseline map is A_base(f). The energy amplitude difference ΔA(f) = A(f) - A_base(f) at each frequency point is calculated to obtain the real-time vibration signal spectrum difference distribution map.

[0098] Step S211: Identify frequency ranges where the spectral energy distribution difference exceeds a preset spectral energy distribution difference threshold based on the real-time vibration signal spectral difference distribution map as abnormal vibration frequency ranges, and extract the center frequency parameter corresponding to the abnormal vibration frequency range.

[0099] A preset threshold for the difference in spectral energy distribution, ΔA_th, is set. Continuous intervals containing frequency points where |ΔA(f)|>ΔA_th are marked as abnormal vibration frequency intervals. For each abnormal vibration frequency interval, the center value f_c of its frequency range is calculated and used as the center frequency parameter.

[0100] Step S212: Map the center frequency parameter to the surface wear morphology parameter space to obtain the predicted surface wear morphology parameter set. Perform correlation verification processing on the predicted surface wear morphology parameter set and the remaining effective thickness parameter in the brake life prediction result. Normalize each parameter in the predicted surface wear morphology parameter set and the remaining effective thickness parameter respectively. Calculate the correlation parameter between the two based on the normalized parameters.

[0101] Through the pre-established backpropagation neural network mapping model, the center frequency parameter \(f_c\) is input into the model, and the predicted set of surface wear morphology parameters is output, including the surface roughness parameter \(Ra\), the waviness parameter \(W\), and the wear depth parameter \(h_{wear}\). Perform min-max normalization processing on each of the above parameters and the remaining effective thickness parameter \(h_{min}\) respectively to obtain the normalized values \(Ra_{norm}\), \(W_{norm}\), \(h_{wear norm}\), and \(h_{min norm}\). Calculate the Pearson correlation coefficient \(r_p\) between the normalized surface wear morphology parameters and the normalized remaining effective thickness parameter as the correlation degree parameter.

[0102] Step S213: When the correlation degree parameter is less than the preset correlation degree threshold, trigger the operation of re-executing the stress field inversion process by inputting the surface topography reconstruction model into the preset stress field evolution network, and generate an updated internal stress field inversion map.

[0103] Set the preset correlation degree threshold \(r_{th}\). If \(r_p < r_{th}\), then trigger the re-execution of step S130 and its sub-steps to generate an updated internal stress field inversion map.

[0104] Step S214: When the correlation degree parameter is greater than or equal to the preset correlation degree threshold, mark the brake life prediction result as the verified brake life prediction result.

[0105] If \(r_p \geq r_{th}\), then mark the current brake life prediction result as the verified brake life prediction result.

[0106] Step S215: Perform a maintenance level classification process on the remaining effective thickness parameter in the verified brake life prediction result to obtain the brake maintenance level identifier corresponding to the remaining effective thickness parameter.

[0107] Compare the minimum remaining effective thickness parameter \(h_{min}\) in the verified brake life prediction result with multiple preset maintenance level thresholds. Set the first maintenance level threshold \(H1\), the second maintenance level threshold \(H2\), and the third maintenance level threshold \(H3\). If \(h_{min}>H1\), it corresponds to the low maintenance level identifier; if \(H2 < h_{min} \leq H1\), it corresponds to the medium maintenance level identifier; if \(H3 < h_{min} \leq H2\), it corresponds to the high maintenance level identifier; if \(h_{min} \leq H3\), it corresponds to the emergency maintenance level identifier.

[0108] Step S216: Call the preset maintenance strategy template library according to the brake maintenance level identifier, and extract the maintenance operation instruction sequence associated with the brake maintenance level identifier from the preset maintenance strategy template library.

[0109] The preset maintenance strategy template library stores the maintenance operation instruction sequence corresponding to different maintenance levels. Based on the brake maintenance level identifier obtained in step S215, the corresponding maintenance operation instruction sequence is extracted.

[0110] Step S217: Encapsulate the maintenance operation instruction sequence with the verified brake life prediction result to generate a brake maintenance decision instruction containing the maintenance operation instruction sequence and the verified brake life prediction result.

[0111] The maintenance operation instruction sequence and the verified brake life prediction results are encapsulated into brake maintenance decision instructions.

[0112] Step S310: Obtain the material composition parameters and initial surface texture parameters of the target brake, input the material composition parameters and the initial surface texture parameters into a preset microstructure initialization network, and perform grain structure distribution simulation processing on the material composition parameters through the microstructure initialization network to generate an initial distribution map of the material grain structure.

[0113] The material composition parameters of the target brake are obtained, including the percentage of alloying elements, statistical distribution parameters of grain size, and initial surface texture parameters, including surface roughness parameters and processing texture direction parameters. These parameters are input into a preset microstructure initialization network. This network, based on a phase-field model, determines the nucleation rate and growth rate of grains according to the alloying element content in the material composition parameters, simulates the grain nucleation and growth process in a three-dimensional spatial domain, and generates an initial distribution map of the material's grain structure.

[0114] Step S311: The surface initial texture parameters are processed by the microstructure initialization network to generate a surface texture morphology map.

[0115] The microstructure initialization network generates a Gaussian random surface with a specified autocorrelation function based on the initial surface texture parameters using a filtering method, which serves as the initial surface texture topography map.

[0116] Step S312: Spatially superimpose the initial distribution map of the material grain structure with the initial surface texture map to generate an initial microstructure model containing grain structure distribution information and surface texture information.

[0117] The initial distribution map of the material's grain structure and the initial texture map of the surface are spatially superimposed. The initial texture map of the surface is used as the upper surface boundary condition of the three-dimensional model to generate an initial microstructure model containing information on grain structure distribution and surface texture.

[0118] Step S313: Obtain the set of real-time braking pressure change curves and the set of real-time braking temperature change curves generated when the target brake is continuously tested on a standard braking condition test bench. Perform pressure waveform feature extraction processing on the set of real-time braking pressure change curves to obtain a pressure waveform feature parameter sequence.

[0119] Continuous braking tests were conducted on a standard braking condition test bench, and the real-time braking pressure change curves during each braking process were recorded to form a set of real-time braking pressure change curves. For each curve, the peak pressure P_peak, pressure rise time t_rise, pressure hold time t_hold, pressure fall time t_fall, and pressure fluctuation amplitude ΔP were extracted to form a sequence of pressure waveform characteristic parameters.

[0120] Step S314: Perform temperature gradient feature extraction processing on the set of real-time braking temperature change curves to obtain a temperature gradient feature parameter sequence.

[0121] Simultaneously, the real-time braking temperature change curves during each braking process are recorded to form a set of real-time braking temperature change curves. For each curve, the maximum temperature T_max, temperature rise rate v_rise, temperature fall rate v_fall, and temperature fluctuation amplitude ΔT are extracted to form a temperature gradient feature parameter sequence.

[0122] Step S315: Normalize the pressure waveform feature parameter sequence and the temperature gradient feature parameter sequence respectively. Perform time-series correlation modeling on the normalized pressure waveform feature parameter sequence and temperature gradient feature parameter sequence to generate a working condition coupling evolution feature vector that reflects the coupling evolution relationship between braking pressure and braking temperature. Map the working condition coupling evolution feature vector to each spatial location point of the initial microstructure model to generate the working condition loading parameters corresponding to each spatial location point.

[0123] The pressure waveform characteristic parameter sequence and temperature gradient characteristic parameter sequence are normalized and input into a long short-term memory network, outputting a working condition coupling evolution feature vector. This vector is mapped to each spatial location point of the initial microstructure model and converted into working condition loading parameters at each spatial location point, including equivalent pressure P_eq and equivalent temperature T_eq, through a spatial attenuation function f(d)=e^{-αd}, where d is the distance from the spatial location point to the friction pair surface and α is the attenuation coefficient.

[0124] Step S316: Perform stress-strain response calculation on the material grain structure in the initial microstructure model according to the working condition loading parameters corresponding to each spatial location point to obtain the grain structure deformation parameters corresponding to each spatial location point. Superimpose the grain structure deformation parameters corresponding to each spatial location point onto the corresponding spatial location point of the initial microstructure model to generate an initial service microstructure model containing grain structure deformation information.

[0125] Using the crystal plastic finite element method, for each spatial location point, based on its equivalent pressure P_eq and equivalent temperature T_eq, considering the crystal orientation, slip system, and hardening law of the grains, the equilibrium equation and constitutive equation are solved to obtain the grain structure deformation parameters, including the grain rotation angle θ_g, dislocation density ρ_d, and plastic strain tensor ε_p. These parameters are then superimposed onto the corresponding spatial location point of the initial microstructure model to generate the initial microstructure model for service.

[0126] Step S317: Store the initial microstructure model of the service as the initial input parameter of the preset stress field evolution network for subsequent execution of the operation of inputting the surface morphology reconstruction model into the preset stress field evolution network for stress field inversion processing.

[0127] The initial microstructure model is stored as the initial input parameter for the preset stress field evolution network.

[0128] Step S410: Obtain the historical braking failure record data set generated by the target brake during its historical service cycle. The historical braking failure record data set includes parameters of the number of braking events before failure and the remaining thickness at the time of failure for each braking failure event.

[0129] Collect brake failure records of the target brake during its historical service life to form a historical brake failure record data set. Each record contains the parameter N_fail, which represents the number of braking attempts before failure, and the parameter h_fail, which represents the remaining thickness at the time of failure.

[0130] Step S411: Extract the correspondence between the number of braking events before failure and the remaining thickness parameter at the time of failure from the historical braking failure record data set, and construct an association mapping table between the number of braking events before failure and the remaining thickness parameter at the time of failure.

[0131] Extract all (N_fail, h_fail) pairs from the historical brake failure record dataset, and perform curve fitting using the least squares method to obtain the fitting function h_fail=a. N_fail^b+c, construct the association mapping table.

[0132] Step S412: Input the remaining effective thickness parameter in the brake life prediction result into the association mapping table for matching query processing to obtain the predicted remaining braking number parameter corresponding to the remaining effective thickness parameter.

[0133] The minimum remaining effective thickness parameter h_min from the brake life prediction results is input into the association mapping table, and the corresponding predicted remaining braking number parameter N_pred is calculated through the fitting function.

[0134] Step S413: Based on the historical average braking frequency of the target brake, convert the remaining life cycle parameter into an equivalent remaining braking number parameter, perform consistency analysis on the predicted remaining braking number parameter and the equivalent remaining braking number parameter, and calculate the conversion deviation parameter between the predicted remaining braking number parameter and the equivalent remaining braking number parameter.

[0135] Based on the historical average braking frequency f_avg of the target brake, the remaining life cycle parameter N_corrected is converted into the equivalent remaining braking count parameter N_eq=N_corrected. f_avg. Calculate the conversion bias parameter δ = |N_pred - N_eq| / N_pred.

[0136] Step S414: When the conversion deviation parameter is greater than the preset conversion deviation threshold, a correction process is triggered on the remaining life cycle parameter in the brake life prediction result to generate a corrected brake life prediction result; when the conversion deviation parameter is less than or equal to the preset conversion deviation threshold, the brake life prediction result is marked as a verified brake life prediction result.

[0137] A preset conversion deviation threshold δ_th is set. If δ > δ_th, then the remaining lifespan parameters are corrected. The correction method is N_corrected_new = N_corrected (N_pred / N_eq) generates the corrected brake life prediction result; if δ≤δ_th, the brake life prediction result is marked as a verified brake life prediction result.

[0138] Step S415: Perform three-dimensional color mapping processing on the material fatigue location distribution map to generate a three-dimensional color distribution map of material fatigue degree, and perform three-dimensional model sectioning and display processing on the remaining effective thickness parameter in the verified brake life prediction result to generate a three-dimensional section model of the friction pair containing the remaining effective thickness indicator.

[0139] The fatigue damage degree parameters in the material fatigue location distribution map are mapped to color values ​​to generate a three-dimensional color distribution map of material fatigue degree. The minimum remaining effective thickness parameter h_min in the verified brake life prediction results is sectioned and displayed in the three-dimensional friction pair model, and the location of the remaining effective thickness is marked on the section surface to generate a three-dimensional sectioned model of the friction pair.

[0140] Step S416: Spatial overlay display processing is performed on the three-dimensional color distribution map of the material fatigue degree and the three-dimensional cross-sectional model of the friction pair to generate a life prediction visualization result containing fatigue degree distribution information and remaining effective thickness information, and the life prediction visualization result is output to the display terminal for display.

[0141] The three-dimensional color distribution map of material fatigue degree is spatially superimposed with the three-dimensional cross-sectional model of the friction pair to generate a life prediction visualization result, which is then output to the display terminal.

[0142] For example, the method may further include: step S510: obtaining the surface temperature field distribution sequence of the target brake during the continuous braking process within a preset time window before the current moment, and performing temperature field spatiotemporal evolution feature extraction processing on the surface temperature field distribution sequence of the continuous braking process to obtain the temperature field spatiotemporal evolution feature vector.

[0143] During the service life of the brake, surface temperature field distribution data are continuously collected within a preset time window (i.e., during each of the past M braking actions) using a thermocouple array or infrared thermal imager arranged on the surface of the brake disc. The temperature field distribution data collected for each braking action is a two-dimensional matrix, where each element corresponds to the temperature value at a spatial location on the brake disc surface. The temperature field distribution data for all braking actions are arranged in chronological order, forming a continuous braking process surface temperature field distribution sequence. This sequence is then processed to extract the spatiotemporal evolution features of the temperature field. A three-dimensional convolutional neural network (3D Convolutional Neural Network) is used to process the sequence data. This 3D Convolutional Neural Network contains multiple 3D convolutional layers and pooling layers. The 3D convolutional kernels perform convolution operations simultaneously in both spatial and temporal dimensions to extract the spatial distribution features and temporal evolution features of the temperature field. The input data dimension is M multiplied by H multiplied by W, where M is the number of braking actions within the time window, and H and W are the height and width of the temperature field matrix, respectively. After passing through multiple three-dimensional convolutional and pooling layers, a fixed-length spatiotemporal evolution feature vector of the temperature field is output. This feature vector comprehensively reflects the evolution of the surface temperature field in both time and space during the braking process.

[0144] Step S511: Perform thermal stress cycle accumulation calculation on the spatiotemporal evolution feature vector of the temperature field to obtain the distribution map of the cumulative degree of surface thermal fatigue damage.

[0145] The spatiotemporal evolution feature vector of the temperature field is input into a pre-defined thermal stress calculation model. This model, based on thermoelasticity theory, calculates the thermal stress value at each spatial location point at different times based on the temperature distribution and temperature change rate information contained in the spatiotemporal evolution feature vector. For each spatial location point, the number of times its thermal stress value exceeds the material's thermal fatigue limit is counted, obtaining the cumulative thermal stress cycle count parameter for that point. Based on the cumulative thermal stress cycle count parameter for each spatial location point and the material's thermal fatigue life curve, the thermal fatigue damage degree parameter for each spatial location point is calculated. The thermal fatigue damage degree parameter is equal to the cumulative thermal stress cycle count parameter divided by the failure cycle count under the corresponding thermal stress amplitude. The thermal fatigue damage degree parameters of all spatial location points are filled according to their spatial coordinates to generate a surface thermal fatigue damage cumulative degree distribution map.

[0146] Step S512: The surface thermal fatigue damage accumulation distribution map and the material fatigue location distribution map are spatially superimposed to generate a comprehensive fatigue damage distribution map that includes both mechanical fatigue damage and thermal fatigue damage.

[0147] The surface thermal fatigue damage accumulation distribution map and the material fatigue location distribution map share the same spatial coordinate system, both constructed based on the three-dimensional spatial coordinate system of the brake disc. The fatigue damage degree parameters at corresponding spatial locations in the two distribution maps are weighted and summed to obtain a comprehensive fatigue damage degree parameter. During the weighted summation, the weighting coefficients for the mechanical fatigue damage degree parameter and the thermal fatigue damage degree parameter are determined according to the contribution ratio of mechanical load and thermal load during braking conditions. After traversing all spatial locations, a comprehensive fatigue damage degree distribution map is generated. Each spatial location in this map corresponds to a comprehensive fatigue damage degree parameter, comprehensively reflecting the coupling effect of mechanical fatigue and thermal fatigue.

[0148] Step S513: Based on the comprehensive fatigue damage degree distribution map, identify the spatial location points where the comprehensive fatigue damage degree exceeds the preset comprehensive fatigue damage threshold as the comprehensive failure starting point set, and perform spatial connectivity analysis on the comprehensive failure starting point set to obtain multiple comprehensive failure connectivity regions.

[0149] A preset comprehensive fatigue damage threshold D_comb_th is set. All spatial locations in the comprehensive fatigue damage distribution map are traversed, and spatial locations with a comprehensive fatigue damage parameter greater than D_comb_th are marked as comprehensive failure initiation points, forming a set of comprehensive failure initiation points. This set is then processed using a connected component labeling algorithm to perform spatial connectivity analysis, grouping all spatially adjacent (i.e., 26-directionally adjacent) comprehensive failure initiation points into the same connected region, resulting in multiple comprehensive failure connected regions.

[0150] Step S514: Perform penetration analysis in the region thickness direction on each of the multiple integrated failure connected regions, calculate the maximum penetration depth parameter of each integrated failure connected region in the direction perpendicular to the friction pair surface, and obtain the integrated penetration depth parameter of each integrated failure connected region.

[0151] For each integrated failure connected region, determine the normal direction of the friction pair surface. Traverse all spatial points within this region and calculate the distance from each point along the normal direction to the friction pair surface, i.e., the depth of that point. Take the maximum depth among all points as the integrated penetration depth parameter h_comb_max for this integrated failure connected region.

[0152] Step S515: Perform difference calculation processing based on the comprehensive penetration depth parameter of each comprehensive failure connected region and the preset total thickness parameter of the friction pair to obtain the comprehensive remaining effective thickness parameter corresponding to each comprehensive failure connected region.

[0153] The preset total thickness parameter of the friction pair is H_total. For each integrated failure connected region, its integrated remaining effective thickness parameter h_comb_rem is equal to H_total minus the integrated penetration depth parameter h_comb_max of that region, that is, h_comb_rem=H_total-h_comb_max.

[0154] Step S516: Extract the minimum value among the comprehensive remaining effective thickness parameters corresponding to all comprehensive failure connected regions as the comprehensive minimum remaining effective thickness parameter, compare the comprehensive minimum remaining effective thickness parameter with the remaining effective thickness parameter in the brake life prediction result, and calculate the difference parameter between the comprehensive minimum remaining effective thickness parameter and the remaining effective thickness parameter.

[0155] The minimum value among the comprehensive remaining effective thickness parameters h_comb_rem for all integrated failure connected regions is taken as the comprehensive minimum remaining effective thickness parameter h_comb_min. The minimum remaining effective thickness parameter h_min is extracted from the brake life prediction results, and the difference parameter Δh = |h_comb_min - h_min| is calculated between the two.

[0156] Step S517: When the difference parameter is greater than the preset difference threshold, the comprehensive minimum remaining effective thickness parameter is used as the updated remaining effective thickness parameter to replace the remaining effective thickness parameter in the brake life prediction result, an updated brake life prediction result is generated, and the updated brake life prediction result is output to the brake life prediction result storage database for storage.

[0157] A preset difference threshold Δh_th is set. If Δh > Δh_th, it indicates a significant difference between the remaining effective thickness parameter obtained by considering mechanical fatigue alone and the comprehensive remaining effective thickness parameter after considering thermal fatigue coupling, requiring an update to the prediction result. The minimum remaining effective thickness parameter h_min in the brake life prediction result is replaced with the comprehensive minimum remaining effective thickness parameter h_comb_min, and steps S154 to S158 are re-executed. Based on the updated remaining effective thickness parameter, the remaining life cycle parameter is recalculated to generate an updated brake life prediction result. This updated brake life prediction result is output to the brake life prediction result storage database for storage, for subsequent maintenance decisions and life tracking.

[0158] For example, the method may further include: step S610: acquiring the time-series signal of the electrochemical corrosion potential on the friction pair surface generated by the target brake during continuous braking, and performing time-frequency domain joint feature decomposition processing on the electrochemical corrosion potential time-series signal to obtain corrosion potential fluctuation feature components and corrosion potential trend feature components.

[0159] During the service life of the brake, electrochemical corrosion potential signals are continuously acquired by electrochemical sensors arranged on the surface of the friction pair. These signals are time-series signals that vary over time, denoted as E(t). The time-frequency domain joint eigenvalue decomposition process is performed on this electrochemical corrosion potential signal, and an empirical mode decomposition algorithm is used to decompose the signal into multiple intrinsic mode function (EMF) components. The EMF algorithm iteratively extracts EMF components from the original signal in descending order of frequency through an iterative selection process. Several EMF components from the high-frequency components are reconstructed to obtain corrosion potential fluctuation characteristic components, which reflect the instantaneous fluctuations of the corrosion potential and are related to surface fretting wear and localized corrosion processes. Several EMF components from the low-frequency components are reconstructed to obtain corrosion potential trend characteristic components, which reflect the long-term trend of the corrosion potential and are related to the growth and destruction of the corrosion product film.

[0160] Step S611: Perform waveform pattern matching processing on the corrosion potential fluctuation characteristic components to identify the potential fluctuation segment dominated by fretting wear on the friction pair surface and the potential fluctuation segment dominated by electrochemical corrosion.

[0161] A sliding window method was used to segment the corrosion potential fluctuation characteristic components into multiple time segments, with the length of each time segment set to the duration of a typical braking action. For each time segment, waveform feature parameters were extracted, including peak amplitude, trough amplitude, waveform rise time, waveform fall time, waveform frequency, and waveform energy. These waveform feature parameters were input into a pre-set classifier, which employed a support vector machine algorithm and was pre-trained using a large amount of labeled sample data dominated by fretting wear and electrochemical corrosion. Based on the input waveform feature parameters, the classifier output a classification result indicating whether the time segment belonged to a potential fluctuation segment dominated by fretting wear or electrochemical corrosion, thereby identifying the two different types of potential fluctuation segments.

[0162] Step S612: Based on the potential fluctuation segment dominated by fretting wear and the potential fluctuation segment dominated by electrochemical corrosion, extract the braking condition parameter sequence and surface morphology evolution parameter sequence within the corresponding time interval, and construct a time-series correlation mapping diagram between fretting wear and electrochemical corrosion.

[0163] For each identified potential fluctuation segment dominated by fretting wear, braking condition parameters within the corresponding time interval are extracted, including braking pressure, braking temperature, and braking speed, forming a fretting wear condition parameter sequence. For each identified potential fluctuation segment dominated by electrochemical corrosion, surface morphology evolution parameters within the corresponding time interval are extracted, including changes in surface roughness and wear depth, forming a corrosion morphology evolution parameter sequence. A time-series correlation analysis is performed between the fretting wear condition parameter sequence and the corrosion morphology evolution parameter sequence in subsequent time intervals to calculate the time-delay cross-correlation function between them, determining the lag time of the impact of fretting wear on subsequent electrochemical corrosion. Based on the analysis results, a time-series correlation mapping graph is constructed. This graph is a directed graph structure, where nodes represent fretting wear events and electrochemical corrosion events, and edges represent the temporal causal relationship and influence weight between them.

[0164] Step S613: Perform film thickness inversion calculation on the corrosion potential trend characteristic components to obtain the spatiotemporal distribution thickness map of the corrosion product film on the friction pair surface.

[0165] A definite electrochemical relationship exists between the corrosion potential trend characteristic component and the thickness of the corrosion product film. According to the Nernst equation, the corrosion potential is linearly related to the logarithm of the film thickness. A film thickness inversion model is established, which takes the corrosion potential trend characteristic component as input and outputs the corresponding film thickness value. Since the corrosion potential trend characteristic component changes with time, and the corrosion potential may differ at different spatial locations, it is necessary to combine the spatial location information of multiple electrochemical sensors and spatially interpolate the film thickness inversion results at each sensor location to generate a spatiotemporal distribution map of the corrosion product film thickness on the entire friction pair surface. This film thickness inversion map is three-dimensional spatiotemporal data, containing the film thickness value at each spatial location at different times.

[0166] Step S614: Perform spatial coordinate alignment processing on the spatiotemporal distribution thickness map of the corrosion product film layer on the surface of the friction pair and the material fatigue positioning distribution map to generate a spatial superimposed distribution map of the thickness of the corrosion product film layer and the degree of fatigue damage on the surface of the friction pair.

[0167] The spatiotemporal distribution thickness map of the corrosion product film on the friction pair surface and the material fatigue location distribution map are both based on the three-dimensional spatial coordinate system of the brake disc. The two maps are aligned to ensure that the same spatial location point corresponds in both maps. For each spatial location point, the corrosion product film thickness and fatigue damage degree parameters are extracted and used as a set of feature values. The feature values ​​of all spatial location points are visualized to generate a spatially overlaid distribution map. This map, based on spatial coordinates, simultaneously displays the corrosion product film thickness and fatigue damage degree at each location point, typically using a two-color mapping or layered rendering method.

[0168] Step S615: Identify spatially overlapping regions where the thickness of the corrosion product film exceeds a preset film thickness threshold and the degree of fatigue damage exceeds a preset fatigue damage threshold based on the spatial overlay distribution map, and mark the spatially overlapping regions as corrosion fatigue coupled damage sensitive areas.

[0169] Set preset film thickness thresholds h_film_th and D_fatigue_th. Traverse all spatial locations in the spatial overlay distribution map, marking those locations that simultaneously satisfy the conditions that the corrosion product film thickness is greater than h_film_th and the fatigue damage parameter is greater than D_fatigue_th. Perform spatial connectivity analysis on these marked points, grouping spatially adjacent marked points into the same connected region, resulting in multiple spatially overlapping regions. These regions are marked as corrosion-fatigue coupled damage sensitive areas.

[0170] Step S616: Perform corrosion fatigue coupling damage acceleration factor calculation for each spatial location point within the corrosion fatigue coupling damage sensitive area. Based on the corrosion product film thickness parameter and fatigue damage degree parameter of each spatial location point, query the preset corrosion fatigue coupling damage acceleration factor mapping table to obtain the coupling damage acceleration factor corresponding to each spatial location point.

[0171] The pre-defined corrosion-fatigue coupled damage acceleration factor mapping table was established using extensive corrosion-fatigue coupled test data. This table takes corrosion product film thickness and fatigue damage severity as input dimensions and the coupled damage acceleration factor as output. For each spatial location within the corrosion-fatigue coupled damage sensitive area, bilinear interpolation is performed in the mapping table based on the corrosion product film thickness h_film and the fatigue damage severity parameter D_fatigue to obtain the corresponding coupled damage acceleration factor α_coupling. This coupled damage acceleration factor reflects the acceleration factor of the fatigue damage accumulation rate compared to the individual mechanical fatigue condition due to the coupling effect of corrosion and fatigue.

[0172] Step S617: Accelerate the correction process of the fatigue damage degree parameters in the fatigue location distribution map of the material to generate a corrected fatigue damage degree distribution map that includes the corrosion fatigue coupling effect.

[0173] For each spatial location point in the material fatigue location distribution map, determine whether it belongs to the corrosion fatigue coupled damage sensitive area. If it does, multiply the fatigue damage degree parameter D_original at that point by the corresponding coupled damage acceleration factor α_coupling to obtain the corrected fatigue damage degree parameter D_corrected=D_original. α_coupling. If it does not belong to the range, the original fatigue damage degree parameter remains unchanged. After traversing all spatial location points, a modified fatigue damage degree distribution map is generated, which reflects the cumulative fatigue damage after considering the corrosion-fatigue coupling effect.

[0174] Step S618: Based on the modified fatigue damage distribution map, re-execute the remaining life prediction process for the target brake to generate a modified brake life prediction result that includes the effects of corrosion fatigue coupling.

[0175] Using the corrected fatigue damage distribution map as input, step S150 and its related sub-steps are re-executed. Specifically, the corrected fatigue damage degree parameters are extracted from the corrected fatigue damage degree distribution map. Spatial locations where the corrected fatigue damage degree parameters are greater than the preset fatigue damage failure threshold are marked as the set of corrected failure starting points. Spatial connectivity analysis is performed to obtain the corrected failure connected region. The corrected remaining effective thickness parameters are calculated, and the remaining life cycle parameters are recalculated based on the corrected remaining effective thickness parameters to generate the corrected brake life prediction result.

[0176] Step S619: Perform a difference comparison analysis between the corrected brake life prediction result and the brake life prediction result, calculate the remaining life prediction cycle deviation rate between the two, and output the remaining life prediction cycle deviation rate as the brake life prediction confidence evaluation index.

[0177] The remaining life cycle parameter N_original is extracted from the brake life prediction results, and the remaining life cycle parameter N_corrected is extracted from the corrected brake life prediction results. The remaining life prediction cycle deviation rate ε = |N_corrected - N_original| / N_original is calculated. This remaining life prediction cycle deviation rate reflects the degree of influence of the corrosion-fatigue coupling effect on the life prediction results. The larger the deviation rate, the more significant the corrosion-fatigue coupling effect, and the lower the confidence of the prediction results considering only mechanical fatigue. This deviation rate is used as the brake life prediction confidence evaluation index to guide subsequent maintenance decisions. When the deviation rate exceeds a preset threshold, the corrected prediction results that consider the corrosion-fatigue coupling effect should be used first.

[0178] For example, the method may further include: step S710: obtaining the braking action execution time sequence and braking intensity parameter corresponding to each braking action recorded by the target brake during its historical service process, performing time interval statistical analysis on the braking action execution time sequence to obtain braking action time interval distribution characteristic parameters.

[0179] Extract the braking action execution time sequence from the historical brake operation data, denoted as t1, t2, t3, ..., tK, where K is the total number of historical braking actions. Calculate the time interval between adjacent braking actions Δt_j = t_{j+1} - t_j, obtaining the time interval sequence. Perform statistical analysis on this time interval sequence, calculating the mean μ_Δt, standard deviation σ_Δt, skewness coefficient γ_Δt, kurtosis coefficient κ_Δt, and the probability distribution function of the time intervals. These statistics constitute the characteristic parameters of the braking action time interval distribution.

[0180] Step S711: Cluster the spatiotemporal distribution characteristics of braking actions to identify multiple typical braking behavior patterns of the target brake and their temporal patterns.

[0181] For each braking action, its feature vector is extracted, including the braking intensity parameter s, braking duration parameter d, vehicle speed parameter v before braking, vehicle speed change Δv after braking, and the time interval Δt between the braking actions. The feature vectors of all braking actions are compiled into a dataset, and a density-based clustering algorithm is used for clustering, grouping braking actions with similar features into the same class, with each class representing a typical braking behavior pattern. For each identified braking behavior pattern, its temporal occurrence pattern is statistically analyzed, including the frequency of occurrence of the pattern at different times of day, the frequency of occurrence on different days of week, and the transition probability in a continuous braking sequence, i.e., the probability of transitioning from one pattern to another.

[0182] Step S712: Based on the various typical braking behavior modes and their timing patterns, construct a braking behavior mode damage weight allocation matrix that reflects the mapping relationship between braking behavior modes and the cumulative rate of fatigue damage of friction pairs.

[0183] For each typical braking behavior mode, the corresponding friction pair fatigue damage accumulation rate parameter is determined through finite element simulation or bench testing. This parameter represents the increment of fatigue damage caused by the mode per unit number of braking cycles. This damage accumulation rate parameter is used as the basic damage weight for that mode. Simultaneously, considering the influence of temporal combinations between different modes on damage accumulation, a coupling damage correction coefficient β_{ij} is introduced for the case where mode i is immediately followed by mode j. This coefficient is determined based on the temporal interaction between the two modes. Therefore, a braking behavior mode damage weight allocation matrix W is constructed, where rows correspond to preceding braking behavior modes and columns correspond to subsequent braking behavior modes. The matrix element W_{ij} = w_i. β_{ij}, where w_i is the basic damage weight of pattern i.

[0184] Step S713: Perform braking behavior mode weighting on the fatigue damage degree parameters at different spatial locations in the material fatigue location distribution map to generate a dynamic fatigue damage cumulative distribution map that considers the temporal evolution of braking behavior modes.

[0185] The material fatigue location distribution map records the static fatigue damage degree parameters at each spatial location point. However, these parameters are calculated based on historical average working conditions and do not consider the differentiated contributions of different braking behavior modes to damage accumulation. Based on the actual braking behavior mode sequence, the fatigue damage degree parameters at each spatial location point are dynamically weighted and corrected. Specifically, for each braking action, according to its corresponding braking behavior mode, the damage weight factor corresponding to that mode is obtained from the braking behavior mode damage weight allocation matrix. This weight factor is multiplied by the instantaneous damage increment generated by the braking action at that spatial location point, and then added to the cumulative damage at that point. After traversing all historical braking actions, the dynamic fatigue damage degree parameters at each spatial location point are obtained, generating a dynamic fatigue damage accumulation distribution map.

[0186] Step S714: Extract spatial location points from the dynamic fatigue damage cumulative distribution map where the growth rate of the fatigue damage degree parameter exceeds a preset growth rate threshold as fatigue damage hotspot areas, and obtain the dominant braking behavior mode identifier corresponding to the fatigue damage hotspot areas.

[0187] The growth rate of fatigue damage severity parameters at each spatial location point in the dynamic fatigue damage accumulation distribution map with the number of braking cycles is calculated, i.e., the damage increment per unit braking cycle. A preset growth rate threshold v_th is set, and spatial locations with growth rates greater than v_th are marked. These marked points are then spatially clustered to obtain multiple fatigue damage hotspot regions. For each fatigue damage hotspot region, the main contributors to damage accumulation are analyzed, i.e., the main braking behavior patterns leading to damage accumulation in that region are identified. By statistically analyzing the proportion of damage increment generated by each braking behavior pattern within the region, the braking behavior pattern with the largest proportion is determined as the dominant braking behavior pattern, and its identifier is obtained.

[0188] Step S715: Based on the dominant braking behavior mode identifier corresponding to the fatigue damage hotspot area, call the preset braking behavior mode optimization strategy library, and extract the braking parameter adjustment suggestion instruction sequence associated with the dominant braking behavior mode identifier from the braking behavior mode optimization strategy library.

[0189] The pre-defined braking behavior mode optimization strategy library stores optimization strategies for different braking behavior modes. For example, for high-frequency, low-intensity braking modes, the optimization strategy is to appropriately reduce the braking pressure response sensitivity to reduce the number of braking operations; for high-intensity emergency braking modes, the optimization strategy is to provide early warning and optimize the braking pressure rise slope to reduce peak pressure; for long-term continuous braking modes, the optimization strategy is to increase the braking interval time and optimize the heat dissipation strategy. Based on the dominant braking behavior mode identifier corresponding to the fatigue damage hotspot area, the corresponding braking parameter adjustment suggestion instruction sequence is extracted from the strategy library. This braking parameter adjustment suggestion instruction sequence contains a series of specific braking parameter adjustment instructions, such as target braking pressure value, braking pressure rise time, braking interval time, etc.

[0190] Step S716: Convert the braking parameter adjustment suggestion instruction sequence into a braking parameter adjustment control signal executable by the brake electronic control unit, and send it to the electronic control unit of the target brake to perform online adaptive adjustment of braking parameters, thereby changing the braking intensity parameter and braking time interval parameter of subsequent braking actions.

[0191] The commands in the braking parameter adjustment suggestion sequence are converted into control signal formats conforming to the brake electronic control unit (ECU) communication protocol and sent to the ECU via the vehicle bus. Upon receiving the control signals, the ECU adjusts the braking control parameters online, including modifying the target pressure value of the braking pressure control algorithm, adjusting the braking pressure rise rate limit, and modifying the braking interval threshold. This alters the braking intensity and braking time interval parameters of subsequent braking actions, shifting the subsequent braking behavior pattern in a direction more conducive to mitigating fatigue damage.

[0192] Step S717: Obtain the updated friction pair surface vibration signal sequence generated by the target brake within a preset monitoring time window after the online adaptive adjustment of the braking parameters is performed, and perform fatigue damage state verification processing on the updated friction pair surface vibration signal sequence to generate fatigue damage mitigation effect evaluation parameters after the braking parameters are adjusted.

[0193] After performing online adaptive adjustment of braking parameters, vibration signals of the friction pair surface are continuously collected within a preset monitoring time window to form an updated vibration signal sequence of the friction pair surface. This sequence is then subjected to fatigue damage state verification processing. Using the same method as steps S210 to S212, spectral features are extracted from the vibration signals, and the correlation parameter with the remaining effective thickness parameter is calculated. Alternatively, other damage assessment methods are used to calculate the adjusted fatigue damage accumulation rate. The fatigue damage accumulation rate before and after adjustment is compared to calculate the fatigue damage mitigation effect assessment parameter η = (v_before - v_after) / v_before, where v_before is the damage accumulation rate before adjustment, and v_after is the damage accumulation rate after adjustment.

[0194] Step S718: Compare the fatigue damage mitigation effect evaluation parameter with the preset mitigation effect threshold. When the fatigue damage mitigation effect evaluation parameter is greater than the preset mitigation effect threshold, record the braking parameter combination corresponding to the braking parameter adjustment control signal into the braking behavior mode optimization strategy library as an optimization strategy sample.

[0195] A preset mitigation effect threshold η_th is set. If η > η_th, it indicates that the braking parameter adjustment has achieved a significant mitigation effect. The braking parameter combination corresponding to the braking parameter adjustment control signal is marked as an effective optimization strategy and recorded in the braking behavior pattern optimization strategy library as an optimization strategy sample. This optimization strategy sample will be used for subsequent strategy library updates and optimizations, enabling the strategy library to continuously accumulate effective braking parameter adjustment experience and improve the accuracy and effectiveness of subsequent braking behavior pattern optimizations.

[0196] In one exemplary embodiment, a brake life prediction system is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the brake life prediction system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a brake life prediction method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the housing of the brake life prediction system, or an external keyboard, touchpad, or mouse, etc.

[0197] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for predicting brake life, characterized in that, The method includes: A wear deformation atlas of the target brake under continuous braking conditions is constructed. The wear deformation atlas is a set of graph structures that reflect the continuous deformation process of wear morphology after mapping the wear morphology changes of the friction pair surface under continuous braking action in the form of a three-dimensional grid structure according to the braking action sequence. The wear deformation atlas is subjected to surface morphology reconstruction processing to obtain a surface morphology reconstruction model. The surface morphology reconstruction model is a three-dimensional mesh model composed of multiple reconstructed triangular facets. Each reconstructed triangular facet is associated with and stored as a surface curvature value as a morphology feature attribute label. The surface morphology reconstruction model is input into a preset stress field evolution network for stress field inversion processing to generate an internal stress field inversion map. The preset stress field evolution network is a deep learning network model that includes a wear morphology feature extraction layer, a morphology spatial encoding layer, a graph convolution processing layer, a stress field evolution simulation layer, and an inversion mapping layer. It is used to map surface morphology information into an internal stress field distribution. The internal stress field inversion map is a three-dimensional spatial distribution map containing the correspondence between surface morphology and internal stress field generated by spatially superimposing the isosurface of the stress field inversion value with the surface morphology reconstruction model. Based on the internal stress field inversion diagram, material fatigue location processing is performed to obtain a material fatigue location distribution map. The material fatigue location distribution map is a three-dimensional distribution map containing the fatigue cumulative damage degree parameter of each spatial location point after filling in the fatigue cumulative damage degree parameter of each spatial location point according to the spatial coordinate position. Based on the material fatigue location distribution map, the remaining life of the target brake is predicted, generating a brake life prediction result that includes the minimum remaining effective thickness parameter and the remaining life cycle parameter.

2. The brake life prediction method according to claim 1, characterized in that, The construction of the wear and deformation atlas of the target brake under continuous braking conditions includes: The surface wear morphology image corresponding to each braking action of the target brake under continuous braking conditions is obtained. The surface wear morphology images corresponding to each braking action are arranged according to the execution time sequence of the braking action to form a wear morphology image time sequence with braking action time sequence identifier. The image pixel grayscale value difference calculation processing is performed on the wear morphology images corresponding to adjacent braking actions in the time sequence of the wear morphology images to obtain the grayscale difference distribution map of the wear morphology images between adjacent braking actions. According to the grayscale difference distribution map, the pixel point regions whose grayscale change gradient exceeds the preset grayscale change gradient threshold are extracted and marked as regions with significant wear morphology changes. Spatial coordinate calibration is performed on the pixels in the area of ​​significant wear morphology change to obtain the set of boundary contour points of each area of ​​significant wear morphology change in the coordinate system of the friction pair surface. The set of boundary contour points is then input into a preset wear deformation map generator to perform triangulation on the set of boundary contour points to generate a mesh structure of significant wear morphology change area composed of multiple triangular facets. The normal vector of each triangular facet in the mesh structure of the area with significant wear change is calculated to obtain the normal vector direction parameter of each triangular facet, and the normal vector direction parameter of each triangular facet is used as the deformation attribute label of the corresponding triangular facet. Based on the deformation attribute label of each triangular facet, the mesh structure of the area with significant wear change is segmented, and triangular facets with the same deformation attribute label and adjacent spatial positions are aggregated and classified into the same wear change sub-region. The normal vector direction parameters of all triangular facets within the same wear change sub-region are processed by vector synthesis. The direction of the synthesized vector of the normal vector direction parameters of all triangular facets within the same wear change sub-region is calculated as the regional deformation direction parameter of the wear change sub-region. This parameter is then associated with and stored with the braking action timing identifier corresponding to the wear change sub-region to generate a wear change sub-region deformation direction record for each braking action. Based on the deformation direction record of the wear change sub-region corresponding to each braking action, the wear change sub-regions corresponding to adjacent braking actions are associated and mapped according to the spatial position matching relationship to construct a wear deformation atlas that reflects the wear pattern of the friction pair surface under continuous braking action and the continuous deformation process of the structure.

3. The brake life prediction method according to claim 1, characterized in that, The surface topography reconstruction process performed on the wear deformation atlas to obtain a surface topography reconstruction model includes: The wear deformation map corresponding to each braking action is extracted from the wear deformation map set, and the wear deformation map corresponding to each braking action is sorted according to the time sequence of the braking action to form a wear deformation map sequence with time sequence identifier; For each wear deformation map in the wear deformation map sequence, the grid vertex coordinates are extracted to obtain the set of vertex spatial coordinates of all triangular facets in each wear deformation map. Based on the set of vertex spatial coordinates, the geometric center point coordinates of the vertex spatial coordinates of all triangular facets in each wear deformation map are calculated to obtain the geometric center point coordinate parameters of each wear deformation map. Using the geometric center point coordinates of each wear deformation map as a reference point, the spatial coordinates of the vertices of all triangular facets in each wear deformation map are transformed relative to each other to generate a set of relative coordinates of all triangular facet vertices in each wear deformation map. This set of relative coordinates is then input into the feature encoding layer of a preset surface topography reconstruction network to extract spatial position features from the set of relative coordinates of all triangular facet vertices in each wear deformation map, thus obtaining the spatial position feature encoding vector corresponding to each wear deformation map. The spatial position feature encoding vector corresponding to each wear deformation map is input into the feature fusion layer of the surface topography reconstruction network. The spatial position feature encoding vectors corresponding to adjacent braking actions are spliced ​​together in temporal order to generate a temporal feature fusion vector that reflects the continuous evolution of wear morphology. The temporal feature fusion vector is input into the topography reconstruction decoding layer of the surface topography reconstruction network. The temporal feature fusion vector is subjected to inverse feature mapping to generate a reconstructed surface three-dimensional point cloud data set. The reconstructed surface 3D point cloud data set is processed into a point cloud mesh. By connecting the point clouds in the reconstructed surface 3D point cloud data set according to spatial proximity, a surface 3D mesh model composed of multiple reconstructed triangular facets is generated. The surface curvature of each reconstructed triangular facet in the surface 3D mesh model is calculated to obtain the surface curvature value of each reconstructed triangular facet. The surface curvature value of each reconstructed triangular facet is used as the shape feature attribute label of the corresponding reconstructed triangular facet. The shape feature attribute labels of each reconstructed triangular facet are associated and stored with the surface three-dimensional mesh model to generate a surface shape reconstruction model containing the shape feature attribute labels of each reconstructed triangular facet.

4. The brake life prediction method according to claim 1, characterized in that, The step of inputting the surface morphology reconstruction model into a preset stress field evolution network for stress field inversion processing to generate an internal stress field inversion map includes: The surface morphology reconstruction model is input into the morphology feature extraction layer of the stress field evolution network. The morphology feature attribute labels of each reconstructed triangular facet in the surface morphology reconstruction model are processed into feature vectors to obtain the morphology feature vector of each reconstructed triangular facet. The topographic feature vector of each reconstructed triangular facet is input into the topographic space encoding layer of the stress field evolution network. The topographic feature vector of each reconstructed triangular facet is mapped to a preset three-dimensional spatial coordinate system to generate the spatial encoding position coordinates corresponding to each reconstructed triangular facet. Based on the spatial encoding position coordinates corresponding to each reconstructed triangular facet, a topographic space topology graph structure is constructed with each reconstructed triangular facet as a node and the spatial connection relationship between adjacent reconstructed triangular facets as edges. The topological spatial graph structure is input into the graph convolution processing layer of the stress field evolution network, and the neighborhood information aggregation processing is performed on the topological feature vector of each node in the topological spatial graph structure to obtain the neighborhood aggregation feature vector corresponding to each node. Through the stress field simulation layer of the stress field evolution network, the neighborhood aggregation feature vector corresponding to each node is input into the preset stress field differential equation solver, and the neighborhood aggregation feature vector of each node is iteratively solved to obtain the intermediate state vector of the stress field corresponding to each node. The intermediate state vector of the stress field corresponding to each node is input into the inversion mapping layer of the stress field evolution network. The intermediate state vector of the stress field of each node is subjected to inverse mapping transformation to obtain the stress field inversion value corresponding to each node. Based on the stress field inversion value corresponding to each node, a three-dimensional stress field inversion point cloud data set is constructed with the spatial coding position coordinates of each node as the horizontal and vertical coordinates and the stress field inversion value of each node as the vertical coordinate. Spatial interpolation processing is performed on the three-dimensional stress field inversion point cloud data set. By interpolating the point clouds in the three-dimensional stress field inversion point cloud data set according to spatial distance weights, a continuously distributed spatial distribution field of stress field inversion values ​​is generated. Isosurface extraction processing is performed on the spatial distribution field of stress field inversion values ​​to extract spatial location points in the continuously distributed spatial distribution field of stress field inversion values ​​where the stress field inversion value is equal to a preset stress field inversion threshold, and the spatial location points are connected to form an isosurface surface. The isosurface and the surface topography reconstruction model are spatially superimposed to generate an internal stress field inversion map that includes the correspondence between the surface topography and the internal stress field.

5. The brake life prediction method according to claim 1, characterized in that, The step of performing material fatigue localization processing based on the internal stress field inversion map to obtain a material fatigue localization distribution map includes: Extract the stress field inversion value corresponding to each spatial location point from the internal stress field inversion map, and mark the spatial location points whose stress field inversion value is greater than the preset material yield stress threshold as the set of plastic deformation starting points; For each plastic deformation starting point in the set of plastic deformation starting points, a neighborhood expansion process is performed. Taking each plastic deformation starting point as the center point, a search is performed on the surrounding spatial locations. Spatial locations whose stress field inversion value difference from the center point is less than a preset stress field inversion value difference threshold are assigned to the plastic deformation region corresponding to the center point, thus obtaining multiple plastic deformation regions. For each of the plurality of plastic deformation regions, the area of ​​each plastic deformation region is calculated to obtain the area parameter of each plastic deformation region, and plastic deformation regions whose area parameters are greater than the preset area parameter threshold are selected as the main plastic deformation regions. For each of the main plastic deformation regions, the region boundary curvature is calculated to obtain a region boundary curvature distribution map of each main plastic deformation region. Based on the region boundary curvature distribution map, the boundary location points where the region boundary curvature value exceeds the preset region boundary curvature threshold are identified as candidate crack initiation points. For each crack initiation candidate point, the stress cycle count is accumulated to obtain the stress cycle count parameter corresponding to each crack initiation candidate point. The stress cycle count parameter corresponding to each crack initiation candidate point is matched with the preset material fatigue life curve to obtain the fatigue damage degree parameter corresponding to each crack initiation candidate point. Based on the fatigue damage degree parameter corresponding to each crack initiation candidate point, the spatial coordinates of each crack initiation candidate point are associated and stored with the corresponding fatigue damage degree parameter to generate a fatigue damage degree record of the crack initiation candidate point. All crack initiation candidate points in the fatigue damage degree record are sorted in descending order of fatigue damage degree parameter, and a set number of crack initiation candidate points with the highest fatigue damage degree parameter in the sorting result are extracted as key fatigue damage points. Taking the spatial coordinates of the key fatigue damage point as the center, fatigue damage degree diffusion processing is performed to the surrounding spatial points. By distributing the fatigue damage degree parameter of the key fatigue damage point to the surrounding spatial points according to the spatial distance attenuation weight, fatigue damage degree parameter of each spatial point is generated. The fatigue damage degree parameters of each spatial location point are filled according to the spatial coordinate position to generate a material fatigue location distribution map containing the fatigue damage degree parameters of each spatial location point.

6. The brake life prediction method according to claim 1, characterized in that, The remaining life prediction processing of the target brake based on the material fatigue location distribution map generates a brake life prediction result containing the minimum remaining effective thickness parameter and the remaining life cycle parameter, including: Extract the fatigue damage degree parameter corresponding to each spatial location point from the material fatigue location distribution map, and mark the spatial location points whose fatigue damage degree parameter is greater than the preset fatigue damage failure threshold as the failure starting point set; For each failure starting point in the set of failure starting points, spatial connectivity analysis is performed to aggregate and classify spatially adjacent failure starting points into the same failure connectivity region, resulting in multiple failure connectivity regions. For each failed connected region, a penetration analysis is performed in the region thickness direction. The maximum penetration depth parameter of each failed connected region in the direction perpendicular to the friction pair surface is calculated to obtain the penetration depth parameter of each failed connected region. The difference between the penetration depth parameter of each failed connected region and the preset total thickness parameter of the friction pair is calculated to obtain the remaining effective thickness parameter corresponding to each failed connected region. The minimum value screening process is performed on the remaining effective thickness parameter corresponding to each failed connected region. The minimum value among the remaining effective thickness parameters corresponding to all failed connected regions is extracted as the minimum remaining effective thickness parameter. The ratio of the minimum remaining effective thickness parameter to the preset thickness decay rate parameter is calculated to obtain the initial remaining lifetime cycle parameter. The initial remaining life cycle parameters are subjected to working condition correction processing. The initial remaining life cycle parameters are weighted and adjusted according to the actual braking frequency parameters and actual braking intensity parameters of the target brake to obtain the corrected remaining life cycle parameters. The confidence interval of the corrected remaining life cycle parameter is estimated by performing confidence interval estimation. The upper and lower fluctuation range of the predicted cycle is calculated based on the distribution dispersion of the fatigue damage degree parameter in the material fatigue location distribution map, and the confidence interval parameter of the remaining life cycle is obtained. The corrected remaining lifetime lifespan parameter is associated and stored with the confidence interval parameter of the remaining lifetime lifespan to generate an intermediate remaining lifetime prediction result containing the remaining lifetime lifespan parameter and its confidence interval parameter. The intermediate results of the remaining life prediction are fused with the minimum remaining effective thickness parameter to generate a brake life prediction result that includes the minimum remaining effective thickness parameter and the remaining life cycle parameter.

7. The brake life prediction method according to claim 1, characterized in that, After performing remaining life prediction processing on the target brake based on the material fatigue location distribution map to generate brake life prediction results including minimum remaining effective thickness parameters and remaining life cycle parameters, the method further includes: The real-time surface vibration signal sequence generated by the target brake under the current braking action is obtained. The vibration signal spectrum feature extraction processing is performed on the real-time surface vibration signal sequence to obtain a real-time vibration signal spectrum feature map. The spectrum energy distribution difference processing is performed between the real-time vibration signal spectrum feature map and the preset healthy state vibration spectrum reference map to obtain a real-time vibration signal spectrum difference distribution map. Based on the real-time vibration signal spectrum difference distribution map, frequency ranges with spectrum energy distribution differences exceeding a preset spectrum energy distribution difference threshold are identified as abnormal vibration frequency ranges, and the center frequency parameters corresponding to the abnormal vibration frequency ranges are extracted. The center frequency parameter is mapped to the surface wear morphology parameter space to obtain the predicted surface wear morphology parameter set. The correlation verification process is performed between the predicted surface wear morphology parameter set and the remaining effective thickness parameter in the brake life prediction result. Each parameter in the predicted surface wear morphology parameter set and the remaining effective thickness parameter are normalized respectively. The correlation degree parameter between the two is calculated based on the normalized parameters. When the correlation parameter is less than the preset correlation threshold, the operation of inputting the surface morphology reconstruction model into the preset stress field evolution network for stress field inversion processing is triggered to generate an updated internal stress field inversion map. When the correlation parameter is greater than or equal to the preset correlation threshold, the brake life prediction result is marked as a verified brake life prediction result. The remaining effective thickness parameter in the verified brake life prediction result is processed to classify the maintenance level, and the brake maintenance level identifier corresponding to the remaining effective thickness parameter is obtained. The system calls a preset maintenance strategy template library based on the brake maintenance level identifier and extracts a maintenance operation instruction sequence associated with the brake maintenance level identifier from the preset maintenance strategy template library. The maintenance operation instruction sequence and the verified brake life prediction result are encapsulated to generate a brake maintenance decision instruction that includes the maintenance operation instruction sequence and the verified brake life prediction result.

8. The brake life prediction method according to claim 1, characterized in that, Before constructing the wear and deformation atlas of the target brake under continuous braking conditions, the method further includes: The material composition parameters and initial surface texture parameters of the target brake are obtained. The material composition parameters and the initial surface texture parameters are input into a preset microstructure initialization network. The microstructure initialization network is used to simulate the grain structure distribution of the material composition parameters to generate an initial grain structure distribution map of the material. The microstructure initialization network is used to perform surface texture morphology generation processing on the initial surface texture parameters to generate an initial surface texture morphology map. The initial distribution map of the material grain structure and the initial texture morphology map of the surface are spatially superimposed to generate an initial microstructure model containing grain structure distribution information and surface texture morphology information. The set of real-time braking pressure change curves and the set of real-time braking temperature change curves generated by the target brake during continuous braking test on a standard braking condition test bench are obtained. The set of real-time braking pressure change curves is subjected to pressure waveform feature extraction processing to obtain a pressure waveform feature parameter sequence. Temperature gradient feature extraction processing is performed on the set of real-time braking temperature change curves to obtain a temperature gradient feature parameter sequence; The pressure waveform feature parameter sequence and the temperature gradient feature parameter sequence are normalized respectively. The normalized pressure waveform feature parameter sequence and temperature gradient feature parameter sequence are then subjected to time-series correlation modeling to generate a working condition coupling evolution feature vector that reflects the coupling evolution relationship between braking pressure and braking temperature. The working condition coupling evolution feature vector is then mapped to each spatial location point of the initial microstructure model to generate the working condition loading parameters corresponding to each spatial location point. Based on the loading parameters corresponding to each spatial location point, the stress-strain response of the material grain structure in the initial microstructure model is calculated to obtain the grain structure deformation parameters corresponding to each spatial location point. The grain structure deformation parameters corresponding to each spatial location point are then superimposed onto the corresponding spatial location point of the initial microstructure model to generate an initial service microstructure model containing grain structure deformation information. The initial microstructure model of the service is stored as the initial input parameter of the preset stress field evolution network, and is used for subsequent execution of the operation of inputting the surface morphology reconstruction model into the preset stress field evolution network for stress field inversion processing.

9. The brake life prediction method according to claim 1, characterized in that, After performing remaining life prediction processing on the target brake based on the material fatigue location distribution map to generate brake life prediction results including minimum remaining effective thickness parameters and remaining life cycle parameters, the method further includes: Acquire a set of historical braking failure record data generated by the target brake during its historical service cycle. The set of historical braking failure record data includes parameters of the number of braking events before failure and the remaining thickness at the time of failure for each braking failure event. Extract the correspondence between the number of braking events before failure and the remaining thickness at the time of failure from the historical braking failure record data set, and construct an association mapping table between the number of braking events before failure and the remaining thickness at the time of failure; The remaining effective thickness parameter in the brake life prediction result is input into the association mapping table for matching query processing to obtain the predicted remaining braking number parameter corresponding to the remaining effective thickness parameter. Based on the historical average braking frequency of the target brake, the remaining life cycle parameter is converted into an equivalent remaining braking number parameter. The predicted remaining braking number parameter and the equivalent remaining braking number parameter are subjected to consistency analysis, and the conversion deviation parameter between the predicted remaining braking number parameter and the equivalent remaining braking number parameter is calculated. When the conversion deviation parameter is greater than the preset conversion deviation threshold, the remaining life cycle parameter in the brake life prediction result is corrected to generate a corrected brake life prediction result; when the conversion deviation parameter is less than or equal to the preset conversion deviation threshold, the brake life prediction result is marked as a verified brake life prediction result. The material fatigue location distribution map is subjected to three-dimensional color mapping processing to generate a three-dimensional color distribution map of material fatigue degree, and the remaining effective thickness parameter in the verified brake life prediction result is subjected to three-dimensional model section display processing to generate a three-dimensional section model of friction pair containing the remaining effective thickness indicator. The three-dimensional color distribution map of the material fatigue degree is spatially superimposed on the three-dimensional cross-sectional model of the friction pair to generate a life prediction visualization result containing fatigue degree distribution information and remaining effective thickness information. The life prediction visualization result is then output to the display terminal for display.

10. A brake life prediction system, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the brake life prediction method according to any one of claims 1 to 9 by executing the machine-executable instructions.

Images