Tire maintenance prediction, analysis and diagnosis system based on AI

Abstract

The invention discloses an AI-based tire maintenance prediction analysis and diagnosis system, and relates to the technical field of tire maintenance analysis, an acquisition layer obtains maintenance environment parameters, the maintenance environment parameters comprise a polishing tool state and a polishing depth, after the maintenance environment parameters are subjected to standardization processing, an environment state vector is generated, and the environment state vector is stored in a storage layer; an operation behavior parameter is obtained through real-time sensing equipment, after feature extraction is conducted on the operation behavior parameter, an operation state vector is generated, and a model building layer takes an environment state vector and the operation state vector as input variables and inputs the input variables into a time sequence neural network to build a damage evolution dynamic prediction model. And the control layer predicts a repair process state through a damage evolution dynamic prediction model, and dynamically adjusts a maintenance strategy to control a damage evolution process. The diagnosis system can describe the damage expansion rate, the interface bonding aging amplitude and the local stress change amplitude in a high-precision mode, prospective prediction of the state of the repairing process is achieved, the repairing quality is remarkably improved, and the follow-up tire service life is remarkably prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of tire repair analysis, and particularly relates to an AI-based tire repair prediction analysis and diagnosis system. Background Art

[0002] With the rapid development of the transportation industry, the vehicle ownership continues to grow. As a key component directly contacting the ground during vehicle operation, the health status of tires is directly related to driving safety and operation efficiency. During long-term use, tires are prone to be affected by various factors such as road conditions, driving behaviors, and climate changes, resulting in faults such as wear, bulging, cracking, and air leakage. If these problems cannot be detected and addressed in a timely manner, it will lead to a decline in vehicle performance at best and serious traffic accidents at worst. With the rapid development of artificial intelligence technology, introducing AI into tire health management has become a new technological trend. By intelligently analyzing vehicle sensor data (such as tire pressure, temperature, vibration signals, etc.), historical repair records, driving environment data, etc., potential faults can be identified in advance, tire life can be predicted, and personalized maintenance plans can be formulated, thus realizing the transformation from passive repair to active prediction and preventive maintenance.

[0003] The prior art has the following defects: Currently, during the tire repair process, the understanding of the dynamic process of damage expansion still remains at the stage of empirical summary, lacking a scientific prediction system based on time series characteristics. Diagnostic systems usually use static data modeling and cannot capture the microscopic evolution processes of crack propagation, interface aging, and stress changes during the repair process, resulting in a low prediction accuracy for future damage trends, delaying the best intervention time, and easily causing problems such as repair failure and premature tire scrapping.

[0004] Based on this, the present invention proposes an AI-based tire repair prediction analysis and diagnosis system, which can accurately depict the damage expansion rate, the amplitude of interface bonding aging, and the amplitude of local stress changes, realize the forward-looking prediction of the repair process state, and significantly improve the repair quality and the subsequent tire service life. Summary of the Invention

[0005] The purpose of the present invention is to provide an AI-based tire repair prediction analysis and diagnosis system to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An AI-based tire repair prediction analysis and diagnosis system, comprising an acquisition layer, a model construction layer, and a control layer; Acquisition layer: Obtain maintenance environment parameters and generate an environmental state vector, and obtain operation behavior parameters and generate an operation state vector; Model construction layer: Use the environmental state vector and the operation state vector as input variables and input them into a time series neural network to establish a damage evolution dynamic prediction model; Control layer: Predict the state of the repair process through the dynamic prediction model of damage evolution, and dynamically adjust the repair strategy when the prediction result is abnormal.

[0007] In a preferred embodiment, the time series prediction data output by the dynamic prediction model of damage evolution in the control layer includes the crack propagation amplitude, the interfacial bonding aging amplitude, and the local stress change amplitude, and whether there is an abnormality is predicted based on the crack propagation amplitude, the interfacial bonding aging amplitude, and the local stress change amplitude.

[0008] In a preferred embodiment, the crack propagation amplitude prediction formula: , where is the crack propagation amplitude, represents the crack propagation amplitude at the current moment. If , is the expansion threshold, predicting accelerated crack propagation; The interfacial bonding aging amplitude prediction formula: , where is the interfacial bonding aging amplitude, represents the interfacial bonding aging amplitude at the current moment. If , is the aging threshold, predicting a decrease in interfacial stability; The local stress change amplitude prediction formula: , where is the local stress change amplitude, represents the local stress value at the current moment, represents the time interval. If , is the change threshold, predicting a potential risk of structural imbalance.

[0009] In a preferred embodiment, when an abnormality is predicted, the adjustment mechanism is as follows: Grinding pressure adjustment: , where represents the adjusted grinding pressure, is the current grinding pressure, is the crack propagation amplitude, is the expansion threshold, is the adjustment coefficient; Grinding speed adjustment: , where is the adjusted grinding speed, is the current grinding speed, is the interfacial bonding aging amplitude, is the aging threshold, is the adjustment coefficient; Grinding path correction: , where is the corrected offset of the grinding trajectory, is the offset of the current center line of the grinding trajectory, is the amplitude of the local stress change, is the change threshold, is the adjustment coefficient.

[0010] In a preferred embodiment, the model construction layer concatenates the environmental state vector and the operation state vector to form a unified time series input, and constructs a three-dimensional input tensor from the concatenation results of all time steps; The position encoding mechanism is introduced to add time identifiers to the input sequence, and the th position encoding vector of the time step is obtained. The constructed input sequence is fed into a time series model based on a time series neural network structure, and the time series model uses the self-attention mechanism to model the dependence relationship between any two time steps; From the hidden state of each time step output by the time series neural network encoder , the structure is extracted to construct a damage evolution prediction model, and finally the time series prediction of the damage evolution index is output: For each time step a target function is constructed to obtain the th predicted output vector of the time step. The prediction results of all time steps are output in chronological order to form a complete dynamic damage evolution trajectory sequence.

[0011] In a preferred embodiment, the model construction layer concatenates the environmental state vector and the operation state vector to form a unified time series input: , and constructs a three-dimensional input tensor from the concatenation results of all time steps: , where represents the environmental state vector of the th time step, represents the operation state vector of the th time step, represents the fused input vector of the th time step, represents the vector concatenation operation, is the length of the time series, represents the dimension of the environmental state vector, represents the dimension of the operation state vector.

[0012] In a preferred embodiment, the model construction layer introduces a position encoding mechanism to add time identifiers to the input sequence. The position encoding function is defined as: , where Indicates the position encoding vector at the th time step, Indicates the position encoding dimension index, which is used to control the alternating calculation of sine and cosine, Indicates the current time step number, is the total vector dimension, Indicates the dimension of the environmental state vector, Indicates the dimension of the operation state vector. The final time-aware input vector is: , Indicates the final time-aware input vector, Indicates the th fused input vector at the time step.

[0013] In a preferred embodiment, the model construction layer constructs an objective function for each time step : , where, Indicates the predicted output vector at the th time step, Indicates the hidden state vector output by the temporal neural network, which is used to capture historical inputs and long-term dependencies, Indicates the th predicted value of the crack propagation amplitude at the time step, Indicates the th predicted value of the interfacial adhesion aging amplitude at the time step, The th local stress change amplitude value at the time step, is a non-linear mapping function; Outputs the prediction results of all time steps in chronological order to form a dynamic damage evolution trajectory sequence: , where, Indicates the prediction sequence matrix for the entire repair process, Indicates the length of the time series, Indicates the th predicted output vector at the time step.

[0014] In a preferred embodiment, the acquisition layer synchronously starts the environmental parameter acquisition task, collects real-time data during the grinding process, including the rotation speed of the grinding tool, the degree of tool wear, and the dynamic change of the grinding depth, and performs feature aggregation processing on the standardized grinding tool state data and grinding depth data to form a unified-dimension environmental state vector for describing the state representation of the current repair environment; Collect operation behavior parameters in real time, including grinding path trajectory, grinding angle, force variation, tool tilt angle, and operations in the repeated operation area. Obtain the trajectory deviation rate and local over-grinding risk factor for the collected original action sequence, and integrate the operation behavior parameters into an operation state vector.

[0015] In a preferred embodiment, the calculation expression of the trajectory deviation rate is: , where in the formula, is the trajectory deviation rate, is the total length of the ideal path, is the arc length coordinate along the grinding trajectory, is the spatial position vector of the actual grinding trajectory at parameter s, is the spatial position vector of the ideal grinding trajectory at parameter s, represents the Euclidean norm of the vector; The calculation expression of the local over-grinding risk factor is: , where in the formula, is the local over-grinding risk factor, is the projection range of the grinding action area on the two-dimensional plane, is the grinding depth measured at the position point (x, y), is the grinding depth threshold, represents only counting the positive residuals of the over-grinding area.

[0016] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention obtains the maintenance environment parameters through the acquisition layer. The maintenance environment parameters include the state of the grinding tool and the grinding depth. After standardizing the maintenance environment parameters, an environment state vector is generated. The operation behavior parameters are obtained through the real-time perception device, and after feature extraction of the operation behavior parameters, an operation state vector is generated. The model construction layer uses the environment state vector and the operation state vector as input variables and inputs them into a time series neural network to establish a dynamic prediction model of damage evolution. Control layer: Predict the state of the repair process through the dynamic prediction model of damage evolution, send a warning signal when the prediction result is abnormal, and dynamically adjust the repair strategy to control the damage evolution process. This diagnostic system introduces an acquisition layer, a model construction layer, and a control layer, forming a complete dynamic adaptive closed-loop mechanism, which can accurately depict the damage expansion rate, the interface bonding aging amplitude, and the local stress change amplitude, realize the forward-looking prediction of the repair process state, effectively inhibit the damage deterioration process, and significantly improve the repair quality and the subsequent service life of the tire. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0018] Figure 1 This is the system architecture diagram of the diagnostic system of the present invention.

[0019] Figure 2 This is the timing diagram of the diagnostic system of the present invention.

[0020] Figure 3 This is the operation flowchart of the diagnostic system of the present invention. Detailed implementation manners

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0022] Embodiment 1: Please refer to Figure 1 - Figure 2 As shown, the present embodiment provides an AI-based tire repair prediction and analysis diagnostic system, which includes an acquisition layer, a model construction layer, and a control layer; Acquisition layer: Obtain repair environment parameters through various sensors. The repair environment parameters include the status of the grinding tool and the grinding depth. After standardizing the repair environment parameters, an environmental status vector is generated. The operation behavior parameters are obtained through a real-time perception device, and after feature extraction of the operation behavior parameters, an operation status vector is generated. The environmental status vector and the operation status vector are sent to the model construction layer; Model construction layer: Use the environmental status vector and the operation status vector as input variables, and input them into a time series neural network (such as Transformer) to establish a damage evolution dynamic prediction model. The damage evolution dynamic prediction model is sent to the control layer; Control layer: Predict the status of the repair process through the damage evolution dynamic prediction model. The status of the repair process includes the crack propagation amplitude, the interface bonding aging amplitude, and the local stress change amplitude. When the prediction result is abnormal, a warning signal is sent, and the repair strategy is dynamically adjusted to control the damage evolution process.

[0023] This application obtains the maintenance environment parameters through the acquisition layer. The maintenance environment parameters include the state of the grinding tool and the grinding depth. After standardizing the maintenance environment parameters, an environmental state vector is generated. The operation behavior parameters are obtained through the real-time perception device, and after feature extraction of the operation behavior parameters, an operation state vector is generated. The model construction layer uses the environmental state vector and the operation state vector as input variables and inputs them into a time series neural network to establish a dynamic prediction model for damage evolution. The control layer predicts the state of the repair process through the dynamic prediction model for damage evolution, sends a warning signal when the prediction result is abnormal, and dynamically adjusts the maintenance strategy to control the damage evolution process. This diagnostic system introduces the acquisition layer, the model construction layer, and the control layer to form a complete dynamic adaptive closed-loop mechanism, which can accurately depict the damage expansion rate, the interface bonding aging amplitude, and the local stress change amplitude, realize the forward-looking prediction of the repair process state, effectively inhibit the damage deterioration process, and significantly improve the repair quality and the subsequent service life of the tire.

[0024] Please refer to Figure 3 as shown in the figure, the specific working process of the diagnostic system is as follows: The diagnostic system obtains the maintenance environment parameters through multiple types of sensors. The maintenance environment parameters include the state of the grinding tool and the grinding depth. After standardizing the maintenance environment parameters, an environmental state vector is generated. The operation behavior parameters are obtained through the real-time perception device, and after feature extraction of the operation behavior parameters, an operation state vector is generated. The environmental state vector and the operation state vector are used as input variables and input into a time series neural network (such as Transformer) to establish a dynamic prediction model for damage evolution. The state of the repair process is predicted through the dynamic prediction model for damage evolution. The state of the repair process includes the crack expansion amplitude, the interface bonding aging amplitude, and the local stress change amplitude. A warning signal is sent when the prediction result is abnormal, and the maintenance strategy is dynamically adjusted to control the damage evolution process.

[0025] Embodiment 2: The acquisition layer obtains the maintenance environment parameters through multiple types of sensors. The maintenance environment parameters include the state of the grinding tool and the grinding depth. After standardizing the maintenance environment parameters, an environmental state vector is generated. The operation behavior parameters are obtained through the real-time perception device, and after feature extraction of the operation behavior parameters, an operation state vector is generated. The environmental state vector and the operation state vector are sent to the model construction layer.

[0026] Multiple types of sensor devices are arranged in the tire repair operation area, specifically including grinding tool state monitoring sensors (such as current, voltage, vibration, temperature rise sensors) and grinding depth monitoring sensors (such as laser ranging, ultrasonic thickness measurement sensors), and parameters such as the acquisition frequency, accuracy threshold, and data buffer are initialized and configured.

[0027] After the maintenance operation is started, the acquisition layer synchronously starts the environmental parameter acquisition task to collect real-time data during the grinding process, including but not limited to: multi-dimensional data such as the rotation speed of the grinding tool, the change of motor load, the temperature change trend, the degree of tool wear, and the dynamic change of grinding depth.

[0028] Through the sliding time window and normalization processing algorithms (such as z-score normalization, min-max normalization), the original collected maintenance environmental parameter data is standardized to eliminate the data distribution differences caused by different maintenance environments and equipment models, and enhance the generalization ability of subsequent modeling.

[0029] The standardized grinding tool state data and grinding depth data are subjected to feature aggregation processing, and dimension fusion technologies (such as principal component analysis PCA or convolutional feature compression network) are used to form an environmental state vector with a unified dimension to describe the state representation of the current maintenance environment. The relevant code examples are as follows: import_torch import_torch.nn_as_nn import_torch.nn.functional_as_F class_Feature_Compressor(nn.Module): def__init__(self): super(FeatureCompressor,self).__init__() # Tool state convolution channel self.conv_tool=nn.Sequential( nn.Conv1d(in_channels=8,out_channels=16,kernel_size=3,padding=1), nn.ReLU(), nn.Conv1d(in_channels=16,out_channels=16,kernel_size=3,padding=1), nn.ReLU() ) # Grinding depth convolution channel self.conv_depth=nn.Sequential( nn.Conv1d(in_channels=4,out_channels=16,kernel_size=3,padding=1), nn.ReLU(), nn.Conv1d(in_channels=16, out_channels=16, kernel_size=3, padding=1), nn.ReLU() ) # Aggregation + compression to a unified vector dimension self.pool = nn.AdaptiveAvgPool1d(output_size=1) # Aggregate the time dimension self.fc = nn.Linear(in_features=32, out_features=32) # Unify the dimension def forward(self, tool_input, depth_input): # Input shape: (batch_size, T, feature_dim) tool_input = tool_input.permute(0, 2, 1) # Convert to (batch_size, feature_dim, T) depth_input = depth_input.permute(0, 2, 1) tool_feat = self.conv_tool(tool_input) # Shape: (batch_size, 16, T) depth_feat = self.conv_depth(depth_input) # Shape: (batch_size, 16, T) fused = torch.cat([tool_feat, depth_feat], dim=1) # Shape: (batch_size, 32, T) pooled = self.pool(fused) # Shape: (batch_size, 32, 1) pooled = pooled.squeeze(-1) # Shape: (batch_size, 32) output = self.fc(pooled).unsqueeze(1) # Shape: (batch_size, 1, 32) return output Usage example: model = FeatureCompressor() tool_data = torch.randn(16, 50, 8) # batch_size = 16, time steps = 50, tool features = 8 depth_data = torch.randn(16, 50, 4) # batch_size = 16, time steps = 50, depth features = 4 env_vector = model(tool_data, depth_data) # Output shape: [16, 1, 32] print(env_vector.shape) In the above code example, the shape of the grinding tool state data is [batch_size, T, 8], indicating that the time series length is T and it contains 8 state dimensions; the shape of the grinding depth data is [batch_size, T, 4], indicating that the time series length is T and it contains 4 depth-related dimensions; the dimension of the output unified environmental state vector is [batch_size, 1, 32], which extracts local patterns of the time series through convolutional layers; AdaptiveAvgPool1d compresses different time series lengths into a fixed length; the fully connected layer compresses to the target dimension; the output is a unified environmental state vector, which is convenient for subsequent input into model building layers such as Transformer.

[0030] Through an inertial measurement unit (IMU), motion capture system, or machine vision module installed on the maintenance tool, the motion characteristics of the operator are collected in real time, including operation behavior information such as the grinding path trajectory, grinding angle, force change, tool tilt angle, and repeated operation area. Feature extraction is performed on the collected original motion sequence, such as the trajectory deviation rate and local over-grinding risk factor. The above-mentioned extracted operation behavior characteristics are integrated into an operation state vector in a unified format, which is used as a multi-dimensional state description vector of the maintenance behavior to characterize the operation quality and potential error risk.

[0031] The trajectory deviation rate measures the geometric deviation degree between the operation trajectory and the ideal grinding path, and its calculation expression is: , where is the trajectory deviation rate, is the total length of the ideal path, is the arc length coordinate along the grinding trajectory, is the spatial position vector of the actual grinding trajectory at parameter s, is the spatial position vector of the ideal grinding trajectory at parameter s, represents the Euclidean norm (i.e., distance) of the vector.

[0032] The local over-polishing risk factor measures the degree of over-polishing of the polishing depth in a certain area relative to the preset depth threshold per unit area, and the expression is: , where, is the local over-polishing risk factor, is the projection range of the polishing action area on the two-dimensional plane, is the polishing depth measured at the position point (x, y), is the polishing depth threshold, that is, the maximum safe polishing depth allowed by the material, represents only the positive residuals of the "over-polished" area (the part exceeding the threshold).

[0033] Through the data bus or the edge computing module, the generated environmental state vector and operation state vector are merged and packed in time series and sent to the model construction layer for further damage evolution modeling and prediction analysis.

[0034] The model construction layer takes the environmental state vector and operation state vector as input variables and inputs them into a time series neural network (such as Transformer) to establish a damage evolution dynamic prediction model, and the damage evolution dynamic prediction model is sent to the control layer.

[0035] The core work of the model construction layer is first to receive and fuse two types of input information from the acquisition layer, namely the standardized environmental state vector and operation state vector, which respectively represent multi-dimensional features related to operation behaviors such as the state of the maintenance tool, polishing depth, and tool vibration frequency. Since these two types of vectors have significant differences in dimension, time series characteristics, and semantics, feature dimension alignment and fusion must be carried out first.

[0036] Assume: represents the environmental state vector at the th time step (including sensor parameters such as the state of the polishing tool (such as rotation speed, temperature rise), polishing depth, and environmental humidity): represents the operation state vector at the th time step (reflecting maintenance behavior characteristics such as hand trajectory, polishing trajectory angle, contact force, etc.); the two are concatenated to form a unified time series input: , represents the fused input vector at the th time step, represents the vector concatenation operation, and the concatenation results of all time steps are formed into a three-dimensional input tensor: , and this fused input retains the time series structure, forms a complete input sequence, and serves as the input basis for Transformer.

[0037] Since the temporal neural network is essentially a fully connected model based on the self-attention mechanism and its structure does not have natural temporal perception ability, a position encoding mechanism needs to be introduced to add time identifiers to the input sequence, so that the temporal neural network can understand the evolution trend of events over time.

[0038] Let the input at the -th time step be , then the position encoding function is defined as: , where represents the position encoding vector at the -th time step, represents the position encoding dimension index, which is used to control the alternating calculation of sine and cosine, represents the current time step number, is the total vector dimension, represents the dimension of the environmental state vector, represents the dimension of the operation state vector. The final time-aware input vector is: . The addition of time identifiers enables the model to capture the dynamic laws of damage over time during learning, such as how stress concentration points migrate with grinding time and the fatigue accumulation trend of the adhesive layer.

[0039] The constructed input sequence is fed into the temporal model based on the Transformer structure. The core mechanism in Transformer is the self-attention mechanism, which can model the dependency relationship between any two time steps, thereby uncovering the delayed and non-linear association between the grinding behavior and the damage behavior. The attention mechanism is defined as follows: , where: represents the Query matrix, which is obtained by linearly mapping the input vector, represents the Key matrix, which is also obtained by linearly transforming the input vector, represents the Value matrix, which contains information representing values; is the dimension of the key; is the length of the time series (i.e., the discrete number of steps of the maintenance process duration), is the dimension of the Key vector, is the dimension of the Value vector, and softmax is used to normalize the importance weights of each time step. By stacking multiple Transformer encoder layers, the model can capture multi-level semantic interactions, such as whether the fluctuation of the tool vibration frequency triggers the crack propagation in the bonding area after multiple time steps.

[0040] The hidden state at each time step output from the Transformer encoder is used to extract key structures for constructing the final output of the damage evolution prediction model, which is the time series prediction of the following three types of key damage evolution indicators: Crack growth amplitude: measuring the crack growth amount per unit time; Interface bonding aging amplitude: measuring the trend of the bonding layer strength deterioration; Local stress change amplitude: measuring the stress change amplitude in a specific area.

[0041] For each time step a target function is constructed: , where represents the predicted output vector at the -th time step, which contains the damage evolution prediction indicators in three dimensions. represents the hidden state vector (high-dimensional feature) output by the Transformer, which is used to capture the historical input and long-term dependencies. represents the predicted value of the crack growth amplitude at the -th time step (the crack length growth rate per unit time). represents the predicted value of the interface bonding aging amplitude at the -th time step. The -th time step's local stress change amplitude value, representing the stress concentration evolution trend, is a non-linear mapping function, which can be implemented by a feed-forward neural network.

[0042] Finally, the prediction results of all time steps are output in chronological order to form a complete dynamic damage evolution trajectory sequence, which is used as the input reference for the control layer: , where represents the prediction sequence matrix of the entire repair process, and each row is the predicted value of the damage state at a time step, represents the length of the time series, represents the predicted value of the three-dimensional damage evolution indicator at the -th time step. This sequence can not only be used to identify local abnormal states (such as rapid crack growth in a certain period of time), but also can evaluate the impact of the overall repair operation on the tire structure stability through continuous trends, realizing: prospective modeling of the future damage development trajectory; quantitative feedback on the impact results of the repair operation behavior; real-time warning of potential risk points (such as local stress anomalies).

[0043] The design of this model construction layer realizes the deep integration of environment and behavior, the dynamic tracking of multi-dimensional damage, and the deep learning of time-series evolution trends. Compared with traditional empirical rules or single-variable prediction methods, it has the following technical advantages: it can adapt to complex and changeable maintenance environments and different workers' operating habits; it can effectively model the delay response and non-linear damage mechanism in the maintenance process; it can realize the real-time prediction and early intervention of damage expansion during maintenance; it supports the dynamic adaptability and real-time optimization of maintenance control strategies in the time dimension.

[0044] This model can be widely applied to scenarios such as tire defect repair, military heavy-duty tire maintenance, intelligent maintenance robots, etc., and is a key support for the deep integration of AI and the physical behavior of tire structures.

[0045] The control layer predicts the state of the repair process through a dynamic damage evolution prediction model. The state of the repair process includes the crack propagation amplitude, the interface bonding aging amplitude, and the local stress change amplitude. When the prediction result is abnormal, a warning signal is sent, and the maintenance strategy is dynamically adjusted to control the damage evolution process.

[0046] In the AI-based tire repair prediction and analysis system, the control layer undertakes the key decision-making scheduling and dynamic intervention tasks. Its core goal is to: based on the "dynamic damage evolution prediction result" output by the model construction layer, perform real-time perception, abnormal identification, and intervention adjustment on risk states such as crack evolution, bonding aging, and local stress distribution during the tire repair process, so as to improve the repair accuracy, reduce the secondary risk of repair, and extend the service life of the tire.

[0047] The entire control process is divided into three major stages: (1) receiving and evaluating the damage state prediction result, (2) abnormal identification and prediction, (3) dynamic control and feedback optimization of the maintenance strategy.

[0048] The time-series prediction data output by the system's dynamic damage evolution prediction model focuses on the trends of the following three physical indicators: Crack -Propagation -Rate: It represents the crack propagation speed per unit time and reflects the severity of material damage expansion in the tire structure during the repair and grinding process.

[0049] Adhesive- Interface- Aging -Rate: It measures the rate of performance degradation at the interface between the adhesive layer or patch and the original tire structure, and determines whether the repair adhesion is durable and reliable.

[0050] Local -Stress- Variation- Trend: It monitors the local stress fluctuations caused by excessive grinding or abnormal load distribution in the repair area and reflects the structural mechanical stability.

[0051] To ensure that the maintenance process is always within the safe and stable dynamic balance range, the system sets the following prediction formula for trend monitoring.

[0052] (1) Crack propagation amplitude prediction formula: , where is the crack propagation amplitude, represents the crack propagation amplitude at the current moment, represents the time interval, which is the sampling period between adjacent moments. Calculated with the differential idea, it calculates the change slope of crack propagation per unit time to predict whether the current crack is accelerating. If , is the expansion threshold, it is considered that the crack propagation is accelerating and in a dangerous state.

[0053] (2) Bonding aging amplitude prediction formula: , where is the bonding aging amplitude, represents the bonding aging amplitude at the current moment, represents the time interval, which is the sampling period between adjacent moments, representing the degradation rate of bonding performance per unit time, identifying interface aging caused by grinding heat, interface contamination, etc. If , is the aging threshold, the interface stability decreases, indicating that the gluing may fail.

[0054] (3) Local stress change amplitude prediction formula: , where is the local stress change amplitude, represents the local stress value at the current moment, represents the time interval, which is the sampling period between adjacent moments, evaluating the non-uniform change degree of the local stress field, capturing whether there is instability caused by excessive grinding during manual operation or unrelieved residual stress of the material. If , is the change threshold, there is a hidden danger of structural imbalance.

[0055] When the prediction result exceeds the safety threshold, the system will actively adjust the grinding parameters to inhibit further damage expansion. The adjustment mechanism is as follows: (1) Grinding pressure adjustment model: , where represents the adjusted grinding pressure, is the current grinding pressure, is the crack propagation amplitude, is the expansion threshold, reducing the excessive crack propagation rate. When the crack propagates too fast, the grinding pressure is automatically reduced, and the adjustment coefficient Indicates the sensitivity of the system to abnormal crack propagation, usually with a value range of .

[0056] (2) Grinding speed adjustment model: , where is the adjusted grinding speed, is the current grinding speed, is the adhesive aging amplitude, is the aging threshold, reducing the aging of the bonding interface caused by excessive grinding. By slowing down the operation time, the heat conduction and the accumulation of interfacial shear stress are controlled, and the adjustment coefficient represents the control of aging sensitivity, and the recommended value range is .

[0057] (3) Grinding path correction formula: , where is the offset of the corrected grinding trajectory, is the offset of the current center line of the grinding trajectory, is the amplitude of local stress change, is the change threshold, correcting the local stress concentration caused by path deviation or repeated grinding. The path adjustment amount is proportional to the amplitude of local stress change, and the adjustment coefficient indicates that it can be set based on the tire material or the repair position to control the degree of path deviation correction.

[0058] Example 3: In the above Example 2, the processing logic of the model construction layer is given, that is, the detailed steps of establishing a dynamic damage evolution prediction model by taking the environmental state vector and the operation state vector as input variables and inputting them into a temporal neural network (such as Transformer). In this example, an example of the establishment steps of the dynamic damage evolution prediction model is given, as follows: Purpose: Convert multi-source heterogeneous sensor data and behavior parameters into a unified time series structure as the basic input of the model.

[0059] Collect original data: Grinding tool state: including tool speed (rpm), temperature rise (°C), vibration value, etc. Grinding depth: Obtained by laser ranging or structured light scanning. Trajectory deviation rate: Calculated from the geometric difference between the actual trajectory and the ideal trajectory. Local over-grinding risk factor: Calculated based on grinding intensity, time, and remaining material thickness.

[0060] All data are aligned according to the time stamp and sampled at equal intervals (such as every second) to construct a series of time step data Use the z-score or min-max method to normalize each parameter to prevent model deviation caused by inconsistent numerical dimensions, and construct the input vector: , where represents the The state of the step grinding tool (such as rotational speed or temperature rise). Indicates the grinding depth of the step. Indicates the trajectory deviation rate.

[0061] Concatenate the environmental state vector and the operation state vector to form a unified input sequence: , construct a sequence for the entire time period: , introduce positional encoding and feed it into the Transformer encoder. Since the Transformer lacks temporal awareness, positional encoding needs to be added: , where: Indicates the positional encoding vector at the th time step. The overall input is fed into the Transformer-Encoder:

[0062] Define the output for each time step as: , and the model training objective is to make the output predict the damage state evolution, including: Crack-Growth-Rate: predicting the growth rate of crack length in the current state. Adhesion-Aging: predicting the degradation trend of the bonding interface. Local stress change amplitude: predicting the evolution of stress concentration that may cause fatigue damage.

[0063] For example, the input raw data (assuming the unit is the normalized value) is as follows: Grinding tool state (wear rate): 0.75; Grinding depth: 1.20; Trajectory deviation rate: 0.35; Grinding factor at the same part: 0.40.

[0064] We fuse these features into two vectors: Environmental state vector ; Operation state vector ; Merge them into the joint vector sequence of the model input (assuming the current time window is : ; Damage evolution dynamic prediction model (Transformer): Feature embedding transformation (Linear -projection): Transform the input vector: Let: ; ; Calculate: ; ; ; Obtain the embedding vector: ; This part is processed by the encoder for the time series data (we only process single-step examples here), and the predicted damage state is output: Prediction targets: crack propagation amplitude, bonding aging amplitude, local stress change; Assume that the trained Transformer maps the embedding vector to the following function (simplified simulation): ; Substitute into the calculation: ; ; .

[0065] The real-time prediction of the damage evolution trend during maintenance is realized through the damage evolution dynamic prediction model; it supports dynamically adjusting the grinding path or repair method according to the results; it improves the service life and safety stability of the tire; it can be extended to a multi-time-step prediction model for dynamic monitoring of the whole maintenance process.

[0066] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0067] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An AI-based tire maintenance prediction analysis and diagnosis system, characterized by: Includes acquisition layer, model building layer and control layer; Acquisition layer: obtain maintenance environment parameters and generate environment state vectors, obtain operation behavior parameters and generate operation state vectors; Model building layer: The environmental state vector and the operating state vector are used as input variables and input into the time series neural network to establish a dynamic prediction model for damage evolution; Control layer: Predict the state of the repair process through the dynamic prediction model of damage evolution, and dynamically adjust the maintenance strategy when there are abnormalities in the prediction results.

2. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 1, characterized in that: The time series prediction data output by the dynamic prediction model for damage evolution of the control layer includes crack extension amplitude, interface adhesion aging amplitude and local stress change amplitude, and whether there is an abnormality is predicted based on the crack extension amplitude, interface adhesion aging amplitude and local stress change amplitude.

3. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 2, characterized in that: Crack extension amplitude prediction formula: , where is the crack extension amplitude, represents the crack extension amplitude at the current moment. If , is the extension threshold, predicting the acceleration of crack growth; Bond aging amplitude prediction formula: , where is the bonding aging amplitude, Indicates the bonding aging amplitude at the current moment. If , is the aging threshold, predicting the decrease in interface stability; The prediction formula of local stress change amplitude is: , where is the amplitude of local stress variation, represents the local stress value at the current moment, Represents a time interval, if , It is the change threshold, and it is predicted that there is a risk of structural imbalance.

4. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 3, characterized in that: When there are anomalies in the forecast, the adjustment mechanism is: Grinding pressure adjustment: , where Indicates the adjusted grinding pressure. is the current grinding pressure, is the crack extension amplitude, is the expansion threshold, is the adjustment coefficient; Grinding speed adjustment: , where For the adjusted grinding speed, is the current grinding speed, is the bonding aging amplitude, is the aging threshold, is the adjustment coefficient; Grinding path correction: , where is the corrected grinding track offset, is the center line offset of the current grinding track, is the amplitude of local stress variation, is the change threshold, is the adjustment coefficient.

5. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 4, characterized in that: The model building layer concatenates the environment state vector and the operation state vector to form a unified time series input, and concatenates all time steps into a three-dimensional input tensor; Introduce the position encoding mechanism to add time stamps to the input sequence and obtain the The position encoding vector of time steps is constructed as the input sequence It is fed into a time series model based on a time series neural network structure. The time series model uses a self-attention mechanism to model the dependency between any two time steps. Each time step hidden state output from the temporal neural network encoder In the above example, the extracted structure is used to construct a damage evolution prediction model, and finally the time series prediction of damage evolution indicators is output: For each time step Construct the objective function and obtain the The prediction output vector of each time step is generated, and the prediction results of all time steps are output in chronological order to form a complete sequence of dynamic damage evolution trajectories.

6. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 5, characterized in that: The model building layer concatenates the environment state vector and the operation state vector to form a unified time series input: , and concatenate the results of all time steps into a three-dimensional input tensor: ,in, Indicates The environment state vector for each time step is Indicates The operation state vector of time steps, Indicates The fused input vector of time steps, represents the vector concatenation operation, is the length of the time series, represents the dimension of the environment state vector, Represents the dimension of the operation state vector.

7. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 6, characterized in that: The model building layer introduces a position encoding mechanism to add time tags to the input sequence. The position encoding function Defined as: ,in, Indicates The position encoding vector of time steps, Represents the position encoding dimension index, which is used to control the alternating calculation of sine and cosine. represents the current time step number, is the total vector dimension, represents the dimension of the environment state vector, Represents the dimension of the operation state vector, and the final time-aware input vector is: , represents the final time-aware input vector, Indicates The fused input vector for each time step.

8. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 7, characterized in that: The model building layer for each time step Construct the objective function: ,in, Indicates The predicted output vector for time steps is Represents the hidden state vector of the temporal neural network output, used to capture historical inputs and long-term dependencies, Indicates The predicted value of crack extension amplitude at time step, Indicates The predicted value of the interface adhesion aging amplitude for time steps, No. The local stress change amplitude value of the time step, is a nonlinear mapping function; The prediction results of all time steps are output in chronological order to form a sequence of dynamic damage evolution trajectories: ,in, Represents the prediction sequence matrix of the entire patching process, represents the length of the time series, Indicates The predicted output vector for each time step.

9. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 8, characterized in that: The acquisition layer synchronously starts the environmental parameter acquisition task to collect real-time data during the grinding process, including the grinding tool rotation speed, tool wear degree, and dynamic changes in grinding depth, and performs feature aggregation processing on the standardized grinding tool state data and grinding depth data to form an environmental state vector of unified dimension, which is used to describe the state representation of the current maintenance environment; The operation behavior parameters are collected in real time, including grinding path trajectory, grinding angle, force change, tool tilt angle and repeated operation area operation. The trajectory deviation rate and local over-grinding risk factor of the collected original action sequence are characterized and the operation behavior parameters are integrated into the operation state vector.

10. The AI-based tire maintenance prediction analysis and diagnosis system according to claim 9, characterized in that: The calculation expression of the trajectory deviation rate is: , where is the trajectory deviation rate, is the total length of the ideal path, is the arc length coordinate along the grinding trajectory, is the spatial position vector of the actual grinding trajectory at parameter s, is the spatial position vector of the ideal grinding trajectory at parameter s, Represents the Euclidean norm of a vector; The calculation expression of the local over-grinding risk factor is: , where is the risk factor for local over-polishing, is the projection range of the polishing action area on the two-dimensional plane, is the grinding depth measured at the position point (x, y), To polish the depth threshold, Indicates that only the positive residuals of the polished area are counted.

Images