A tire life prediction method and system based on big data

Abstract

The application relates to the technical field of tire life prediction, in particular to a tire life prediction method and system based on big data, which comprises the following steps: acquiring multi-source operation data and vehicle state information of a target tire, and a reference tire life database and sample characteristics corresponding to the target tire; cleaning and extracting features from the multi-source operation data to obtain a wear characteristic sequence and a load characteristic sequence of the target tire; and constructing a wear trend sequence and a load evaluation sequence of the target tire according to the reference tire life database, the wear characteristic sequence and the load characteristic sequence.The application can accurately predict the remaining life of the target tire through the collection and fusion of multi-source data and the combination of an attention mechanism, and can help to provide more accurate and personalized life prediction by analyzing the wear deviation and load overrun at different time nodes.

Description

Technical Field

[0001] This invention relates to the field of tire life prediction technology, specifically to a tire life prediction method and system based on big data. Background Technology

[0002] Currently, traditional methods are usually based on static models for prediction, which lacks adaptability to real-time changes. This means that they are difficult to effectively cope with the effects of different working conditions, environmental changes and load fluctuations. The prediction results may be inaccurate and cannot reflect the actual use status of the tire in real time. Moreover, traditional methods rely on only a single data source for life prediction and ignore other potential influencing factors. This makes the prediction accuracy of traditional methods relatively low and unable to fully reflect the overall condition of the tire.

[0003] Furthermore, traditional methods generally use a uniform life prediction standard, failing to provide personalized predictions for each tire based on specific usage conditions. This makes the prediction results rather general and unable to meet the specific needs of different users and vehicles. Moreover, traditional methods often fail to detect potential safety hazards in tires in a timely manner, such as excessive wear or exceeding load capacity limits, causing users to discover problems only when the tires reach their critical point, increasing the risk of safety accidents. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a tire life prediction method based on big data, comprising:

[0005] Acquire multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire;

[0006] Multi-source operational data are cleaned and feature extracted to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, the wear trend sequence and load assessment sequence of the target tire are constructed.

[0007] According to the time sequence position of the timestamp in the reference tire life database, the wear trend sequence and the load assessment sequence are compared to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism.

[0008] Global average pooling is performed on the second historical monitoring data sequence from multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is the historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types;

[0009] The pooling features, attention weight features, correlation features, and fusion feature sequences are concatenated to obtain concatenated features. These concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features.

[0010] Based on the mask feature, the vehicle state information is extracted to obtain multi-dimensional state features. The multi-dimensional state features are then subjected to regression processing to output the target tire's target remaining life information. The target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted probability of tire blowout.

[0011] Preferably, the multi-source operational data includes wear monitoring data, mileage data, road condition data, and load history data;

[0012] Acquire multi-source operational data and vehicle status information for the target tire, including:

[0013] Acquire initial sensing data of the target tire, including tread depth, tire temperature, driving speed and ground reaction force for multiple monitoring cycles;

[0014] Outlier removal and missing value imputation are performed on the initial sensor data to obtain multi-source operational data;

[0015] Obtain driving behavior information, environmental climate information, and maintenance record information of the vehicle to which the target tire belongs;

[0016] According to the preset driving behavior classification rules, the driving behavior information is tagged to obtain the vehicle's driving style characteristic information.

[0017] The temperature and humidity index of the environmental climate information is calculated to obtain the temperature and humidity characteristic information of the environmental climate information. The maintenance record information and the temperature and humidity characteristic information are summarized to obtain the environmental maintenance characteristic information of the vehicle.

[0018] Vehicle status information is obtained based on driving style characteristics and environmental maintenance characteristics.

[0019] Preferably, the wear characteristic sequence includes wear depth value, mileage value, and road surface roughness value; the load characteristic sequence includes vertical load value, lateral load value, and tire pressure fluctuation value; the wear trend sequence includes the baseline wear curve and the actual wear curve; and the load evaluation sequence includes the working condition load sequence and the structural stress sequence.

[0020] Based on the reference tire life database, wear characteristic sequence, and load characteristic sequence, a wear trend sequence and load assessment sequence for the target tire are constructed, including:

[0021] According to the time sequence position of the timestamp in the reference tire life database, the load values ​​under each working condition in the actual load curve and the impact load values ​​of each abnormal high load working condition are sorted and combined to obtain the working condition load sequence.

[0022] The estimated structural stress values ​​for each working condition in the actual load curve and the peak structural stress values ​​for each abnormally high load condition are sorted and combined to obtain the structural stress sequence.

[0023] The wear curve and the actual wear curve are differentially calculated to determine the wear deviation at each time node in the fused feature sequence. Based on the wear deviation at each time node, the wear deviation vector at each time node is generated. The wear deviation vectors at each time node are combined to obtain the wear deviation matrix of the fused feature sequence.

[0024] By calculating the ratio between the baseline load curve and the actual load curve, the load over-limit situation of each working condition in the fused feature sequence is determined. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence.

[0025] The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

[0026] Preferably, based on the wear deviation at each time point, a wear deviation vector for each time point is generated, including:

[0027] For any given time point, construct the initial wear vector for that time point; the initial wear vector includes normal wear interval positions, abnormal wear interval positions, multiple levels of mild wear indicator positions, and multiple levels of severe wear indicator positions.

[0028] If the wear deviation at a time point is greater than a preset threshold, the values ​​of the abnormal wear interval position in the initial wear vector and the values ​​of the severe wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value to generate the wear deviation vector at the time point.

[0029] If the wear deviation at a time point is less than or equal to a preset threshold, the values ​​of the normal wear interval position in the initial wear vector and the values ​​of the mild wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value, thereby generating the wear deviation vector at the time point.

[0030] Preferably, based on the load exceeding the limit under each working condition, a load coefficient vector for each working condition is generated, including:

[0031] For any given load condition, construct an initial load vector for that load condition; the initial load vector includes multiple candidate load level identifier bits;

[0032] Among multiple candidate load level identifiers, determine the target load level identifier corresponding to the actual load of the working condition, set the value of the target load level identifier to the first preset value, and set the values ​​of other load level identifiers in the initial load vector to the second preset value to obtain the load coefficient vector of the working condition.

[0033] Preferably, based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through an attention mechanism, including:

[0034] The first historical monitoring data sequence is obtained from multiple historical monitoring data sequences, and the type of monitoring data included in the first historical monitoring data sequence is a preset type;

[0035] The tire identifier of each target tire is processed with the monitoring data of each associated monitoring point contained in the first historical monitoring data sequence using the target attention mechanism to obtain the attention weight features of each target tire corresponding to different associated monitoring points.

[0036] The monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points are obtained from the first historical monitoring data sequence; the current associated monitoring point is any one associated monitoring point, and the other associated monitoring points are all associated monitoring points other than the current associated monitoring point.

[0037] Construct a similarity matrix based on the cosine distance between the monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points;

[0038] Based on the largest matrix element in the similarity matrix and the pre-constructed normal distribution, the associated features corresponding to the target tire are obtained.

[0039] Preferably, the lifetime prediction model includes a feature encoding layer, a feature fusion layer, and a lifetime decoding layer;

[0040] Based on mask features, operational state features are extracted from vehicle state information to obtain multi-dimensional state features. Regression processing is then performed on these multi-dimensional state features to output the target tire's remaining lifespan information, including:

[0041] Wear load features are extracted from the fused feature sequence based on the feature coding layer, and operating status features of vehicle status information are extracted.

[0042] The wear load features and operating status features are stitched together based on the feature fusion layer, and the stitched features are subjected to nonlinear transformation to obtain multidimensional status features.

[0043] The target tire's remaining life information is output by performing regression processing on the multidimensional state features based on the life decoding layer.

[0044] Preferably, the feature coding layer includes a sequence wear coding unit; the fused feature sequence includes a reference wear curve, an actual wear curve, a reference load curve, and an actual load curve;

[0045] Wear load features are extracted from the fused feature sequence based on the feature coding layer, including:

[0046] The wear deviation of each time node in the fused feature sequence is determined by performing differential calculation on the reference wear curve and the actual wear curve through the sequence wear coding unit. Based on the wear deviation of each time node, the wear deviation vector of each time node is generated. The wear deviation vectors of each time node are combined to obtain the wear deviation matrix of the fused feature sequence.

[0047] The ratio between the reference load curve and the actual load curve is calculated by the sequential wear coding unit to determine the load over-limit situation of each working condition in the fused feature sequence. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence.

[0048] The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

[0049] Preferably, the feature coding layer includes text coding units and numerical coding units;

[0050] Extracting operational status features from vehicle status information includes:

[0051] Driving style features are extracted from vehicle status information based on text encoding units, and environmental temperature and humidity features are extracted from vehicle status information based on numerical encoding units.

[0052] Determine the operating status characteristics based on driving style characteristics and ambient temperature and humidity characteristics;

[0053] The lifetime decoding layer consists of a fully connected layer and an output layer;

[0054] Based on the lifetime decoding layer, regression processing is performed on the multi-dimensional state features to output the target tire's target remaining lifetime information, including:

[0055] Based on the fully connected layer, the multidimensional state features are compressed to obtain a one-dimensional life feature vector of the target tire.

[0056] Based on the output layer, an activation function is applied to the one-dimensional lifespan feature vector to predict the remaining lifespan probability distribution of the target tire at each future time point. The time point corresponding to the maximum remaining lifespan probability is determined as the target remaining lifespan information of the target tire.

[0057] A tire life prediction system based on big data, applicable to the aforementioned tire life prediction method based on big data, includes:

[0058] The data acquisition module is used to acquire multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire;

[0059] The feature extraction module is used to clean and extract features from multi-source operational data to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, the wear trend sequence and load evaluation sequence of the target tire are constructed.

[0060] The feature fusion module is used to compare the wear trend sequence and the load assessment sequence according to the time sequence position of the timestamp in the reference tire life database to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism.

[0061] The average pooling module is used to perform global average pooling on the second historical monitoring data sequence from multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is a historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types;

[0062] The feature concatenation module is used to concatenate pooling features, attention weight features, correlation features, and fused feature sequences to obtain concatenated features. The concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features.

[0063] The life prediction module is used to extract the operating state features of the vehicle state information based on the mask features, obtain multi-dimensional state features, perform regression processing on the multi-dimensional state features, and output the target remaining life information of the target tire; among which, the target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted probability of tire blowout.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] This invention, through the collection and fusion of multi-source data and combined with an attention mechanism, can accurately predict the remaining life of a target tire. By analyzing wear deviations and overload conditions at different time points, it helps to provide more accurate and personalized life predictions. Moreover, by comprehensively analyzing multiple data sources, it provides a multi-dimensional and multi-angle analysis framework that can better reflect the actual working conditions of the tire. This fusion of multi-source data improves the accuracy and reliability of the prediction model.

[0066] This invention utilizes big data analytics and attention mechanisms to dynamically adjust the prediction model, making it highly adaptable to different operating conditions and time points. In particular, by comparing wear characteristic sequences and load characteristic sequences, the model can adjust and update prediction results in real time, avoiding errors caused by static models. Furthermore, by accurately predicting the remaining life of tires, users can take necessary maintenance or replacement measures in a timely manner, avoiding malfunctions or safety hazards caused by insufficient tire life prediction, thereby improving the operational safety and efficiency of vehicles. Attached Figure Description

[0067] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0069] In the diagram: 1. Data acquisition module; 2. Feature extraction module; 3. Feature fusion module; 4. Average pooling module; 5. Feature concatenation module; 6. Lifetime prediction module. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1, please refer to Figure 1 This invention provides a technical solution: a tire life prediction method based on big data, comprising:

[0072] S1. Obtain multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire;

[0073] S2. Clean and extract features from multi-source operating data to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, construct the wear trend sequence and load evaluation sequence of the target tire.

[0074] S3. According to the time sequence position of the timestamp in the reference tire life database, the wear trend sequence and the load assessment sequence are compared to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism.

[0075] S4. Perform global average pooling on the second historical monitoring data sequence from multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is a historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types;

[0076] S5. The pooling features, attention weight features, correlation features and fusion feature sequences are concatenated to obtain concatenated features. The concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features.

[0077] S6. Based on the mask feature, the vehicle state information is extracted to obtain multi-dimensional state features. The multi-dimensional state features are then subjected to regression processing to output the target tire's target remaining life information. The target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted tire blowout probability.

[0078] It should be noted that various operational data of the target tire are obtained, such as: multi-source operational data including vehicle speed, acceleration, braking frequency, road conditions, and temperature; reference tire life databases from historically used tires, recording their life data (such as tire wear time, mileage, and blowout occurrences); sample characteristics including tire manufacturing date, model, and usage environment; for example, assuming a truck, data such as the truck's speed, load, and temperature under different road conditions are recorded. In addition, life data of tires of the same type as the truck under similar environments are also obtained.

[0079] The collected multi-source data is cleaned to remove noise or invalid data, and useful features are extracted. These features include: wear feature sequences, such as the surface wear of tires, which may be obtained by periodically inspecting the tire surface or using data collected by sensors; load feature sequences, such as the load borne by the tires, which may be obtained through vehicle load sensors or GPS data. For example, a truck tire may be affected by different loads and different road conditions over a period of time, and the wear and load information of the tire at different times can be obtained after data cleaning.

[0080] Based on the extracted wear and load characteristics, a wear trend and load assessment sequence for the target tire are constructed. The wear trend sequence can represent the tire's wear over time or mileage, while the load assessment sequence represents the stress changes of the tire under different loads. For example, by analyzing historical data, it is found that tires wear faster under long-term high loads, so the future wear of the tire can be predicted based on these trends.

[0081] Following a chronological order, the wear trend and load assessment sequence of the target tire are compared with corresponding data in a reference tire life database. Through this comparison, a fused feature sequence is obtained, which combines all the key information of wear and load. For example, by comparing the wear data of truck tires with historical tire data under similar conditions, a comprehensive wear trend is obtained, which can better predict the actual service life of the tire.

[0082] Through the attention mechanism, it can automatically learn which features are more important at different monitoring points (such as different positions of the tire); in this way, by weighting different features, it can more accurately capture information that has a greater impact on the prediction results; for example, it may identify that the left front side of the tire wears faster because this position often bears a higher load; therefore, the data at this position will be given higher weight.

[0083] Global average pooling is performed on multiple historical monitoring data sequences (i.e., data from different time periods) to obtain a pooled feature, which represents the overall trend over different monitoring periods. Pooling is used here to reduce data redundancy and extract representative features. For example, wear data from different time periods, after pooling, yields an overall trend of wear throughout the entire life cycle.

[0084] Pooling features, attention weight features, correlation features, and fused feature sequences are concatenated together to form a new concatenated feature. These concatenated features are then input into the lifetime prediction model for further processing to generate a mask feature, which masks some unimportant features and focuses on processing key prediction information. For example, wear trends, load features, and historical data are fused to obtain a multi-dimensional feature. After being processed by the masking network, this feature ignores unimportant data and focuses on the core information that affects lifetime prediction.

[0085] In one alternative embodiment, the multi-source operational data includes wear monitoring data, mileage data, road condition data, and load history data;

[0086] Acquire multi-source operational data and vehicle status information for the target tire, including:

[0087] Acquire initial sensing data of the target tire, including tread depth, tire temperature, driving speed and ground reaction force for multiple monitoring cycles;

[0088] Outlier removal and missing value imputation are performed on the initial sensor data to obtain multi-source operational data;

[0089] Obtain driving behavior information, environmental climate information, and maintenance record information of the vehicle to which the target tire belongs;

[0090] According to the preset driving behavior classification rules, the driving behavior information is tagged to obtain the vehicle's driving style characteristic information.

[0091] The temperature and humidity index of the environmental climate information is calculated to obtain the temperature and humidity characteristic information of the environmental climate information. The maintenance record information and the temperature and humidity characteristic information are summarized to obtain the environmental maintenance characteristic information of the vehicle.

[0092] Vehicle status information is obtained based on driving style characteristics and environmental maintenance characteristics.

[0093] It should be noted that multi-source operational data refers to data from different sources regarding the target tire and vehicle's operating conditions. This data helps to comprehensively understand the tire's wear condition and the vehicle's driving conditions. Common data include: Wear monitoring data: tire wear condition, monitored by sensors to track changes in tire surface depth; Mileage data: total mileage traveled by the vehicle, helping to understand the tire's lifespan; Road condition data: the impact of different road surface types (such as highways, dirt roads, potholes, etc.) on tires; Load history data: the load carried by the vehicle during driving, such as changes in load. For example, suppose a truck travels 50,000 kilometers under different road conditions, including highway driving, mountain dirt roads, and urban roads, and the wear monitoring sensor records the tire wear depth data.

[0094] Initial sensor data provides crucial information about the initial condition of the tires and typically includes: tread depth: measures changes in the depth of the tire surface, indicating the degree of tire wear; tire temperature: monitors the tire's temperature during driving; excessively high temperatures may accelerate wear; vehicle speed: the speed at which the vehicle is traveling, as wear patterns differ at different speeds; and ground reaction force: the reaction force exerted by the ground on the tires, influencing tire wear patterns. For example, for the same vehicle, initial sensor data might record the tire temperature at start-up, the maximum speed achieved, and changes in tread depth under different road conditions.

[0095] The initial sensor data undergoes cleaning, including outlier removal and missing value imputation: Outlier removal: For example, if the data record shows a tire temperature of 500°C, this is obviously erroneous data and should be removed; Missing value imputation: Some data may be missing due to sensor malfunction or other reasons. In this case, these missing values ​​can be filled in using interpolation methods to ensure data integrity; For example, if tire temperature data is missing for a certain period, it can be filled in using an interpolation algorithm based on the temperature values ​​before and after the period.

[0096] Vehicle status information includes driving behavior, environmental climate, and maintenance records: Driving behavior information records the driver's driving style, such as rapid acceleration, hard braking, and frequent lane changes, all of which can affect tire wear; environmental climate information includes temperature, humidity, and precipitation, which affect the tire wear rate; for example, tires may be more easily damaged in wet environments; maintenance record information includes the tire's maintenance history, such as tire replacement, inspection, and inflation, which can help determine the tire's health condition; for example, if a driver frequently brakes hard and the vehicle is frequently driven on slippery roads, this information can be recorded as part of the vehicle's status.

[0097] Based on preset driving behavior classification rules, driving behaviors are tagged; for example: aggressive driving: frequent rapid acceleration and braking; mild driving: smooth acceleration and braking; for example: if a car drives smoothly on the highway, the recorded driving behavior will be marked as mild driving, while if the car frequently accelerates or brakes suddenly, it will be marked as aggressive driving.

[0098] Among environmental climate information, temperature and humidity are the most critical factors. By calculating the temperature and humidity index, the impact of these two factors on tires can be comprehensively considered. By calculating the combined effect of temperature and humidity on tires, a characteristic value representing environmental conditions can be obtained. For example, in a hot and humid environment in summer, the temperature and humidity index may be high, indicating that tires may be more prone to overheating or getting damp in such an environment, leading to faster wear.

[0099] By combining vehicle maintenance records with environmental and climatic characteristics, we can obtain the vehicle's environmental maintenance characteristics. For example, in a humid environment, regular tire checks and replacements by the owner may reduce the risk of tire damage. For instance, if a vehicle in a humid area checks its tires every six months and performs appropriate tire maintenance, such records, along with environmental characteristics, will generate the vehicle's environmental maintenance characteristics.

[0100] By combining driving style characteristics and environmental maintenance characteristics, comprehensive vehicle status information can be obtained. This information will be used for subsequent tire life prediction. For example, based on the vehicle's aggressive driving behavior and wet and slippery environment characteristics, accelerated tire wear rate can be predicted, and suggestions for early tire replacement can be provided.

[0101] In an optional embodiment, the wear characteristic sequence includes wear depth value, mileage value, and road surface roughness value; the load characteristic sequence includes vertical load value, lateral load value, and tire pressure fluctuation value; the wear trend sequence includes a baseline wear curve and an actual wear curve; and the load evaluation sequence includes a working condition load sequence and a structural stress sequence.

[0102] Based on the reference tire life database, wear characteristic sequence, and load characteristic sequence, a wear trend sequence and load assessment sequence for the target tire are constructed, including:

[0103] According to the time sequence position of the timestamp in the reference tire life database, the load values ​​under each working condition in the actual load curve and the impact load values ​​of each abnormal high load working condition are sorted and combined to obtain the working condition load sequence.

[0104] The estimated structural stress values ​​for each working condition in the actual load curve and the peak structural stress values ​​for each abnormally high load condition are sorted and combined to obtain the structural stress sequence.

[0105] The wear curve and the actual wear curve are differentially calculated to determine the wear deviation at each time node in the fused feature sequence. Based on the wear deviation at each time node, the wear deviation vector at each time node is generated. The wear deviation vectors at each time node are combined to obtain the wear deviation matrix of the fused feature sequence.

[0106] By calculating the ratio between the baseline load curve and the actual load curve, the load over-limit situation of each working condition in the fused feature sequence is determined. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence.

[0107] The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

[0108] It should be noted that the wear characteristic sequence includes wear depth value, mileage value, and road surface roughness value. These data help to understand the wear of tires under different conditions. Wear depth value: the depth change of the tire surface due to friction, usually monitored in real time by sensors. Mileage value: the total mileage driven by the vehicle since the tires were installed. Road surface roughness value: indicates the unevenness of the road surface, affecting the tire wear rate. For example, after a truck has driven 10,000 kilometers on urban roads, sensors record the change in tire wear depth, and also record the road surface roughness data. These data together form the wear characteristic sequence.

[0109] The load characteristic sequence includes vertical load values, lateral load values, and tire pressure fluctuation values. These data help to understand how tires perform under different load conditions. Vertical load value: the pressure exerted on the tires by the vehicle's vertical downward weight; Lateral load value: the lateral pressure exerted on the tires during cornering; Tire pressure fluctuation value: changes in tire pressure, which may be caused by temperature or load variations. For example, on a truck transporting goods, when the load is heavy, the vertical load value of the tires will increase, the lateral load value will also increase during sharp cornering, and the tire pressure fluctuation value will fluctuate due to the increase in tire temperature.

[0110] Wear trend sequence includes a baseline wear curve and an actual wear curve. These two curves help to understand the changing trend of tire wear. The baseline wear curve is a theoretical wear pattern derived from tests or experiments. The actual wear curve is tire wear data obtained through actual measurements. For example, the baseline wear curve may represent the tire wear rate under ideal conditions, while the actual wear curve may vary due to factors such as road surface and driving habits.

[0111] The load assessment sequence includes the operating condition load sequence and the structural stress sequence. These sequences are used to evaluate the load conditions of the tire under different operating conditions. The operating condition load sequence records the load changes of the vehicle under different road conditions, including abnormally high load conditions (e.g., sudden braking or sharp turning). The structural stress sequence reflects the structural stress of the tire under different load conditions and is usually obtained through simulation calculations of the tire's materials. For example, when a truck brakes suddenly, the load on the tire increases instantaneously, and this condition is recorded as part of the operating condition load sequence. At the same time, the stress state of the tire's material structure at that moment is also recorded as part of the structural stress sequence.

[0112] By using timestamps, the load values ​​for each working condition in the actual load curve are sorted and combined with the impact load values ​​of abnormally high load conditions to obtain a complete working condition load sequence; the goal of this step is to accurately reflect the load changes of the tire under different working conditions. The estimated structural stress values ​​for each working condition in the actual load curve are combined with the peak structural stress values ​​under abnormally high load conditions to obtain a structural stress sequence. The reference wear curve and the actual wear curve are differentially calculated to obtain the wear deviation at each time point. These deviation values ​​are used to generate a wear deviation matrix to describe the wear changes of the tire. By calculating the ratio, the reference load curve and the actual load curve are compared to generate load over-limit conditions, and the load change is described by a load coefficient vector, which is finally combined into a load coefficient matrix. The wear deviation matrix, the load coefficient matrix, and the structural stress sequence are spliced ​​together to obtain the final wear load feature matrix.

[0113] In an optional embodiment, generating a wear deviation vector for each time point based on the wear deviation at each time point includes:

[0114] For any given time point, construct the initial wear vector for that time point; the initial wear vector includes normal wear interval positions, abnormal wear interval positions, multiple levels of mild wear indicator positions, and multiple levels of severe wear indicator positions.

[0115] If the wear deviation at a time point is greater than a preset threshold, the values ​​of the abnormal wear interval position in the initial wear vector and the values ​​of the severe wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value to generate the wear deviation vector at the time point.

[0116] If the wear deviation at a time point is less than or equal to a preset threshold, the values ​​of the normal wear interval position in the initial wear vector and the values ​​of the mild wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value, thereby generating the wear deviation vector at the time point.

[0117] It should be noted that an initial wear vector is created for each time point. This vector contains several parts: Normal wear range: indicates that the tire wear is within the normal range and there is no significant abnormality; Abnormal wear range: indicates that the wear has exceeded the normal range and is considered abnormal wear; Mild wear indicator: indicates that the wear is in a mild wear state, which usually does not affect the tire's use; Severe wear indicator: indicates that the wear is more serious and may affect the tire's safety and lifespan.

[0118] After calculating the wear deviation value at a certain time point (i.e., the difference from the baseline wear curve), it is determined whether the deviation exceeds a preset threshold: if the deviation is greater than the threshold, it indicates that the tire wear is relatively severe, and the abnormal wear interval is set to the first preset value (which may be a marker indicating severe wear); the heavy wear indicator that matches the deviation level is set to the first preset value (indicating that the wear in this part has become severe enough to require attention).

[0119] The remaining markings (including the normal wear range markings and the light wear markings) are set to the second preset value (usually markings indicating slight or normal wear).

[0120] If the deviation is less than or equal to the threshold, it means that the tire wear is within the normal range or is relatively light. The following will be done: set the normal wear range position to the first preset value (indicating that the wear is within the normal range); set the light wear indicator position that matches the deviation level to the first preset value (indicating that the wear in this part is relatively light and can continue to be used); set other indicator positions (including the abnormal wear range position and the heavy wear indicator position) to the second preset value.

[0121] Suppose a car's tires are measured during its fifth month of driving, and the wear deviation is found to be 1.2 mm, exceeding a preset threshold of 0.8 mm; then:

[0122] Generate an initial wear vector, including: normal wear interval position: assumed to be 0; abnormal wear interval position: assumed to be 0; light wear indicator position: assumed to be 0; heavy wear indicator position: assumed to be 0; Since the wear deviation is greater than the threshold, the values ​​of the abnormal wear interval position and the heavy wear indicator position are set to 1 (first preset value), indicating that the tire wear is abnormal and relatively serious; other indicator positions (normal wear interval position and light wear indicator position) are set to 2 (second preset value), indicating that these wear levels do not meet the conditions of being serious or abnormal; If the wear deviation is 0.5 mm in the measurement of the 6th month, which is lower than the threshold of 0.8 mm, then:

[0123] Generate an initial wear vector, including: normal wear interval bit: its value is 0; abnormal wear interval bit: its value is 0; slight wear indicator bit: its value is 0; heavy wear indicator bit: its value is 0; since the wear deviation is less than or equal to the threshold, the values ​​of the normal wear interval bit and the slight wear indicator bit are set to 1 (first preset value), indicating that the wear is within the normal range and is slight; the other indicator bits (abnormal wear interval bit and heavy wear indicator bit) are set to 2 (second preset value).

[0124] In an optional embodiment, based on the load over-limit situation of each working condition, a load coefficient vector for each working condition is generated, including:

[0125] For any given load condition, construct an initial load vector for that load condition; the initial load vector includes multiple candidate load level identifier bits;

[0126] Among multiple candidate load level identifiers, determine the target load level identifier corresponding to the actual load of the working condition, set the value of the target load level identifier to a first preset value, and set the values ​​of other load level identifiers in the initial load vector to a second preset value to obtain the load coefficient vector of the working condition.

[0127] It should be noted that for each working condition, an initial load vector is created based on the load conditions of that working condition. This vector includes multiple candidate load level identifiers, which typically represent different load levels. For example: low load level identifier: indicates that the load under the working condition is low; medium load level identifier: indicates that the load under the working condition is in the medium range; high load level identifier: indicates that the load under the working condition is high. The initial load vector initializes these identifiers to certain default values, which typically represent an undetermined state.

[0128] For each load case, the actual load of that load case is evaluated and compared with the standards of each candidate load level; based on this comparison, the target load level corresponding to the actual load is determined; for example, assuming that the actual load of the load case is at a medium level, the medium load level flag will be selected as the target load level.

[0129] Once the target load level is determined, the final load factor vector will be generated according to the following rules: the value of the target load level identifier will be set to the first preset value, indicating that the load of this working condition belongs to this level; the values ​​of the other load level identifiers will be set to the second preset value, indicating that these levels do not match the actual load of the working condition.

[0130] In an optional embodiment, based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through an attention mechanism, including:

[0131] The first historical monitoring data sequence is obtained from multiple historical monitoring data sequences, and the type of monitoring data included in the first historical monitoring data sequence is a preset type;

[0132] The tire identifier of each target tire is processed with the monitoring data of each associated monitoring point contained in the first historical monitoring data sequence using the target attention mechanism to obtain the attention weight features of each target tire corresponding to different associated monitoring points.

[0133] The monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points are obtained from the first historical monitoring data sequence; the current associated monitoring point is any one associated monitoring point, and the other associated monitoring points are all associated monitoring points other than the current associated monitoring point.

[0134] Construct a similarity matrix based on the cosine distance between the monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points;

[0135] Based on the largest matrix element in the similarity matrix and the pre-constructed normal distribution, the associated features corresponding to the target tire are obtained.

[0136] It should be noted that a specific first historical monitoring data series is selected from multiple historical monitoring data series. This series includes some specific types of monitoring data; for example, the monitoring data may include tire temperature, pressure, wear, etc.

[0137] For each target tire, the data is processed through a target attention mechanism, which combines the tire's identification and monitoring data from the first historical monitoring data sequence. The goal of this mechanism is to assign different importance weights to different monitoring points. This means that different monitoring data will receive different attention weight characteristics based on their relationship with the target tire. The greater the attention weight, the greater the impact or relevance of the monitoring data on the target tire.

[0138] Data for the current associated monitoring point is obtained from the first historical monitoring data sequence, along with data for all other associated monitoring points. The current associated monitoring point refers to a specific monitoring point that is being monitored, while other associated monitoring points refer to all monitoring points that are different from the current monitoring point.

[0139] To compare the correlation between these monitoring points, the cosine distance between the current monitoring point and other monitoring points is calculated. The cosine distance measures the similarity between two data points; the smaller the value, the stronger the relationship between the two data points. By calculating the cosine distance between all monitoring points, a similarity matrix is ​​constructed, where each element of the matrix represents the degree of similarity between two monitoring points.

[0140] The correlation features of the target tire are extracted based on the maximum value in the similarity matrix (i.e., the two monitoring points with the highest similarity) and a pre-constructed normal distribution. Here, the maximum matrix element represents the two most relevant monitoring points in the monitoring data, and the normal distribution helps to determine the credibility of these related features.

[0141] In one optional embodiment, the lifetime prediction model includes a feature encoding layer, a feature fusion layer, and a lifetime decoding layer;

[0142] Based on mask features, operational state features are extracted from vehicle state information to obtain multi-dimensional state features. Regression processing is then performed on these multi-dimensional state features to output the target tire's remaining lifespan information, including:

[0143] Wear load features are extracted from the fused feature sequence based on the feature coding layer, and operating status features of vehicle status information are extracted.

[0144] The wear load features and operating status features are stitched together based on the feature fusion layer, and the stitched features are subjected to nonlinear transformation to obtain multidimensional status features.

[0145] The target tire's remaining life information is output by performing regression processing on the multidimensional state features based on the life decoding layer.

[0146] It should be noted that the task of the feature encoding layer is to extract two key features from the raw vehicle state information: wear load features, which refer to the degree of tire wear and the load it bears under different conditions; this feature usually involves factors such as the pressure, load, and road conditions experienced by the tires during vehicle operation; and operating state features, which refer to various state information involved in the vehicle's operation, such as speed, temperature, and tire rotation. For example, suppose a truck is driving on different road conditions (such as mountain roads and flat roads), and the load, temperature, speed, and other information of each road segment will be recorded by the vehicle's sensors; the feature encoding layer will extract wear load features (such as the degree of tire wear under high load conditions) and vehicle operating state features (such as driving speed and engine temperature) from this data.

[0147] The task of the feature fusion layer is to integrate the extracted features and perform nonlinear transformations. Specifically, it concatenates wear load features and operating state features, and then performs a nonlinear transformation to obtain a new multidimensional state feature. For example, suppose the following data is extracted in the feature encoding layer: wear load features: tire wear degree, load size; operating state features: current vehicle speed, engine temperature; in the feature fusion layer, these data will be concatenated together, such as: wear load features: [0.3, 0.5, 0.7] (tire wear degree, load size, other data); operating state features: [60km / h, 85°C] (vehicle speed and engine temperature); through nonlinear transformations (such as through the activation function of a neural network), these concatenated data will be transformed into more complex and representative multidimensional state features, which can capture deeper patterns and regularities.

[0148] The task of the lifespan decoding layer is to perform regression processing based on the fused multidimensional state features, that is, to predict the remaining lifespan information of the target tire. This layer uses the previously extracted and fused features to predict the remaining lifespan of the tire, that is, how much time or kilometers the target tire can still be used. For example, suppose that after the feature fusion layer, the multidimensional state features of the target tire are obtained, which include comprehensive information about load, speed, temperature, etc. The lifespan decoding layer will calculate based on this information to predict the remaining lifespan of the tire. For example, the model may tell you that, based on the current state, the remaining lifespan of the target tire is 3,000 kilometers.

[0149] In an optional embodiment, the feature coding layer includes a sequence wear coding unit; the fused feature sequence includes a reference wear curve, an actual wear curve, a reference load curve, and an actual load curve;

[0150] Wear load features are extracted from the fused feature sequence based on the feature coding layer, including:

[0151] The wear deviation of each time node in the fused feature sequence is determined by performing differential calculation on the reference wear curve and the actual wear curve through the sequence wear coding unit. Based on the wear deviation of each time node, the wear deviation vector of each time node is generated. The wear deviation vectors of each time node are combined to obtain the wear deviation matrix of the fused feature sequence.

[0152] The ratio between the reference load curve and the actual load curve is calculated by the sequential wear coding unit to determine the load over-limit situation of each working condition in the fused feature sequence. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence.

[0153] The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

[0154] It should be noted that the main task of the feature coding layer is to extract and process wear-related features from the vehicle's historical data; the sequence wear coding unit is a key part of the feature coding layer, and its role is to process the wear features and extract the features through certain calculations.

[0155] The fused feature sequence consists of the following curves: Baseline wear curve: This is typically a standard reference curve representing the wear of tires or vehicle components under ideal conditions (e.g., normal driving, reasonable load); Actual wear curve: This represents the actual wear data of tires or components during actual vehicle operation, which may be affected by different road conditions, loads, and other factors; Baseline load curve: Also a standard reference curve, representing the load variation of the vehicle under ideal conditions; Actual load curve: The actual load variation of the vehicle during actual operation. These curves represent two important factors: wear and load. The model analyzes the deviation and over-limit situations of wear and load during actual vehicle operation by comparing the baseline curve and the actual curve.

[0156] It is calculated by comparing the baseline wear curve with the actual wear curve; it represents the wear difference of a vehicle at different time points or under different operating conditions; for example, the actual wear at a certain time point may be higher than the ideal wear curve by a certain value, indicating that the tire wears faster, possibly due to poor driving conditions or other reasons; for example, suppose the baseline wear curve shows that the tire should wear 10mm after a certain mileage, but the actual wear curve shows that the tire has already worn 12mm; this indicates that at this time point, the wear deviation is 2mm;

[0157] By comparing the baseline load curve and the actual load curve, it can be calculated whether the load exceeds the design safety range. Overload refers to the situation where the load borne by the tire exceeds the design load under certain working conditions, which may accelerate tire wear. For example, suppose the baseline load curve indicates that the tire should bear a maximum load of 800 kg, while the actual load curve shows that the tire bears a load of 1000 kg at a certain moment, which means that the load exceeds the limit.

[0158] After calculating the wear deviation at each time point, these deviation values ​​are combined into a matrix. The wear deviation vector at each time point represents the difference between that time point and the baseline wear curve. By using a matrix, all wear deviation information throughout the process is integrated to form a feature set that can be used for subsequent analysis. For example, assuming there are 5 time points, and the wear deviation at each time point is [2mm, 1.5mm, 3mm, 1mm, 2.5mm], these deviation values ​​will be combined into a wear deviation matrix for the model to learn further.

[0159] By calculating the load over-limit situation for each working condition, the load coefficient for each working condition (representing the relative degree of load over-limit) can be obtained. These load coefficients are also combined into a matrix for subsequent analysis. For example, suppose there are 4 working conditions, and the load coefficients for each working condition are [1.2, 0.9, 1.5, 1.1]. These coefficients represent the load over-limit situation under each working condition.

[0160] After obtaining the wear deviation matrix and load coefficient matrix, an additional feature needs to be introduced, namely the structural stress sequence. The structural stress sequence refers to the stress information that the vehicle or tire structure is subjected to under different working conditions. This sequence is usually collected by sensors and reflects the changes in structural stress during vehicle operation. By splicing the wear deviation matrix, load coefficient matrix and structural stress sequence, a composite feature matrix is ​​finally obtained, namely the wear feature matrix. This matrix contains comprehensive information about wear, load and structural stress.

[0161] In an optional embodiment, the feature encoding layer includes text encoding units and numerical encoding units;

[0162] Extracting operational status features from vehicle status information includes:

[0163] Driving style features are extracted from vehicle status information based on text encoding units, and environmental temperature and humidity features are extracted from vehicle status information based on numerical encoding units.

[0164] Determine the operating status characteristics based on driving style characteristics and ambient temperature and humidity characteristics;

[0165] The lifetime decoding layer consists of a fully connected layer and an output layer;

[0166] Based on the lifetime decoding layer, regression processing is performed on the multi-dimensional state features to output the target tire's target remaining lifetime information, including:

[0167] Based on the fully connected layer, the multidimensional state features are compressed to obtain a one-dimensional life feature vector of the target tire.

[0168] Based on the output layer, an activation function is applied to the one-dimensional lifespan feature vector to predict the remaining lifespan probability distribution of the target tire at each future time point. The time point corresponding to the maximum remaining lifespan probability is determined as the target remaining lifespan information of the target tire.

[0169] It should be noted that the purpose of the feature encoding layer is to extract valuable features for prediction from the vehicle's state information; this layer includes two types of encoding units: text encoding units and numerical encoding units, which are used to process different types of data respectively;

[0170] The text encoding unit is used to process text data describing driving styles. Driving styles are usually obtained through vehicle driving behavior records (such as driving habits, acceleration, braking, steering, etc.). These data are usually unstructured text descriptions. The task of the text encoding unit is to convert these descriptions into numerical features that can be used for analysis. For example, suppose there is data recording driving styles, such as smooth driving, rapid acceleration, frequent braking, etc. The text encoding unit will analyze these descriptions to extract features related to driving styles. For example, smooth driving corresponds to low acceleration and less braking, while rapid acceleration may correspond to high acceleration and frequent braking.

[0171] The numerical coding unit is used to process the environmental temperature and humidity features in the vehicle status information. Environmental temperature and humidity data are usually obtained through sensors or external weather data and are structured data. The numerical coding unit converts these data into features that can be further analyzed. For example, suppose the environmental temperature and humidity data are as follows: temperature is 30°C and humidity is 80%. The numerical coding unit will process these data into numerical features and may extract the effects of temperature and humidity on tire wear, such as high temperature and high humidity may accelerate tire aging.

[0172] Based on driving style features obtained from text encoding units and environmental temperature and humidity features obtained from numerical encoding units, the model combines these two types of information to form operating state features. These operating state features will serve as the basis for subsequent predictions. For example, assuming the driving style is rapid acceleration and the ambient temperature is 30°C and the humidity is 80%, the model may conclude that the tires are in a high-load, high-temperature, and high-humidity environment, and predict that the tires will wear faster, thus affecting the tire life.

[0173] The lifetime decoding layer is mainly responsible for predicting the remaining life of the target tire based on multi-dimensional operating state characteristics; this layer consists of two parts: a fully connected layer and an output layer.

[0174] The role of the fully connected layer is to compress the extracted multidimensional state features, that is, to integrate multiple features into a more concise representation, which facilitates subsequent processing and analysis. After passing through the fully connected layer, a simplified one-dimensional lifetime feature vector is obtained, which represents a comprehensive feature of the target tire. For example, assuming that the vehicle state information has multiple features, such as driving style features (e.g., rapid acceleration), temperature and humidity features (e.g., high temperature, high humidity), vehicle speed, road conditions, etc., after being processed by the fully connected layer, this information will be compressed into a simplified lifetime feature vector, such as [0.8, 1.2, 0.5]. These features are convenient for subsequent activation function processing.

[0175] The task of the output layer is to process the one-dimensional lifespan feature vector through an activation function to predict the probability distribution of the remaining lifespan of the target tire at various future time points. Specifically, the output layer performs activation processing based on the one-dimensional lifespan feature vector to generate a probability distribution of the remaining lifespan, indicating the possible remaining lifespan of the tire at different time points. For example, suppose the output layer predicts the following probability distribution of the remaining lifespan of the target tire: 30% remaining lifespan in the next month; 50% remaining lifespan in the next two months; and 20% remaining lifespan in the next three months. By analyzing this probability distribution, the model can determine the time point corresponding to the highest probability, which is the target remaining lifespan of the tire. Two months later is the time point with the highest probability of remaining lifespan, so this time point will be determined as the remaining lifespan of the target tire.

[0176] Example 2, please refer to Figure 2 This invention provides a technical solution: a tire life prediction system based on big data, applicable to the aforementioned tire life prediction method based on big data, comprising:

[0177] Data acquisition module 1 is used to acquire multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire;

[0178] Feature extraction module 2 is used to clean and extract features from multi-source operating data to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, the wear trend sequence and load evaluation sequence of the target tire are constructed.

[0179] Feature fusion module 3 is used to compare the wear trend sequence and the load assessment sequence according to the time sequence position of the timestamp in the reference tire life database to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism.

[0180] The average pooling module 4 is used to perform global average pooling on the second historical monitoring data sequence among multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is a historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types;

[0181] Feature concatenation module 5 is used to concatenate pooling features, attention weight features, correlation features and fusion feature sequences to obtain concatenated features. The concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features.

[0182] The life prediction module 6 is used to extract the operating state features of the vehicle state information based on the mask features, obtain multi-dimensional state features, perform regression processing on the multi-dimensional state features, and output the target remaining life information of the target tire; wherein, the target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted probability of tire blowout.

[0183] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A tire life prediction method based on big data, characterized in that, include: Acquire multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire; Multi-source operational data are cleaned and feature extracted to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, the wear trend sequence and load assessment sequence of the target tire are constructed. According to the time sequence position of the timestamp in the reference tire life database, the wear trend sequence and the load assessment sequence are compared to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism. Global average pooling is performed on the second historical monitoring data sequence from multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is the historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types; The pooling features, attention weight features, correlation features, and fusion feature sequences are concatenated to obtain concatenated features. These concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features. Based on the mask feature, the vehicle state information is extracted to obtain multi-dimensional state features. The multi-dimensional state features are then subjected to regression processing to output the target tire's target remaining life information. The target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted probability of tire blowout.

2. The tire life prediction method based on big data according to claim 1, characterized in that, Multi-source operational data includes wear monitoring data, mileage data, road condition data, and load history data; Acquire multi-source operational data and vehicle status information for the target tire, including: Acquire initial sensing data of the target tire, including tread depth, tire temperature, driving speed and ground reaction force for multiple monitoring cycles; Outlier removal and missing value imputation are performed on the initial sensor data to obtain multi-source operational data; Obtain driving behavior information, environmental climate information, and maintenance record information of the vehicle to which the target tire belongs; According to the preset driving behavior classification rules, the driving behavior information is tagged to obtain the vehicle's driving style characteristic information. The temperature and humidity index of the environmental climate information is calculated to obtain the temperature and humidity characteristic information of the environmental climate information. The maintenance record information and the temperature and humidity characteristic information are summarized to obtain the environmental maintenance characteristic information of the vehicle. Vehicle status information is obtained based on driving style characteristics and environmental maintenance characteristics.

3. The tire life prediction method based on big data according to claim 2, characterized in that, The wear characteristic sequence includes wear depth value, mileage value, and road surface roughness value; The load characteristic sequence includes vertical load values, lateral load values, and tire pressure fluctuation values; the wear trend sequence includes the baseline wear curve and the actual wear curve; the load evaluation sequence includes the working condition load sequence and the structural stress sequence. Based on the reference tire life database, wear characteristic sequence, and load characteristic sequence, a wear trend sequence and load assessment sequence for the target tire are constructed, including: According to the time sequence position of the timestamp in the reference tire life database, the load values ​​under each working condition in the actual load curve and the impact load values ​​of each abnormal high load working condition are sorted and combined to obtain the working condition load sequence. The estimated structural stress values ​​for each working condition in the actual load curve and the peak structural stress values ​​for each abnormally high load condition are sorted and combined to obtain the structural stress sequence. The wear curve and the actual wear curve are differentially calculated to determine the wear deviation at each time node in the fused feature sequence. Based on the wear deviation at each time node, the wear deviation vector at each time node is generated. The wear deviation vectors at each time node are combined to obtain the wear deviation matrix of the fused feature sequence. By calculating the ratio between the baseline load curve and the actual load curve, the load over-limit situation of each working condition in the fused feature sequence is determined. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence. The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

4. The tire life prediction method based on big data according to claim 3, characterized in that, Based on the wear deviation at each time point, a wear deviation vector for each time point is generated, including: For any given time point, construct the initial wear vector for that time point; the initial wear vector includes normal wear interval positions, abnormal wear interval positions, multiple levels of mild wear indicator positions, and multiple levels of severe wear indicator positions. If the wear deviation at a time point is greater than a preset threshold, the values ​​of the abnormal wear interval position in the initial wear vector and the values ​​of the severe wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value to generate the wear deviation vector at the time point. If the wear deviation at a time point is less than or equal to a preset threshold, the values ​​of the normal wear interval position in the initial wear vector and the values ​​of the mild wear flag position that matches the wear deviation level are set to the first preset value, and the values ​​of other flag positions in the initial wear vector are set to the second preset value, thereby generating the wear deviation vector at the time point.

5. The tire life prediction method based on big data according to claim 4, characterized in that, Based on the load exceedance conditions under each working condition, a load coefficient vector for each working condition is generated, including: For any given load condition, construct an initial load vector for that load condition; the initial load vector includes multiple candidate load level identifier bits; Among multiple candidate load level identifiers, determine the target load level identifier corresponding to the actual load of the working condition, set the value of the target load level identifier to a first preset value, and set the values ​​of other load level identifiers in the initial load vector to a second preset value to obtain the load coefficient vector of the working condition.

6. The tire life prediction method based on big data according to claim 5, characterized in that, Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through an attention mechanism, including: The first historical monitoring data sequence is obtained from multiple historical monitoring data sequences, and the type of monitoring data included in the first historical monitoring data sequence is a preset type; The tire identifier of each target tire is processed with the monitoring data of each associated monitoring point contained in the first historical monitoring data sequence using the target attention mechanism to obtain the attention weight features of each target tire corresponding to different associated monitoring points. The monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points are obtained from the first historical monitoring data sequence; the current associated monitoring point is any one associated monitoring point, and the other associated monitoring points are all associated monitoring points other than the current associated monitoring point. Construct a similarity matrix based on the cosine distance between the monitoring data of the current associated monitoring point and the monitoring data of all other associated monitoring points; Based on the largest matrix element in the similarity matrix and the pre-constructed normal distribution, the associated features corresponding to the target tire are obtained.

7. The tire life prediction method based on big data according to claim 6, characterized in that, The lifetime prediction model includes a feature encoding layer, a feature fusion layer, and a lifetime decoding layer; Based on mask features, operational state features are extracted from vehicle state information to obtain multi-dimensional state features. Regression processing is then performed on these multi-dimensional state features to output the target tire's remaining lifespan information, including: Wear load features are extracted from the fused feature sequence based on the feature coding layer, and operating status features of vehicle status information are extracted. The wear load features and operating status features are stitched together based on the feature fusion layer, and the stitched features are subjected to nonlinear transformation to obtain multidimensional status features. The target tire's remaining life information is output by performing regression processing on the multidimensional state features based on the life decoding layer.

8. The tire life prediction method based on big data according to claim 7, characterized in that, The feature coding layer includes a sequence wear coding unit; the fused feature sequence includes a reference wear curve, an actual wear curve, a reference load curve, and an actual load curve. Wear load features are extracted from the fused feature sequence based on the feature coding layer, including: The wear deviation of each time node in the fused feature sequence is determined by performing differential calculation on the reference wear curve and the actual wear curve through the sequence wear coding unit. Based on the wear deviation of each time node, the wear deviation vector of each time node is generated. The wear deviation vectors of each time node are combined to obtain the wear deviation matrix of the fused feature sequence. The ratio between the reference load curve and the actual load curve is calculated by the sequential wear coding unit to determine the load over-limit situation of each working condition in the fused feature sequence. Based on the load over-limit situation of each working condition, the load coefficient vector of each working condition is generated. The load coefficient vectors of each working condition are combined to obtain the load coefficient matrix of the fused feature sequence. The wear deviation matrix, load coefficient matrix, and structural stress sequence are spliced ​​together to obtain the wear characteristic matrix, which is then used to determine the wear load characteristics.

9. The tire life prediction method based on big data according to claim 8, characterized in that, The feature coding layer includes text coding units and numerical coding units; Extracting operational status features from vehicle status information includes: Driving style features are extracted from vehicle status information based on text encoding units, and environmental temperature and humidity features are extracted from vehicle status information based on numerical encoding units. Determine the operating status characteristics based on driving style characteristics and ambient temperature and humidity characteristics; The lifetime decoding layer consists of a fully connected layer and an output layer; Based on the lifetime decoding layer, regression processing is performed on the multi-dimensional state features to output the target tire's target remaining lifetime information, including: Based on the fully connected layer, the multidimensional state features are compressed to obtain a one-dimensional life feature vector of the target tire. Based on the output layer, an activation function is applied to the one-dimensional lifespan feature vector to predict the remaining lifespan probability distribution of the target tire at each future time point. The time point corresponding to the maximum remaining lifespan probability is determined as the target remaining lifespan information of the target tire.

10. A tire life prediction system based on big data, applicable to the tire life prediction method based on big data as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire multi-source operating data and vehicle status information of the target tire, as well as the reference tire life database and sample features corresponding to the target tire; The feature extraction module is used to clean and extract features from multi-source operational data to obtain the wear feature sequence and load feature sequence of the target tire. Based on the reference tire life database, wear feature sequence and load feature sequence, the wear trend sequence and load evaluation sequence of the target tire are constructed. The feature fusion module is used to compare the wear trend sequence and the load assessment sequence according to the time sequence position of the timestamp in the reference tire life database to obtain the fused feature sequence. Based on the fused feature sequence and sample features, the attention weight features of each associated monitoring point and the associated features corresponding to the target tire are obtained through the attention mechanism. The average pooling module is used to perform global average pooling on the second historical monitoring data sequence from multiple historical monitoring data sequences to obtain pooling features; wherein, the second historical monitoring data sequence is a historical monitoring data sequence other than the first historical monitoring data sequence; the monitoring data types included in the first historical monitoring data sequence are preset types; The feature concatenation module is used to concatenate pooling features, attention weight features, correlation features, and fused feature sequences to obtain concatenated features. The concatenated features are then input into the masking network of the lifetime prediction model to obtain masking features. The life prediction module is used to extract the operating state features of the vehicle state information based on the mask features, obtain multi-dimensional state features, perform regression processing on the multi-dimensional state features, and output the target remaining life information of the target tire; among which, the target remaining life information includes the predicted remaining safe mileage, the predicted wear depth, and the predicted probability of tire blowout.

Images