Aircraft tire state intelligent evaluation and residual life prediction system and method

Abstract

The invention discloses an aircraft tire state intelligent evaluation and residual life prediction system and method, and the system comprises a multi-mode intelligent image collection module which is used for obtaining the multi-mode data of a tire; the data processing and analyzing module is used for processing and analyzing the multi-modal data and outputting a state evaluation result and a residual service life prediction result of the aircraft tire; and the interactive result display and intelligent decision module is used for displaying the state evaluation result and the residual service life prediction result and generating a maintenance suggestion. Through software and hardware collaborative design and multi-source information fusion, objective, accurate and comprehensive evaluation of the state of the aircraft tire and more reliable life-cycle prediction are realized, the detection reliability and the flight safety are remarkably improved, and technical support is provided for predictive maintenance.

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft maintenance and safety assurance, and in particular relates to a system and method for intelligently evaluating the state of aircraft tires and predicting their remaining life. Background Art

[0002] Aircraft tires are critical components for safe aircraft takeoff and landing, making their condition monitoring crucial. Traditional inspection methods, such as manual visual inspection, tapping, and tread depth measurement, have significant limitations: They are highly subjective and inconsistently standardized, leading to missed detections and misjudgments; they are inefficient and costly, making them difficult to meet the demands of high-frequency inspections; they are limited to surface inspections, making it difficult to detect internal defects and premature aging; they are unable to accurately and quantitatively assess key aging factors, such as thermal-oxidative aging (which requires correlation with age and historical temperature data in the service area), chemical factors, and their coupling effects with wear under specific operating conditions; in particular, they fail to fully integrate programmatically imported historical usage data (such as takeoff and landing times, regional temperature) with initial tire condition data for accurate and dynamic assessments; and they rely on fixed replacement cycles, making it difficult to predict remaining useful life (RUL) based on actual health conditions, resulting in wasteful use or risks. Existing automation attempts (such as laser scanning and general image processing) face significant bottlenecks in comprehensive data acquisition, accurate damage and aging assessment, reliable dynamic life prediction based on historical data, and intelligent decision support. Therefore, this field urgently needs an integrated intelligent technology solution to achieve accurate and comprehensive evaluation of aircraft tire status and reliable RUL prediction to meet the safety, efficiency and economy requirements of modern aviation. Summary of the Invention

[0003] This invention aims to overcome the limitations of existing technologies and provide an integrated, intelligent aircraft tire condition assessment and lifecycle prediction system and method. Its core objectives are to achieve precise wear assessment based on initial condition comparison, thermal oxidative aging assessment based on usage time and regional temperature, surface quality assessment based on microscopic morphology big data analysis, and precise, quantifiable, simultaneous assessment of potential damage. The system also integrates real-time assessment results with historical lifecycle data to perform high-precision, dynamic RUL predictions with uncertainty analysis. Furthermore, it generates intelligent maintenance recommendations to improve aviation safety, optimize maintenance strategies, and reduce operating costs.

[0004] To achieve the above objectives, the present invention provides an intelligent aircraft tire condition assessment and remaining life prediction system. Specifically, the intelligent aircraft tire condition assessment and remaining life prediction system includes:

[0005] Multimodal intelligent image acquisition module, used to obtain multimodal data of tires;

[0006] A data processing and analysis module, configured to process and analyze the multimodal data and output a condition assessment result and a remaining service life prediction result of the aircraft tire;

[0007] The interactive result display and intelligent decision-making module is used to display the condition assessment results and remaining service life prediction results, and generate maintenance recommendations.

[0008] Preferably, the multimodal intelligent image acquisition module includes:

[0009] An image acquisition unit, used to acquire image data of aircraft tires by combining multiple imaging technologies;

[0010] The light source system is used to provide controllable and diverse light sources, optimize lighting parameters according to the detection target, and intelligently select light source combinations and adjust brightness according to acquisition requirements;

[0011] Positioning and attitude adjustment unit, used for precise positioning and rotation platform driven by servo motors, suitable for angular positioning and 360° rotation scanning of tires of different specifications;

[0012] The adaptive acquisition unit is used to dynamically adjust imaging parameters based on the feedback of preliminary analysis results, tire historical data or identified tire model, and enhance the acquisition of specific damage or aging characteristics.

[0013] Preferably, the image acquisition unit includes:

[0014] A camera array configured with a number of cameras, arranged as needed, to capture visual images of the tread and sidewall from various preset angles; wherein the preset angles include 0°, ±30°, ±60° relative to the tread centerline, and specific angles relative to the sidewall;

[0015] A structured light 3D imaging unit, configured to reconstruct 3D point cloud data of the tire surface by using a structured light projector and at least two cameras to work together;

[0016] Multispectral / hyperspectral imaging unit, used to acquire multi-channel images using a specific wavelength light source array in conjunction with a camera with a filter or a dedicated multispectral camera;

[0017] At least one sensing unit, the sensing unit comprising at least one item from the following group: an integrated thermal imaging sensor mounted on a robotic arm capable of performing synchronous scanning as the tire rotates, positioned directly above the tire during tread testing and to the side of the tire during sidewall testing; and an integrated ultrasonic flaw detection unit, which utilizes a dedicated fixture to allow the probe to closely adhere to the tire surface at a specific angle of incidence and to scan along the circumference of the tire.

[0018] Preferably, the adaptive acquisition unit includes:

[0019] Pre-scanning and region of interest positioning unit, used to automatically identify tire models through scanning and preliminarily locate key areas;

[0020] Intelligent lighting optimization unit, which automatically calls preset lighting schemes or dynamically adjusts lighting parameters to maximize contrast of specific features based on a tire model database, historical data, or pre-scan image analysis results;

[0021] Dynamic multi-angle fine scanning unit is used to automatically increase the acquisition angle density or adjust the camera posture for rescanning the identified key areas of interest or suspected defect areas;

[0022] The adaptive multimodal fusion acquisition unit is used to intelligently decide whether to enable some or all imaging modalities for data acquisition and fusion based on preset detection tasks or real-time analysis feedback, balancing detection efficiency and accuracy.

[0023] Preferably, the data processing and analysis module includes:

[0024] Multi-source data preprocessing and fusion unit, used for denoising, registration, correction, segmentation, feature extraction of multimodal image data, and processing of tire historical data;

[0025] The core deep learning model unit is used to perform thermal oxidation aging assessment based on usage time, cumulative damage assessment based on takeoffs and landings, tread profile wear depth quantification, feature decoupling and quantification of tire surface quality analysis, comprehensive state assessment, and dynamic remaining service life prediction and uncertainty quantification through pre-trained deep learning models;

[0026] The model continuous learning and updating unit is used to support online or incremental learning and continuously optimize the model using new data and feedback.

[0027] Preferably, the multi-source data preprocessing and fusion unit includes:

[0028] The first processing unit is used to perform image enhancement preprocessing on the 2D image to highlight subtle defects and texture information on the tire surface. It also uses histogram equalization to improve image contrast and a sharpening filter to enhance edge details. Alternatively, it performs image segmentation first, using a semantic segmentation network to segment different areas of the tread, shoulder, and sidewall and perform feature analysis.

[0029] The second processing unit is used to build a high-quality three-dimensional model of the tire surface, remove outliers and smooth the 3D point cloud / depth map, accurately align point cloud data from different perspectives or multiple scans, and stitch them together to obtain a complete tire model;

[0030] The third processing unit is used to access and process historical data related to tires from a central database or through an interface; the historical data includes weather temperature data of the aircraft usage area, historical take-off and landing data, and tire model data; the processing includes cleaning and standardizing the historical data, and calculating the cumulative damage factor and aging equivalent time based on the processed data.

[0031] Preferably, the deep learning model in the core deep learning model unit adopts a multi-task, multi-modal, spatiotemporal fusion architecture, including:

[0032] Specific modality feature extractor, used to extract features of different modal data respectively;

[0033] Feature fusion module, used to fuse features of different modalities at multiple levels;

[0034] The condition assessment subnetwork is used to output the condition assessment results of the aircraft tire's wear index and aging index based on the fused features;

[0035] The remaining service life prediction subnetwork is used to combine current state characteristics and historical data to predict the remaining service life of aircraft tires and quantify the uncertainty of the prediction.

[0036] Preferably, the method for quantifying the uncertainty of the prediction includes Bayesian neural network and Monte Carlo method.

[0037] Preferably, the interactive result display and intelligent decision-making module includes:

[0038] Visual display unit, used to display the evaluation results and remaining service life prediction results in an intuitive way;

[0039] Intelligent alarm and diagnostic unit, used to trigger graded alarms based on the evaluation results and provide preliminary fault diagnosis information;

[0040] A maintenance suggestion generation unit is used to generate maintenance suggestions based on the evaluation results and the remaining service life prediction results, combined with the maintenance rule base or knowledge graph;

[0041] The maintenance plan optimization unit is used to provide optimized maintenance scheduling recommendations based on the overall fleet situation and resource constraints.

[0042] Preferably, the interactive result display and intelligent decision-making module further integrates a knowledge graph in the field of aircraft tires. The knowledge graph is a structured semantic knowledge base that stores entities, concepts, attributes and complex relationships related to aircraft tires, and is used to assist the maintenance suggestion generation unit in generating more accurate and explainable maintenance suggestions based on the evaluation results and inference rules of the deep learning model.

[0043] The present invention also provides a method for intelligently evaluating the state of an aircraft tire and predicting its remaining life, comprising:

[0044] Acquire multimodal data of the tire, process and analyze the multimodal data, and output a condition assessment result and a remaining service life prediction result of the aircraft tire;

[0045] The condition assessment results and remaining service life prediction results are displayed, and maintenance recommendations are generated.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] Improved comprehensiveness and accuracy of assessments: Simultaneous, distinguishable, and quantifiable assessments of wear, various types of aging (UV / thermal / oxidative / chemical, etc.), and minor damage are now possible, providing deeper insights into the tire's true health.

[0048] Enhanced RUL prediction accuracy and reliability: By integrating real-time multimodal sensor data with full life cycle historical data and quantifying prediction uncertainty, the accuracy of RUL prediction and the credibility of decision-making are greatly improved.

[0049] Enables predictive and prescriptive maintenance: Provides smarter maintenance recommendations based on analysis of specific degradation patterns and uncertainty considerations, supporting the transition to advanced condition-based maintenance.

[0050] Detection efficiency and automation levels have been greatly improved: the highly automated process has shortened detection time, reduced labor costs and subjective errors.

[0051] Dual optimization of safety and economy: Improve safety through early warning of potential risks, accurately assess to avoid waste, extend the effective service life of tires, optimize maintenance plans, and reduce costs.

[0052] Systematic and intelligent management: Provides end-to-end closed-loop solutions that are easy to integrate and promote intelligent maintenance processes.

[0053] Continuous evolution capability: It has online / incremental learning capabilities to maintain long-term high performance and applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0055] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0056] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0057] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment provides an aircraft tire condition intelligent assessment and remaining life prediction system, including:

[0060] Multimodal intelligent image acquisition module, used to obtain multimodal data of tires;

[0061] The data processing and analysis module is used to process and analyze multimodal data and output the aircraft tire's condition assessment results and remaining service life prediction results;

[0062] Interactive result display and intelligent decision-making module, used to display condition assessment results and remaining service life prediction results, and generate maintenance recommendations.

[0063] Furthermore, the multimodal intelligent image acquisition module includes:

[0064] The image acquisition unit is used to acquire image data of aircraft tires by combining multiple imaging technologies. Specifically, this unit uses a combination of imaging technologies, such as high-resolution (e.g., 4K and above) industrial camera arrays for multi-angle visual imaging, structured light projection to acquire three-dimensional morphology, and multispectral / hyperspectral imaging to detect changes in material surfaces. It can integrate thermal imaging and ultrasonic flaw detection modules to achieve full-dimensional perception.

[0065] The light source system is used to provide controllable and diverse light sources, optimize lighting parameters according to the detection target, and intelligently select light source combinations and adjust brightness according to acquisition requirements;

[0066] Specifically, the system is equipped with a controllable and diverse light source system with intelligent lighting control capabilities, optimizing lighting parameters based on the detection target. This includes various light sources, including annular, strip, coaxial, and low-angle. The controller intelligently selects the light source combination and adjusts the brightness based on the acquisition requirements (such as emphasizing texture, eliminating shadows, and enhancing scratch contrast).

[0067] Positioning and attitude adjustment unit, used for precise positioning and rotation platform driven by servo motors, adapting to different specifications of tires, achieving precise angular positioning and 360° rotation scanning;

[0068] Specifically, the integrated precision tire positioning and posture adjustment device ensures stable and comprehensive acquisition geometry. The servo motor-driven precision positioning and rotation platform can adapt to tires of different specifications and achieve precise angular positioning and 360° rotation scanning, ensuring stable and comprehensive acquisition.

[0069] It can also include an environmental control unit to reduce external light interference by configuring a light shield or darkroom environment, and can optionally be equipped with temperature and humidity sensors to record environmental parameters.

[0070] The adaptive acquisition unit is used to dynamically adjust imaging parameters (such as light source and angle) based on preliminary analysis results or historical data, enhance the acquisition of specific damage or aging features, and obtain optimal information.

[0071] Furthermore, the image acquisition unit includes:

[0072] A camera array configured with a number of cameras, arranged at various preset angles, to capture visual images of the tread and sidewall. Preset angles include 0°, ±30°, ±60°, and specific sidewall angles relative to the tread centerline.

[0073] Tire surface micro-morphology acquisition unit: During each maintenance inspection, a high-resolution camera (e.g., with macro shooting capabilities) integrated into the acquisition module is used to capture micro-morphological images of the tire surface (particularly critical areas and suspected defect areas). These images are designed to capture surface details that are difficult to discern with the naked eye, such as fine cracks, early signs of aging, and material fatigue. These images are then used to predict the tire surface quality condition through big data comparison and evaluation.

[0074] Tread profile and wear depth measurement unit: A portable laser profile measurement system is preferred, which uses high-precision laser scanning to obtain three-dimensional profile data of the tire tread and sidewall. The acquired three-dimensional data will be accurately aligned and compared with the tire's initial three-dimensional profile data (factory or last retread baseline data) to calculate the actual wear amount, wear distribution, and maximum wear depth;

[0075] Specifically, six 12-megapixel industrial CMOS cameras with a frame rate of 60fps are configured and arranged from different preset angles to fully capture visual images of the tread and sidewall.

[0076] A structured light 3D imaging unit, configured to reconstruct 3D point cloud data of the tire surface by using a structured light projector and at least two cameras to work together;

[0077] Specifically, the structured light projector uses digital light processing (DLP) technology with a resolution of 1920×1080 and a projection frequency of 100Hz. The visual images of the tread and sidewall include tread depth, wear, and geometric deformation (bulges, dents), etc.

[0078] Multispectral / hyperspectral imaging unit, used to acquire multi-channel images using a specific wavelength light source array in conjunction with a camera with a filter or a dedicated multispectral camera;

[0079] Specifically, a multi-channel image is acquired using an array of light sources in specific wavelengths (e.g., LEDs covering UV 365nm, blue 450nm, green 530nm, red 660nm, near-infrared 850nm, and 940nm) in conjunction with a camera with optical filters or a dedicated multispectral camera. Different wavelengths are used to detect different information. For example, the VIS wavelength performs color and texture analysis to assist in assessing the macroscopic condition of a surface, while the NIR wavelength can be used to detect early changes within a material or variations in specific chemical components.

[0080] At least one sensing unit, wherein the sensing unit is selected from at least one of the following groups:

[0081] Integrated thermal imaging sensor: The thermal imaging sensor is mounted on a robotic arm that scans synchronously with the tire's rotation. It is positioned directly above the tire during tread inspection and to the side of the tire during sidewall inspection. Specifically, the thermal imaging sensor has a resolution of ≥640×480 and an accuracy of ±2°C.

[0082] An integrated ultrasonic flaw detection unit uses a dedicated fixture to position the probe against the tire surface at a specific angle of incidence, enabling scanning along the tire's circumference. The ultrasonic flaw detection unit operates in a 1-10MHz frequency range and supports both B / C scanning.

[0083] The data collected by the sensor unit must be able to be accurately aligned in space and time with the image data collected by the high-resolution camera; the scanning range and scanning path of the selected sensor unit can completely cover the key inspection areas of the tire (such as the shoulder and bead area) and the mandatory inspection areas required by the maintenance manual.

[0084] Furthermore, the adaptive acquisition unit has a built-in adaptive control module that can dynamically adjust imaging parameters (such as light source type, intensity, illumination angle, camera exposure time, acquisition angle combination, etc.) based on the preliminary analysis results fed back from the integrated data processing and analysis module, tire historical data or identified tire model through a preset optimization algorithm, thereby enhancing the acquisition of specific damage or aging characteristics to obtain optimized information for subsequent precise analysis.

[0085] Pre-scanning and region of interest (ROI) positioning unit, used to automatically identify the tire model through scanning (such as by reading a barcode or identifying the tread pattern) and preliminarily locate key areas, such as major wear areas, vulnerable areas such as the shoulder, and critical aging areas such as the sidewall;

[0086] An intelligent lighting optimization unit, which automatically calls preset lighting schemes or dynamically adjusts lighting parameters to maximize contrast of specific features based on a tire model database, historical data, or pre-scanned image analysis results (such as surface roughness and color);

[0087] Dynamic multi-angle fine scanning unit is used to automatically increase the acquisition angle density or adjust the camera posture for the identified key areas of interest or suspected defect areas, and perform a higher resolution scan again;

[0088] The adaptive multimodal fusion acquisition unit is used to intelligently decide whether to enable some or all imaging modalities for data acquisition and fusion based on preset detection tasks (such as rapid screening or fine diagnosis) or real-time analysis feedback, thereby balancing detection efficiency and accuracy.

[0089] Furthermore, the data processing and analysis module is equipped with high-performance computing resources and runs core software. It is the intelligent core of the system and is responsible for processing data, running models and performing analysis. It includes:

[0090] Multi-source data preprocessing and fusion unit, used to perform denoising, registration, correction, segmentation, feature extraction on multimodal image data, and process tire historical data (basic information, usage records, maintenance records, environmental data, etc.);

[0091] All image data is first denoised using general basic preprocessing to improve the signal-to-noise ratio. For example, Gaussian filtering can be used for Gaussian noise, while median filtering can be used for salt and pepper noise. The specific method chosen depends on the actual acquisition environment.

[0092] The core deep learning model unit is used to perform thermal oxidation aging assessment based on usage time, cumulative damage assessment based on takeoffs and landings, tread profile wear depth quantification, feature decoupling and quantification of tire surface quality analysis, comprehensive state assessment, and dynamic remaining service life prediction and uncertainty quantification through pre-trained deep learning models;

[0093] The model continuous learning and updating unit is used to support online or incremental learning, continuously optimize the model using new data and feedback, and can also be combined with active learning strategies.

[0094] Specifically, incremental / online learning capability: The system of this embodiment has the ability to use newly acquired labeled data (such as actual tire failure data, annotations confirmed by expert review) to incrementally update or online learn the existing model without the need for complete retraining, thereby being able to continuously adapt to new tire models, materials or operating conditions and maintain model performance.

[0095] Active learning strategy: The system of this embodiment can integrate an active learning algorithm. The system can identify samples with high model prediction uncertainty and submit them to experts for labeling first, guiding model learning in the most economical and efficient way to accelerate performance improvement.

[0096] It also includes a computing platform, which uses a high-performance server cluster in a fixed system, equipped with multi-core CPUs, large-capacity RAM, high-speed storage (SSD array) and multiple high-performance GPU accelerator cards (such as NVIDIA A100 / H100 series) to support large-scale data processing and training and reasoning of deep learning models.

[0097] Furthermore, the multi-source data preprocessing and fusion unit includes:

[0098] The first processing unit is used to perform image enhancement preprocessing on the 2D image to highlight subtle defects and texture information on the tire surface. It also uses histogram equalization to improve image contrast and a sharpening filter to enhance edge details. Alternatively, it performs image segmentation first, using a semantic segmentation network (U-Net) to segment different areas of the tread, shoulder, and sidewall and perform feature analysis.

[0099] The second processing unit is used to construct a high-quality 3D tire surface model, remove outliers (such as statistical filtering and radius filtering) and smooth the 3D point cloud / depth map (such as MLS smoothing). It also performs precise registration (such as the ICP algorithm and its variants) on point cloud data from different perspectives or multiple scans, and stitches them together to obtain a complete tire model.

[0100] The third processing unit is used to access and process historical data related to tires from a central database or through an interface; the historical data includes weather temperature data of the aircraft usage area, historical take-off and landing data, and tire model data; the processing includes cleaning and standardizing the historical data, and calculating the cumulative damage factor and aging equivalent time based on the processed data.

[0101] Specifically, historical data includes basic information (model, batch, material formula, etc.), programmatically imported usage records (cumulative usage time, cumulative takeoffs and landings, historical weather and temperature data at the main airports where the tire was operated during its service life, load spectra, etc.), maintenance records (historical test results, retreading records, historical surface micro-topography image libraries, etc.), and initial tire condition data (such as initial 3D profile and initial surface characteristic parameters). Historical data (especially takeoffs and landings, usage time, and temperature data) are cleaned and standardized, and a standard total takeoff and landing reference range corresponding to the tire model is established.

[0102] Furthermore, the deep learning model in the core deep learning model unit adopts a multi-task, multi-modal, spatiotemporal fusion architecture (such as a fusion network of CNN, Transformer, GNN, LSTM and other components), including:

[0103] Specific modality feature extractor, used to extract features of different modal data respectively;

[0104] For 2D high-resolution visual images, advanced convolutional neural networks (CNNs) are used as the backbone network, such as the ResNet series (such as ResNet-50) or VisionTransformer (ViT) and its variants, to extract local visual features such as texture, color, fine cracks, scratches, etc. on the tire surface.

[0105] For 3D point cloud data, a network specialized in processing point clouds, such as PointNet++ and DGCNN, is used to learn the three-dimensional spatial features of the tire, including its macroscopic geometric shape, microscopic geometric features (such as tread depth and wear), and point cloud density distribution.

[0106] Feature fusion module, used to fuse features of different modalities at multiple levels;

[0107] The hierarchical design of the feature fusion module adopts a multi-level, gradually deepening fusion strategy:

[0108] Early fusion: For some modalities that naturally have strong correspondences, such as RGB images, pixel-level fusion can be performed at the data input layer.

[0109] Mid-term fusion: After each modality feature extractor (e.g., CNN for 2D images, PointNet++ for 3D point clouds) outputs its own high-dimensional feature vectors, these feature vectors are fed into one or more specially designed fusion modules. These fusion modules can be concatenated and then passed through fully connected layers for dimensionality reduction, element-wise multiplication / summation, or more complex attention-based fusion networks (e.g., cross-modal Transformer layers). The goal is to learn the correlation and complementarity between the features of each modality and generate a more discriminative unified feature representation.

[0110] Late fusion: Each modality or partially fused features are used by independent classifiers or regressors to make preliminary predictions (such as independent wear assessment, aging assessment), and then these prediction results from different modalities are weighted averaged.

[0111] The introduction of an attention mechanism promotes intermodal information fusion, and techniques such as adversarial learning or orthogonal constraints can be used to decouple the characteristics of different degradation modes (wear and tear, various aging patterns). For historical data, recurrent neural networks (RNNs such as LSTM and GRU) or temporal convolutional networks (TCNs) can be used to process time series information. Ultimately, the multimodal features of the current state are combined with historical time series features through cascaded, weighted, or more complex fusion networks.

[0112] The condition assessment subnetwork is used to output the condition assessment results of the aircraft tire's wear index and aging index based on the fused features;

[0113] The remaining service life prediction subnetwork is used to combine current state characteristics and historical data to predict the remaining service life of aircraft tires and quantify the uncertainty of the prediction.

[0114] Furthermore, the model learns the unique feature expressions of different degradation modes (such as wear, UV aging, thermal oxidative aging, etc.) in multimodal data through specific network structures (such as attention mechanism, adversarial learning module), separates the influence of these modes, and outputs their respective quantitative indicators (such as wear percentage, specific aging index, damage severity score, etc.).

[0115] During condition assessment, after feature extraction and multi-level feature fusion, the model generates a comprehensive high-dimensional feature vector that includes the tire's current geometry (from 3D point clouds and 2D images), surface material properties (from 2D images), and historical evolution trends (from time series data). This feature vector serves as input to the subsequent subnetwork for specific assessment tasks.

[0116] For wear assessment: The fused features are fed into a fully connected network to output specific, quantitative wear indicators, such as average tread depth (mm), remaining tread percentage (%), maximum wear (mm), wear volume (cm 3 ), etc. These are usually continuous regressors.

[0117] For tire surface quality assessment: high-resolution images acquired from the tire surface micro-topography acquisition unit are first pre-processed (e.g. denoising, enhancement, segmentation). Subsequently, deep learning models (e.g. convolutional neural networks CNN, generative adversarial networks GAN, etc.) are used to extract micro-topography features. These features are matched and compared with a large-scale database of tire surface images, which contains images of different service stages, different health conditions, and known defect types. Through similarity calculation, anomaly detection, and pattern recognition, the system assesses the crack density, length, width, material spalling, roughening, etc. of the current tire surface, and compares it with historical data to predict its development trend, outputting a comprehensive surface quality score and specific defect description.

[0118] For thermal-oxidative aging assessment: the fused features (especially cumulative usage time, programmatically imported service area historical temperature data, tire material property parameters) are fed into a specialized thermal-oxidative aging assessment subnetwork. This subnetwork uses physical model-based or data-driven algorithms (e.g. Arrhenius equation-based correction models or deep learning time series models) to calculate the cumulative impact of thermal-oxidative aging on material performance, outputting a quantitative thermal-oxidative aging index or aging level (e.g. slight, moderate, severe).

[0119] Comprehensive health index generation: the system calculates a normalized tire health index (0-100 points) based on the above-mentioned independent assessment results (wear indicators, aging indices, damage severity, etc.), combined with pre-set weight factors or dynamic weights learned through end-to-end learning. This value provides the operator with an intuitive overview of the overall health of the tire.

[0120] Further, methods for quantifying the uncertainty of prediction include Bayesian neural networks, Monte Carlo methods.

[0121] Dynamic RUL prediction and uncertainty quantification: combining current state and historical data, the RUL (e.g. remaining takeoff and landing cycles) is predicted and Bayesian deep learning, ensemble learning or similar methods are used to output the probability distribution or confidence interval of the prediction result, quantifying the uncertainty of the prediction.

[0122] Based on the fused current state assessment results and historical spatio-temporal features, the model will predict the remaining useful life (RUL) of the tire, usually expressed as the remaining takeoff and landing cycles, flight hours.

[0123] The specific method for quantifying uncertainty in this embodiment can include:

[0124] Bayesian Neural Network: In a Bayesian neural network, the network weights are no longer fixed scalar values, but are modeled as probability distributions (such as Gaussian distributions). By sampling on these weight distributions (using Markov Chain Monte Carlo (MCMC) methods for posterior sampling), the model can obtain a series of possible RUL values ​​when making a RUL prediction, thereby directly constructing the posterior probability distribution of RUL.

[0125] Monte Carlo method: After model training is complete, the dropout layer in the network remains activated during inference (i.e., a portion of neurons are randomly inactivated). Multiple (e.g., dozens to hundreds) independent forward propagations are performed on the same input sample. Due to the randomness of the dropout layer during each forward propagation, a slightly different set of RUL predictions is obtained. By collecting these predictions, we can approximate the empirical distribution of the RUL, and thus calculate its mean as the final prediction value, and its standard deviation or variance as a measure of uncertainty.

[0126] Data for Uncertainty Quantification: The input data for RUL prediction and its uncertainty quantification includes not only the tire's immediate health status feature vector extracted from current multimodal sensor data (such as high-resolution images and 3D point clouds), but also the tire's lifecycle historical data. This includes detailed usage records (cumulative takeoffs and landings, cumulative flight hours, airports associated with each takeoff and landing, and approximate temperature conditions), maintenance records (including historical wear measurements, thermal oxidation aging assessments, and surface quality assessments), and average storage temperature and humidity. The tire's current cumulative takeoffs and landings are compared with the recommended total takeoff and landing range for its model, serving as a key reference for RUL prediction. This historical data is processed using a time series model (such as LSTM) and fused with the current health status features. Uncertainty quantification methods (such as Bayesian neural networks or MCDropout) learn the relationship between these complex inputs and the true RUL value, as well as its inherent variability, to assess the fluctuation range of RUL predictions when input data is noisy, model parameters are uncertain, or when encountering conditions not fully covered in the training data.

[0127] By outputting the probability distribution or confidence interval of RUL, the system can clearly convey to the user the reliability of the prediction results. For example, a predicted RUL of 100 takeoffs and landings, but its 95% confidence interval is [20, 180] times, is compared with another predicted RUL of 90 takeoffs and landings, but its 95% confidence interval is [80, 100] times. The former has much greater uncertainty and requires more cautious maintenance decisions.

[0128] As a supplementary embodiment, to further enhance the accuracy, interpretability, and intelligence of assessments, the system of this embodiment can integrate a knowledge graph for aircraft tires. This knowledge graph is a structured semantic knowledge base that stores various entities, concepts, attributes, and the complex relationships between them related to aircraft tires in the form of a graph.

[0129] Examples of entity types: Entities in the knowledge graph may include, but are not limited to: specific tire models, various physical components of the tire (such as tread, sidewall, cord layer, bead wire, etc.), manufacturing materials, known defects and damage types (uniform wear, uneven wear, cuts, punctures, aging (mainly referring to thermal oxidation aging based on usage time and regional temperature), etc.), environmental influencing factors (average take-off and landing temperature, humidity range, ultraviolet radiation intensity level, pavement roughness level of major operating airports), maintenance operations and standards (such as routine inspections, refurbishment standards, scrapping standards), sensor types and parameters, and known failure modes and failure mechanisms.

[0130] Integration and reasoning between knowledge graphs and deep learning models: The initial evaluation results of deep learning models through multimodal data analysis can be input into the knowledge graph as queries or evidence. Subsequently, the reasoning engine can use the structured prior knowledge stored in the knowledge graph to perform deep logical reasoning.

[0131] Furthermore, the interactive result display and intelligent decision-making module includes:

[0132] Visual display unit, used to display the evaluation results and remaining useful life prediction results in an intuitive way (3D rendering, heat map, RUL probability density curve);

[0133] Intelligent alarm and diagnostic unit, used to trigger graded alarms based on the evaluation results and provide preliminary fault diagnosis information;

[0134] A maintenance suggestion generation unit is used to generate maintenance suggestions based on the evaluation results and the remaining service life prediction results, combined with the maintenance rule base or knowledge graph;

[0135] Specifically, this unit integrates a decision-making engine that automatically generates maintenance recommendations with clear action directions (for example, "continue service," "re-inspection recommended after X takeoffs and landings," "immediate decommissioning and detailed flaw detection," etc.) based on the RUL prediction value and uncertainty quantification results output by the core deep learning model, the severity assessment of each defect mode, and reference to the built-in configurable maintenance rule library or through interactive reasoning with the knowledge graph module.

[0136] The maintenance plan optimization unit is used to provide optimized maintenance scheduling recommendations based on the overall fleet situation and resource constraints.

[0137] The system also includes a data management and communication module, which is used to store structured data in the database and provide an API interface for data interaction with MMS, ERP and other systems.

[0138] The interactive results display and intelligent decision-making modules also include:

[0139] (1) Hardware: Use a high-resolution (such as 4K) touch screen display and a corresponding workstation.

[0140] (2) Software (example: Web application architecture) includes:

[0141] Front-end interface: Provides a responsive and user-friendly graphical user interface (GUI).

[0142] Visualization engine: Using technologies such as WebGL, the system can visualize and render the three-dimensional tire model. The detected wear distribution (such as through a color depth map), aging degree (such as through heat map mapping), damage location and type (such as through annotation boxes or symbol marks) are intuitively superimposed on the model.

[0143] Dashboards and charts: Dashboards, graphs, bar charts, etc. are used to clearly display quantitative wear indicators, various types of aging indices, RUL prediction values ​​and their probability distribution curves, as well as historical trend charts of these indicators over time.

[0144] Intelligent alarms and notifications: Based on preset thresholds or rules (e.g., excessive wear, severe aging of a specific type, RUL below a safety threshold, excessive prediction uncertainty), visual or audible alarms of different levels (e.g., green-yellow-red) are triggered, and an explanation of the alarm cause is provided.

[0145] Backend service: provides API interface, processes frontend requests and business logic.

[0146] Decision engine: Automatically generates maintenance recommendations. This engine can be based on: (1) a preset rule base (including aviation safety regulations, manufacturer maintenance manual recommendations, airline maintenance experience, etc., which can be configured by the user); (2) more advanced knowledge graph reasoning. The engine comprehensively considers the current assessment results (defect severity, aging type and degree), RUL prediction value and its uncertainty to generate specific and explainable maintenance recommendations, such as: "It is recommended to continue service, and the next inspection cycle is shortened to X takeoffs and landings", "It is recommended to replace the vehicle in a planned manner", "Serious damage is detected, and it is recommended to immediately stop the vehicle for repair", etc.

[0147] Report Generation: Automatically generates standardized PDF inspection reports containing detailed assessment results, key image snapshots, RUL prediction details (including confidence intervals), system-generated maintenance recommendations, and corresponding confidence levels or justifications.

[0148] Human-machine collaboration and feedback loop: Provide an interface that allows maintenance engineers or experts to review, correct, or confirm the system's assessment results. Users can feedback the corrected information (such as the confirmed wear level, the determined aging type, the RUL true value at the time of actual replacement, etc.) to the system through a standardized annotation interface (such as providing predefined label options or simple annotation tools). This high-quality feedback data can be automatically transmitted to the model continuous learning module (223). In particular, in combination with active learning strategies, this feedback is preferentially used to optimize the model, forming an efficient human-machine collaborative optimization loop.

[0149] (3) Central database: used to store and manage various data involved in the system operation process. A hybrid database architecture can be used to meet the storage requirements of different data:

[0150] Relational databases (such as PostgreSQL and Oracle): store structured data, such as basic tire information (unique ID, model, batch, size, material, etc.), user accounts, maintenance rule bases, standardized maintenance records, etc.

[0151] Non-relational databases (such as MongoDB and Cassandra): Store semi-structured or document-based data, such as detailed JSON / XML reports generated by each test and complex feature vectors output by the model.

[0152] Dedicated file / object storage (S3-compatible storage, HDFS): used to store large amounts of unstructured file data such as raw and processed images, 3D point clouds, etc.

[0153] Time series databases (such as InfluxDB and TimescaleDB): Specialized for storing time series data of tire condition indicators (such as wear rate, aging index, and RUL prediction value) that change over time, facilitating trend analysis and historical tracing.

[0154] The communication interface provides a standard application programming interface (API), such as a RESTful WebAPI or the industrial OPC UA protocol interface, to enable secure and reliable data interaction between the system and external information systems such as the airline's existing maintenance management system (MMS), enterprise resource planning (ERP), and aircraft health management (AHM) systems. Interactions can include obtaining tire usage information from external systems, pushing test results / assessment reports / RUL predictions / maintenance recommendations to external systems, and receiving test tasks or work order feedback from external systems. This seamlessly integrates the system into the airline's overall maintenance process, forming an information closed loop.

[0155] Collaborative workflow example: A typical inspection process can be roughly summarized as follows:

[0156] (1) The tire enters the collection unit through the positioning platform.

[0157] (2) The system automatically identifies the tire ID (or manually inputs it) and retrieves the historical data of the tire from the database.

[0158] (3) The acquisition unit performs multimodal data acquisition based on tire information and preset strategies (or adaptive adjustments).

[0159] (4) The collected raw data is transmitted to the processing unit in real time or in batches.

[0160] (5) The preprocessing module processes and aligns images and historical data.

[0161] (6) The core deep learning model infers the processed data and outputs the state assessment results and RUL predictions (including uncertainty measures).

[0162] (7) The results are transmitted to the decision-making unit.

[0163] (8) The decision engine generates maintenance recommendations based on the results and rule base / knowledge graph.

[0164] (9) The evaluation results, RUL predictions, and maintenance recommendations are visualized on the user interface, and an alarm is triggered when necessary.

[0165] (10) The operator / expert can review and confirm, and the system will automatically generate a test report.

[0166] (11) Test results, reports, and recommendations are stored in the database and uploaded to external systems such as MMS through interfaces.

[0167] (12) (If enabled) Subsequent maintenance feedback or expert annotation information is used to trigger the model's continuous learning and updating unit.

[0168] Example 2: Portable intelligent detector with integrated sensor;

[0169] This embodiment can also be embodied as a lightweight, portable handheld or mobile detection device, which is more suitable for rapid field inspections, on-wing inspections, or as a supplement to fixed inspection stations. Its features may include:

[0170] Integrated acquisition head: This combines a miniature high-resolution camera module (for capturing macro and micro topography of the tire surface), a compact laser profiling module (for accurately measuring tread profile and wear depth, and supporting immediate or subsequent comparison with initial data), and the necessary distance sensors (for maintaining scanning distance) into a single handheld or movable scanning head. The operator moves the handheld scanning head along the tire surface to collect data.

[0171] Processing unit: The computing power is relatively limited. Two modes can be used:

[0172] Edge computing mode: Integrate edge AI computing modules (such as the NVIDIA Jetson series or Qualcomm Snapdragon platform) locally on the device to run lightweight or compressed / pruned deep learning models, focusing on real-time key defect detection (such as cracks and foreign objects) and preliminary status assessment.

[0173] Device-cloud collaborative model: Portable devices are responsible for data collection and possible preliminary processing / screening, and then transmit the data to cloud servers via wireless networks (5G / Wi-Fi). The cloud platform has more powerful computing resources and can run complete and complex deep learning models for more sophisticated analysis, such as complex aging assessments and RUL predictions that integrate historical data. The cloud also handles model training, updates, and global data management.

[0174] Display and interaction: The device itself can be equipped with a small touch screen to display basic results and brief suggestions, or connected to a dedicated APP on a tablet or smartphone via Bluetooth / Wi-Fi to provide a richer visual interactive interface.

[0175] This combination of portable devices and cloud platforms (end-cloud collaboration) can combine the flexibility and rapid response capabilities of on-site operations with the powerful computing power and data management capabilities of in-depth background analysis.

[0176] Example 3

[0177] Based on the same inventive concept, this embodiment further provides a method for intelligently assessing aircraft tire status and predicting remaining life. This method corresponds to the system described in the first embodiment and includes the following steps:

[0178] Acquiring multimodal data of the tire, wherein the multimodal data is acquired by a multimodal intelligent image acquisition module;

[0179] Processing and analyzing the multimodal data, where the processing and analysis are performed by a data processing and analysis module, to output a condition assessment result and a remaining service life prediction result of the aircraft tire;

[0180] The condition assessment results and remaining service life prediction results are displayed, and maintenance recommendations are generated. This display and generation are performed by the interactive result display and intelligent decision-making module.

[0181] The method for intelligently evaluating the condition of an aircraft tire and predicting its remaining life provided by this embodiment has all the advantages of the system for intelligently evaluating the condition of an aircraft tire and predicting its remaining life provided by the first embodiment.

[0182] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An intelligent aircraft tire condition assessment and remaining life prediction system, characterized by: include: Multimodal intelligent image acquisition module, used to obtain multimodal data of tires; A data processing and analysis module, configured to process and analyze the multimodal data and output a condition assessment result and a remaining service life prediction result of the aircraft tire; The interactive result display and intelligent decision-making module is used to display the condition assessment results and remaining service life prediction results, and generate maintenance recommendations.

2. The system according to claim 1, wherein: The multimodal intelligent image acquisition module includes: An image acquisition unit, used to acquire image data of aircraft tires by combining multiple imaging technologies; The light source system is used to provide controllable and diverse light sources, optimize lighting parameters according to the detection target, and intelligently select light source combinations and adjust brightness according to acquisition requirements; Positioning and attitude adjustment unit, used for precise positioning and rotation platform driven by servo motors, suitable for angular positioning and 360° rotation scanning of tires of different specifications; The adaptive acquisition unit is used to dynamically adjust imaging parameters based on the feedback of preliminary analysis results, tire historical data or identified tire model to enhance the acquisition of specific damage or aging characteristics.

3. The system according to claim 2, characterized in that The image acquisition unit includes: A camera array configured with a number of cameras, arranged as needed, to capture visual images of the tread and sidewall from various preset angles; wherein the preset angles include 0°, ±30°, ±60° relative to the tread centerline, and specific angles relative to the sidewall; A structured light 3D imaging unit, configured to reconstruct 3D point cloud data of the tire surface by using a structured light projector and at least two cameras to work together; Multispectral / hyperspectral imaging unit, used to acquire multi-channel images using a specific wavelength light source array in conjunction with a camera with a filter or a dedicated multispectral camera; A sensing unit, wherein the sensing unit comprises at least one item from the following group: An integrated thermal imaging sensor, mounted on a robotic arm that scans in sync with the tire's rotation, positioned directly above the tire for tread inspections and to the side of the tire for sidewall inspections; and The integrated ultrasonic flaw detection unit uses a special fixture to enable the probe to fit closely to the tire surface at a specific incident angle and scan along the tire circumference.

4. The system according to claim 2, wherein: The adaptive acquisition unit includes: Pre-scanning and region of interest positioning unit, used to automatically identify tire models through scanning and preliminarily locate key areas; Intelligent lighting optimization unit, which automatically calls preset lighting schemes or dynamically adjusts lighting parameters to maximize contrast of specific features based on a tire model database, historical data, or pre-scan image analysis results; Dynamic multi-angle fine scanning unit is used to automatically increase the acquisition angle density or adjust the camera posture for rescanning the identified key areas of interest or suspected defect areas; The adaptive multimodal fusion acquisition unit is used to intelligently decide whether to enable some or all imaging modalities for data acquisition and fusion based on preset detection tasks or real-time analysis feedback, balancing detection efficiency and accuracy.

5. The system according to claim 1, wherein: The data processing and analysis module includes: Multi-source data preprocessing and fusion unit, used for denoising, registration, correction, segmentation, feature extraction of multimodal image data, and processing of tire historical data; The core deep learning model unit is used to perform thermal oxidation aging assessment based on usage time, cumulative damage assessment based on takeoffs and landings, tread profile wear depth quantification, feature decoupling and quantification of tire surface quality analysis, comprehensive state assessment, and dynamic remaining service life prediction and uncertainty quantification through pre-trained deep learning models; The model continuous learning and updating unit is used to support online or incremental learning and continuously optimize the model using new data and feedback.

6. The system according to claim 5, characterized in that The multi-source data preprocessing and fusion unit includes: The first processing unit is used to perform image enhancement preprocessing on the 2D image to highlight subtle defects and texture information on the tire surface. It also uses histogram equalization to improve image contrast and a sharpening filter to enhance edge details. Alternatively, it performs image segmentation first, using a semantic segmentation network to segment different areas of the tread, shoulder, and sidewall and perform feature analysis. The second processing unit is used to build a high-quality three-dimensional model of the tire surface, remove outliers and smooth the 3D point cloud / depth map, accurately align point cloud data from different perspectives or multiple scans, and stitch them together to obtain a complete tire model; The third processing unit is used to access and process historical data related to tires from a central database or through an interface; the historical data includes weather temperature data of the aircraft usage area, historical take-off and landing data, and tire model data; the processing includes cleaning and standardizing the historical data, and calculating the cumulative damage factor and aging equivalent time based on the processed data.

7. The system according to claim 5, characterized in that The deep learning model in the core deep learning model unit adopts a multi-task, multi-modal, spatiotemporal fusion architecture, including: Specific modality feature extractor, used to extract features of different modal data respectively; Feature fusion module, used to fuse features of different modalities at multiple levels; The condition assessment subnetwork is used to output the condition assessment results of the aircraft tire's wear index and aging index based on the fused features; The remaining service life prediction subnetwork is used to combine current state characteristics and historical data to predict the remaining service life of aircraft tires and quantify the uncertainty of the prediction.

8. The system according to claim 7, characterized in that The methods for quantifying the uncertainty of prediction include Bayesian neural network and Monte Carlo method.

9. The system according to claim 1, wherein: The interactive result display and intelligent decision-making module includes: Visual display unit, used to display the evaluation results and remaining service life prediction results in an intuitive way; Intelligent alarm and diagnostic unit, used to trigger graded alarms based on the evaluation results and provide preliminary fault diagnosis information; A maintenance suggestion generation unit is used to generate maintenance suggestions based on the evaluation results and the remaining service life prediction results, combined with the maintenance rule base or knowledge graph; The maintenance plan optimization unit is used to provide optimized maintenance scheduling recommendations based on the overall fleet situation and resource constraints.

10. A method for intelligently evaluating aircraft tire status and predicting remaining life, characterized in that: include: Acquire multimodal data of the tire, process and analyze the multimodal data, and output a condition assessment result and a remaining service life prediction result of the aircraft tire; The condition assessment results and remaining service life prediction results are displayed, and maintenance recommendations are generated.

Images