CN-121996982-A - Tire pattern depth estimation method based on physical information fusion network

CN121996982ACN 121996982 ACN121996982 ACN 121996982ACN-121996982-A

Abstract

The invention relates to a tire pattern depth estimation method based on a physical information fusion network, and belongs to the technical field of intelligent traffic systems and active safety of vehicles. The method comprises the steps of firstly collecting triaxial acceleration data under different working conditions through a built-in sensor of a tire, then carrying out zero phase shift filtering, phase alignment and resampling pretreatment on signals to form standard waveform data, then constructing a space-time feature extraction network, extracting time sequence features through a convolution layer and a circulation layer, fusing normalized load and speed working condition information, further introducing a physical constraint module based on a tire stiffness mechanism, predicting pattern depth and vertical deformation by utilizing a learnable parameter, and reversely pushing a theoretical load to establish mechanical balance constraint, finally adopting a mixed loss function to synchronously optimize network parameters and physical parameters, and outputting wear state grade and confidence coefficient through a probability statistical method. The invention effectively improves the estimation precision and the physical interpretability of the model under the complex variable load working condition.

Inventors

ZHANG XIAOLONG
LIU YANGYANG
LI ZIJUN
CHENG XUANHAO
HUANG JINGHAN
SONG NA
DENG QIN
WU DONGSONG
MU YANLIANG

Assignees

福州大学

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (10)

1. The tire pattern depth estimation method based on the physical information fusion network is characterized by comprising the following steps of: S1, acquiring triaxial acceleration data of tires with different wear degrees under different load and speed working conditions by adopting an intelligent tire system with triaxial acceleration sensors attached to the inner liner of the tire; S2, performing zero phase shift filtering denoising on the acquired acceleration data, aligning phases of signals of different channels by using a characteristic point matching method, segmenting continuous signals into single-circle samples according to the rotation period characteristics of the tire, and unifying each circle of samples into standard waveform data with fixed length through resampling; Step S3, building a space-time feature extraction neural network, extracting time sequence features of standard waveform data by utilizing a convolution layer and a circulation layer, introducing a physical working condition fusion mechanism, expanding normalized load and speed data into vectors, and splicing the normalized load and speed data with the waveform features in a feature extraction stage to obtain deep features containing working condition information; S4, constructing a physical constraint module based on a tire stiffness mechanism, setting a learnable physical parameter, predicting the pattern depth and the vertical deformation of the tire by using deep features, and reversely calculating a theoretical load by using a predicted value according to a mechanical balance formula; And S5, building a mixed loss function containing data errors and physical consistency errors, training the network, synchronously updating the network weight and physical parameters, processing the continuous predicted value of the network by using a probability statistical method, and outputting the abrasion state grade and the confidence of the tire.
2. The tire pattern depth estimation method based on the physical information fusion network according to claim 1 is characterized in that in the step S1, the data acquisition comprises the steps of selecting an all-steel meridian heavy commercial vehicle tire, pasting and installing a triaxial acceleration sensor at the center position of a tire liner crown, acquiring acceleration vibration signals of the tire along tangential and radial directions in real time through the sensor, synchronously recording real-time vertical load L and running speed V of the tire, and constructing orthogonal experimental conditions covering a full life cycle abrasion state and a combination of multistage loads and speeds.
3. The method for estimating tire tread depth based on physical information fusion network as set forth in claim 1, wherein in step S2, the signal preprocessing comprises constructing a fourth-order Butterworth low-pass filter, denoising the signal with zero phase shift by adopting a bidirectional filtering strategy, calculating local mean value in a dynamic search window of tangential acceleration signal by taking a ground impact extreme point of the radial acceleration signal as an absolute time anchor point And standard deviation of Constructing an adaptive feature search threshold: Wherein k is a statistical sensitivity coefficient, and the search satisfaction is greater than Is less than the peak and immediate vicinity of Calculating the time difference between the center and the anchor point And performing a reverse translational alignment: Wherein, the The value of the aligned tangential strain signal at time t is indicated, Representing the original tangential strain signal, Representing the calculated time lag amount.
4. A tire tread depth estimation method based on a physical information fusion network as in claim 3 wherein in step S2 the signal preprocessing further comprises identifying adjacently impacting anchor points And (3) with Define slice boundaries as A single-turn sample is truncated and resampled to a fixed N-point angle domain waveform by linear interpolation, and finally a Z-Score normalization process is performed.
5. The method for estimating tire tread depth based on the physical information fusion network according to claim 1, wherein in step S3, the construction of the spatio-temporal feature extraction neural network includes using a feature extraction module composed of three stacked time convolution residual blocks, each layer adopts a causal expansion convolution structure with expansion ratio d=1, 2,4 in turn, and configures a Chomp1d causal clipping layer and SiLU activation functions after the convolution layer.
6. The method for estimating tire tread depth based on physical information fusion network as in claim 5, wherein in step S3, the physical condition fusion mechanism comprises setting a load normalization reference Normalized to speed Dimension normalization is carried out on the real-time load L and the speed V: Scalar quantity And (3) with Replication expansion into timing vectors in the time dimension And (3) with Before entering the bidirectional gating circulating unit layer, the expanded time sequence vector and the waveform characteristic tensor output by the convolution layer are carried out Splicing in the channel dimension to form a mixed characteristic tensor : Wherein, the Representing a stitching operator along a feature channel dimension; And then, extracting global time sequence characteristics by using a Bi-directional gating circulating unit Bi-GRU, calculating weights by a self-attention mechanism, and focusing a grounding trace area to output deep characteristic vectors.
7. The tire tread depth estimation method based on the physical information fusion network as claimed in claim 1, wherein in step S4, the construction of the physical constraint module comprises setting a learnable carcass rigidity coefficient And material structural factor And indexing it to ensure physical positive qualification: Wherein, the Unconstrained weight parameters which are actually updated in the back propagation process for the network; network output layer branch prediction pattern depth And vertical deformation Wherein The activation function constraint is positive by Softplus: Wherein, the In order to smooth the coefficient of the coefficient, Is the linear output of the full connection layer.
8. The method for estimating tire tread depth based on physical information fusion network as in claim 7 wherein in step S4, the mechanical balance formula comprises defining tread instantaneous stiffness based on an improved brush model Defining the total vertical stiffness of the tire based on the series spring model And combining Hooke's law to construct a mechanical balance equation and reversely pushing theoretical load : 。
9. A tire tread depth estimation method based on a physical information fusion network as in claim 1, wherein in step S5, the mixing loss function The mathematical expression of (2) is: Wherein, the As a result of the physical weight coefficient, Calculating a data driving error by adopting a mean square error; For physical consistency errors, in order to avoid gradient explosion caused by large load values, a normalized ratio form is adopted to construct: Wherein, the For the theoretical back-thrust load, In order to actually acquire the load, Is a stable constant of the numerical value, and synchronously updates the network weight and the physical parameter by utilizing a gradient descent algorithm in the training process And (3) with 。
10. A tire pattern depth estimation method based on a physical information fusion network as set forth in claim 1, wherein in step S5, the probability statistics method comprises presetting a standard wear state anchor point set covering the full life cycle of the tire Calculating the continuous predictive value z output by the network to belong to each anchor point And is converted into posterior probability distribution by using Gaussian kernel function : Selecting probabilities The largest anchor point is output as the final wear state, and the information entropy H of the distribution is calculated: and outputting a low-confidence early warning signal when the information entropy H is larger than a preset warning threshold value.

Description

Tire pattern depth estimation method based on physical information fusion network Technical Field The invention belongs to the technical field of intelligent traffic systems and active safety of vehicles, and particularly relates to a tire pattern depth estimation method based on a physical information fusion network. Background The state of health of a tire, which is the only component of an automobile in contact with the road surface, directly determines the dynamic performance, braking performance and running safety of the vehicle. The pattern depth is a quantization index representing the most core of the tire wear degree, and gradually decreases along with the increase of the vehicle driving mileage, so that the tire drainage capacity is directly degraded, the wet land gripping force is attenuated, and the water slipping phenomenon and the tire burst accident are extremely easy to induce. Therefore, the method realizes accurate and real-time monitoring of the tire pattern depth, and has important practical significance for ensuring driving safety, optimizing maintenance strategies of motorcades and reducing operation cost. Conventional tire wear detection relies primarily on manual periodic maintenance using a mechanical depth gauge to measure the tread groove depth. The method is low in efficiency, time-consuming and labor-consuming, has a huge monitoring blind area, and cannot reflect the sudden abrasion state in the driving process in real time. With the development of sensor technology, non-contact detection equipment based on laser triangulation and machine vision identification appears on the market. However, such external detection devices are usually fixedly installed at a parking lot entrance or a specific road section, are significantly affected by environmental illumination changes, muddy water coverage on the tire surface and cleanliness, and are expensive in equipment deployment cost, so that the external detection devices are difficult to be popularized to vehicle-mounted terminals on a large scale to realize mobile monitoring. In recent years, indirect monitoring methods based on intelligent tire technology are becoming a research hotspot. The method collects dynamic vibration signals through micro sensors such as accelerometers and strain gauges integrated in the tire inner liner, and inverts the abrasion state by utilizing signal processing and machine learning technologies. However, existing data driven methods still face a number of challenges in practical engineering applications (1) lack of physical interpretability. At present, the mainstream convolutional neural network or the circulating neural network and other networks mostly adopt a black box mapping mode, only the statistical correlation of data input and output is concerned, and the physical characteristics of the tire such as rubber rigidity, contact mechanics and the like are ignored. When the pure data driving model faces working conditions which are not contained in the training set, physical constraint is often lacked, and a prediction result against the physical law is easily output, so that the reliability of the system is doubtful. And (2) generalizing the difficult problem under the complex variable load working condition. The load variation range of the heavy commercial transport vehicle is large, and the data distribution of the vibration signals is severely drifted due to the large difference of physical dimensions. The existing pure data model lacks self-adaptive sensing capability to physical working conditions, and the weight cannot be dynamically adjusted to adapt to the change of physical boundary conditions in the feature extraction stage, so that the estimated accuracy of the model under extreme working conditions is reduced in a cliff type. In view of the foregoing, there is a need for a tire tread depth estimation method that can effectively incorporate physical mechanism constraints and has adaptive perceptibility of complex variable load conditions, so as to break through the bottleneck in the prior art. Disclosure of Invention The invention aims to provide a tire pattern depth estimation method based on a physical information fusion network, which can overcome the defects of poor physical interpretability, weak generalization capability under heavy load working conditions and the like in the prior art. In order to achieve the purpose, the technical scheme of the invention is that the tire pattern depth estimation method based on the physical information fusion network comprises the following steps: S1, acquiring triaxial acceleration data of tires with different wear degrees under different load and speed working conditions by adopting an intelligent tire system with triaxial acceleration sensors attached to the inner liner of the tire; S2, performing zero phase shift filtering denoising on the acquired acceleration data, aligning phases of signals of different channels by using a characteristic poin