Train component condition degradation early warning method, system and storage medium
By performing spatiotemporal alignment processing on historical image sequences of train components and current images, multidimensional state features are extracted and degradation curves are fitted, solving the problem of difficulty in identifying early degradation in existing technologies. This enables quantitative prediction and early warning of train components, improving the accuracy of maintenance and the efficiency of resource utilization.
Patent Information
- Application Number
- CN202611306034.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-17
AI Technical Summary
Existing train component inspection technologies struggle to identify early, subtle degradation features and cannot quantitatively predict component performance degradation, leading to high rates of over-maintenance and unplanned repairs.
By acquiring historical image sequences of train components and performing spatiotemporal alignment with current images, multidimensional state features are extracted, state degradation curves are fitted, remaining lifespan is calculated, and early warning information is output.
It enables sensitive detection of early degradation of train components, reduces over-maintenance costs, and improves driving safety and maintenance resource utilization efficiency.
Smart Images

Figure CN123118516A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent inspection technology for rail transit, and in particular to a method, system and storage medium for early warning of the degradation of train components. Background Technology
[0002] With the rapid development of the rail transit industry, the operational safety of trains (such as high-speed trains) is of paramount importance. During long-term high-speed operation, key components of the train undercarriage (such as bogie bolt connections, brake pads, and rubber joints) are inevitably subject to wear, loosening, corrosion, or deformation due to the combined effects of continuous vibration, physical friction, and environmental erosion. Therefore, frequent condition inspections and health monitoring of these critical components are a core means of ensuring safe train operation.
[0003] Currently, the industry commonly uses trackside detection station systems (such as trackside machine vision inspection systems) to capture images of undercarriage components as trains pass by and then uses image recognition algorithms for defect screening. However, existing conventional visual inspection technologies mainly rely on single-frame static image detection. That is, the system judges whether a component has obvious breaks, missing parts, or severe deformation based solely on a single image captured at a given moment. This traditional inspection method exposes the following significant shortcomings in practical engineering applications: First, traditional single-frame static detection methods can often only identify obvious defects that have deteriorated to a certain extent. For gradual, minor degradation of components in the early stages (such as slight geometric displacement, early fatigue cracks, or slowly accumulating surface wear), the system has difficulty detecting them because single-frame images lack a historical benchmark of normal conditions. This results in an extremely low detection rate of early degradation features, making it impossible to eliminate potential faults in their infancy.
[0004] Second, existing technologies view each image acquisition result in isolation, failing to connect the state of the same component across multiple maintenance cycles. This prevents the system from fitting the non-linear laws of component performance degradation and hinders quantitative prediction of the component's remaining useful life (RUL). Consequently, railway maintenance departments are forced to rely on fixed "experience cycles" for maintenance, which not only easily leads to over-maintenance and resource waste but also fails to effectively reduce unplanned repair rates caused by sudden severe deterioration.
[0005] In conclusion, a new early warning method is urgently needed to address the pain points in the current train operation and maintenance process. Summary of the Invention
[0006] The purpose of this invention is to provide a train component condition degradation early warning method, system, and storage medium that can effectively utilize time-series image sequences to achieve quantitative prediction of component degradation trends, so as to automatically detect components with deteriorating performance in advance.
[0007] To achieve the above objectives, the present invention discloses a method for early warning of degradation of train components, comprising: Acquire a historical image sequence consisting of historical images of the target components of the target train at multiple consecutive historical moments, as well as the image to be detected at the current moment; The historical image sequence and the image to be detected are spatiotemporally aligned to obtain an aligned image sequence mapped to the same reference coordinate system. Based on the aligned image sequence, multidimensional state features at multiple time points are extracted for the key regions of the target component, and the multidimensional state features are combined into a multidimensional state feature sequence according to the time sequence. Based on the multidimensional state feature sequence, a state degradation curve characterizing the relationship between the overall degradation degree of the target component and time is fitted; Determine the predicted failure time when the state degradation curve intersects with a preset limit state threshold; and use the current time corresponding to the image to be detected as the time reference to calculate the time difference between the predicted failure time and the current time to obtain the remaining lifespan of the target component. When the remaining lifespan meets the preset warning conditions, a warning message for the state degradation of the target component is output.
[0008] Preferably, the method for spatiotemporally aligning the historical image sequence with the image to be detected includes: Evaluate and select one image from the historical image sequence as a reference image; Structural feature points are extracted from the reference image, and feature point matching is performed between the remaining images in the historical image sequence and the image to be detected and the reference image, respectively. Based on the feature point matching results, the homography transformation matrix or affine transformation matrix is calculated, and the corresponding images are geometrically corrected using the matrix so that the pixel coordinates of the same physical point in each corrected image are consistent, thereby obtaining the aligned image sequence.
[0009] Preferably, in the process of calculating the homography transformation matrix or the affine transformation matrix, a random sampling consensus algorithm is used to minimize the reprojection error in order to obtain the optimal transformation matrix.
[0010] Preferably, the multidimensional state features include at least a combination of the following dimensional indicators: Geometric displacement representing the degree of loosening or deformation of the target component; The rate of change of texture characterizing the surface condition of the target component, such as rust, oil stains, or fatigue cracks; The profile wear index characterizes the degree of physical wear of the target component.
[0011] Preferably, for any image in the aligned image sequence, the Euclidean distance between the feature points of the target component in that image and the corresponding feature points in the reference image is calculated to obtain the geometric displacement at the corresponding time, wherein the reference image is a selected image that serves as the starting point for state comparison; The contrast features of the target component in the corresponding image region are extracted using the gray-level co-occurrence matrix, and the texture change rate at the corresponding time is calculated. The loss ratio of the edge area of the target component in the corresponding image to the reference image is quantified, and the contour wear index at the corresponding time is calculated.
[0012] Preferably, the method for fitting the state degradation curve includes: Using the multidimensional state feature sequence as input, the state degradation curve is fitted using a long short-term memory network model or an exponential smoothing time series prediction model.
[0013] Preferably, the method for outputting state degradation warning information for the target component includes: Pre-set a multi-level early warning mechanism based on the number of remaining maintenance cycles; The calculated remaining lifespan is converted into the number of remaining maintenance cycles and compared with the thresholds at each level in the multi-level early warning mechanism; The corresponding level of warning information is triggered based on the comparison results.
[0014] The present invention also provides a train component condition degradation early warning system, which includes an image sensor for acquiring images of target components of a target train and a data processor communicatively connected to the image sensor. The data processor performs condition degradation early warning of the target component based on the train component condition degradation early warning method described above.
[0015] The present invention also provides a train component condition degradation early warning system, which includes: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for executing the train component condition degradation early warning method as described above.
[0016] The present invention also provides a computer-readable storage medium comprising a computer program that can be executed by a processor to perform the train component condition degradation early warning method as described above.
[0017] Compared with existing technologies, the train component condition degradation early warning method provided by the above technical solution effectively overcomes the perspective and position deviations generated in dynamic train capture scenarios by introducing historical image sequences and performing high-precision spatiotemporal alignment processing with the current image to be detected, ensuring the accuracy of cross-time dimension feature comparison. By extracting multi-dimensional features and fitting condition degradation curves, it breaks the limitation of traditional single-frame image detection that easily misses early minor hidden dangers, and realizes dynamic quantitative tracking of the progressive degradation process of components. Moreover, this method uses the extrapolation intersection point combined with the current actual image acquisition time as a benchmark to accurately calculate the remaining life of the component and trigger an early warning, providing a reliable basis for condition-based maintenance of trains, significantly improving driving safety and effectively reducing over-maintenance costs. Attached Figure Description
[0018] Figure 1 This is a flowchart of the early warning method in an embodiment of the present invention.
[0019] Figure 2 This is a historical image sequence of the workpiece and an image arrangement diagram to be detected in an embodiment of the present invention.
[0020] Figure 3 This is a schematic diagram illustrating the spatiotemporal alignment principle in an embodiment of the present invention.
[0021] Figure 4 This is a schematic diagram of state degradation curve fitting in an embodiment of the present invention.
[0022] Figure 5 This is a schematic diagram illustrating the principle of remaining lifetime calculation in an embodiment of the present invention.
[0023] Figure 6 This is a trend diagram of geometric displacement in an embodiment of the present invention.
[0024] Figure 7 This is a trend chart of texture change rate in an embodiment of the present invention.
[0025] Figure 8 This is a trend chart of the contour wear index in an embodiment of the present invention. Detailed Implementation
[0026] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.
[0027] This embodiment discloses a method for early warning of the degradation of train components, which is used for daily detection of the degree of functional degradation of the bottom components of trains (such as EMUs).
[0028] like Figure 1 The method includes the following steps: S1: Obtain a historical image sequence consisting of historical images of the target component of the target train at multiple consecutive historical moments, as well as the image to be detected at the current moment.
[0029] Specifically, the detection station system will automatically take pictures of the target components under the train each time the train enters the depot or passes by, based on preset trigger signals (such as trigger pulses from wheel axle sensors or infrared beam detectors).
[0030] The system simultaneously records the carriage number, bogie number, and relative position of the component on the car body corresponding to this capture. Using these as an index, it retrieves historical images of the same component from the historical database that have not been fault-determined within several consecutive historical periods (e.g., the past six months or the past ten maintenance cycles), and arranges them in chronological order to form a historical image sequence. Meanwhile, the image captured in this capture is used as the image to be detected at the current moment.
[0031] like Figure 2 The image shows the bolt connection of a bogie at five time points: t1, t2, t3, t4, and the current time tn. The image clearly shows that the surface condition of the part changes gradually over time from clean and slightly contaminated to rust spots and even local cracks. This is exactly what this invention aims to capture and quantify.
[0032] S2: Perform spatiotemporal alignment processing on the historical image sequence and the image to be detected to obtain an aligned image sequence mapped to the same reference coordinate system.
[0033] Because the train's speed, vehicle attitude, and camera triggering timing inevitably vary slightly each time it passes a detection station, images of the same component acquired at different times will show deviations in viewpoint, scale, and even local location. Directly comparing pixels without processing would introduce a large amount of noise unrelated to the component's actual degradation. Therefore, this step maps all frames in the historical image sequence and the current image to be detected to the same reference coordinate system, obtaining an aligned image sequence. This ensures that during subsequent feature extraction, the same physical location point always corresponds to the same pixel coordinates in images from different times.
[0034] S3: Based on the aligned image sequence, extract multidimensional state features at multiple times for the key regions of the target component, and combine the multidimensional state features into a multidimensional state feature sequence according to the time sequence.
[0035] Specifically, after completing the spatiotemporal alignment, for key areas on the component that are prone to degradation (such as bolt connections, friction surfaces, rubber nodes, etc.), the quantitative indicators of the image at each moment relative to the reference image in the three dimensions of geometry, texture, and contour are calculated, and the three are combined into a multidimensional state feature vector at that moment. Then, the feature vectors at each moment are arranged in chronological order to form a multidimensional state feature sequence.
[0036] S4: Based on the multidimensional state feature sequence, a state degradation curve characterizing the relationship between the overall degradation degree of the target component and time is fitted, such as... Figure 4 .
[0037] S5: As Figure 5 The predicted failure time (T_fail) at which the state degradation curve intersects with the preset limit state threshold is determined; and the time difference between T_fail and T_c is calculated using the current time corresponding to the image to be detected as the time base (T_c) to obtain the remaining lifespan of the target component.
[0038] This step combines historical failure data of this type of component with expert experience to pre-set a comprehensive degradation index critical value, i.e., the limit state threshold, that characterizes the functional failure of the component. The state degradation curve obtained in step S4 is extended along the time axis in the future direction, and the time point at which the curve intersects with the limit state threshold is calculated as the predicted failure time.
[0039] Using the current image acquisition time as the time reference, the remaining lifespan of the target component is obtained by subtracting the predicted failure time from the current time.
[0040] S6: When the remaining lifespan meets the preset warning conditions, output the state degradation warning information for the target component.
[0041] The remaining lifespan calculated in step S5 is compared with the pre-set warning conditions. When the remaining lifespan reaches the trigger condition, the corresponding state degradation warning information is output to prompt the maintenance personnel to take corresponding attention, review or repair measures for the component.
[0042] This embodiment employs a technique that combines historical time-series images with current images, aligns them using a unified coordinate system, and then performs multi-dimensional feature extraction. This allows for the continuous tracking and amplified presentation of minute geometric displacements and texture changes in components that occur early in the process and are difficult to detect in a single frame. Furthermore, the introduction of degradation curve fitting and extrapolation prediction mechanisms enables the system to move beyond static judgments of the component's current state and make quantitative predictions of its future remaining lifespan. This transforms traditional post-fault alarms into proactive predictive warnings, providing a quantitative basis for condition-based maintenance of trains.
[0043] In another embodiment, in order to further solve the problem of image perspective and scale deviation caused by differences in train speed, attitude and camera triggering timing during dynamic capture, this embodiment further explains the specific implementation of the spatiotemporal alignment processing in step S2 above.
[0044] In this embodiment, step S2 specifically includes the following sub-steps: S21: As Figure 3 The sharpness, integrity, and occlusion of each frame in the historical image sequence are evaluated, and the frame with the best quality is selected as the reference image.
[0045] Image quality can be evaluated by using the Laplacian operator to calculate image sharpness, combined with factors such as whether there are obstructions or exposure abnormalities in the part area, and the image with the highest score is selected as the benchmark.
[0046] S22: Extract structural feature points from the reference image, and match the feature points of the remaining frames in the historical image sequence and the current image to be detected with the reference image respectively.
[0047] Feature point extraction can employ feature description algorithms such as SIFT (Scale Invariant Feature Transform), SURF (Speed-Up Robust Feature Transform), and ORB (Oriented Fast and Rotated BRIEF), or other robust local feature extraction methods such as Harris corner detection. This application does not impose any restrictions on these methods, as long as the extracted feature points can establish a reliable correspondence between different images.
[0048] S23: Based on the feature point matching results, calculate the homography transformation matrix or affine transformation matrix, and use the matrix to perform geometric correction on the corresponding image.
[0049] Specifically, let the homogeneous coordinates of the feature points on the reference image be ( x, y, 1) The homogeneous coordinates of the corresponding feature points on the image to be registered are ( x', y', 1) The spatial transformation relationship between the two is determined by the homography matrix. H Represented as: ( x', y', 1) = H · ( x, y, 1) ; Where H is a 3×3 matrix. In scenarios where the surface of the component is approximately planar and the relative pose change between the camera and the component is small, an affine transformation matrix can be used instead of the homography matrix for simplified calculation. Both are optional implementations of the geometric transformation matrix described in this application. After obtaining the transformation matrix, coordinate mapping can be performed on each pixel in the corresponding image to ensure that the pixel coordinates of the same physical point in each corrected image remain consistent, thereby obtaining an aligned image sequence.
[0050] In the process of solving the transformation matrix in step S23 above, since the feature point matching results inevitably contain a small number of mismatched points (outside points), directly using all matching points for least squares solution can easily lead to a large deviation in the transformation matrix. Therefore, this embodiment further employs the Random Sample Consensus (RANSAC) algorithm to optimize the transformation matrix solution. Specifically, it calculates candidate transformation matrices by repeatedly randomly sampling some matching point pairs, and counts the number of interior points whose reprojection error falls within the tolerance range under these candidate matrices. The solution with the largest number of interior points and the smallest reprojection error is selected as the final transformation matrix. Its objective function can be expressed as: ; in f(H, x i , y i ) Indicates a point (x i , y i ) Applying the transformation matrix H The resulting mapped coordinates are summed over the set of interior points involved in this round of calculation. This method eliminates the interference of mismatched points on the final transformation matrix solution, improving the robustness of geometric correction.
[0051] Because this embodiment introduces robust estimation techniques such as RANSAC to eliminate interference from mismatched points based on feature point matching, the final geometric transformation matrix can more accurately reflect the real spatial mapping relationship between images, thereby reducing the feature extraction error caused by image misalignment and laying the foundation for the accurate calculation of small state changes in subsequent steps.
[0052] In another embodiment, for the key areas of the target component, the multidimensional state features extracted in this embodiment include at least the following three dimensions: One is the geometric displacement ( ΔD t (), used to characterize the degree to which a component has become loose or deformed; The second is the texture change rate ( T t(), used to characterize the condition of rust, oil stains or fatigue cracks on the surface of a component; The third is the profile wear index ( W t ( ), used to characterize the degree of physical wear and tear on a component due to long-term friction.
[0053] Combining the above three indicators at the same time yields the multidimensional state feature vector F at that time. t = [ ΔD t , T t , W t ].
[0054] Specifically, the geometric displacement is calculated as follows: for any frame in the aligned image sequence, extract the coordinates of several key feature points (such as bolt edge corners, connector boundary inflection points, etc.) on the target component's critical area in that image, and calculate the Euclidean distance between these coordinates and the corresponding feature point coordinates in the reference image, which is then used as the geometric displacement at that moment. Δ D t : ΔD t = √[( x t - x 0)² + ( y t - y 0)²]; in( x t , y t ) represents the coordinates of the feature points in the currently processed image. x 0 , y 0) represents the coordinates of the corresponding feature point in the reference image. A larger value indicates a more significant physical displacement or deformation of the component relative to its initial state. The geometric displacement tends to increase as the detection cycle progresses, such as... Figure 6 As shown.
[0055] The texture change rate is calculated as follows: Texture features such as contrast in key regions of the target part are extracted using the Gray-Level Co-occurrence Matrix (GLCM), and the change ratio T of texture features relative to the reference image is calculated. t : T t = |C t - C 0 | / C 0× 100%; in Ct , C 0 represents the contrast feature value of the gray-level co-occurrence matrix of this region in the current image and the reference image, respectively. When rust spots, oil stains, or microcracks appear on the surface of a component, the unevenness of its local gray-level distribution will increase accordingly, thus manifesting as a significant change in the contrast feature value. The texture change rate tends to increase as the detection cycle progresses, such as... Figure 7 As shown.
[0056] The contour wear index is calculated as follows: extract the edge contour of the key area of the target component, and quantify the loss ratio of the area enclosed by the contour relative to the corresponding area in the reference image. W t : W t = |A t - A 0 | / A 0× 100%; in A t , A 0 represents the area enclosed by the edge contours of the region in the current image and the reference image, respectively. This index can intuitively reflect the degree of material wear caused by long-term friction of the component. The contour wear index tends to increase as the detection cycle progresses, such as... Figure 8 As shown.
[0057] As an example, in the measured data of a bogie bolt connection and brake pads over ten consecutive maintenance cycles, the geometric displacement gradually increased from 0.117 mm in the first cycle to 0.840 mm in the tenth cycle, the texture change rate increased from 1.7% to 27.7%, and the profile wear index increased from 3.3% to 69.4%. These three factors together drove the comprehensive degradation index to increase from 8.6 to 79.1, showing a clear accelerating upward trend. This is consistent with the observed phenomena of increased rust area and increased bolt loosening in the components.
[0058] Because this embodiment quantifies the state of a component from three independent yet complementary dimensions—geometry, texture, and contour—it avoids the potential for missed detections that may occur when relying on only a single feature dimension. For example, early degradation of some components may only manifest as texture changes while geometric displacement is not yet obvious. The fusion of multi-dimensional features enables such early anomalies to be effectively captured, thereby improving the system's ability to detect early minor degradations.
[0059] Another embodiment, the specific method for fitting a state degradation curve that can reflect the nonlinear degradation law of a component based on a multidimensional state feature sequence is as follows: The obtained multidimensional state feature sequence is input in chronological order and fed into a Long Short-Term Memory (LSTM) network model or an exponential smoothing time series prediction model for learning and fitting. This yields a state degradation curve that reflects the evolution of the overall degradation degree of the component over time, such as... Figure 4 As shown.
[0060] In implementations using exponential smoothing models, the following exponential degradation model can be constructed to fit the comprehensive degradation index: y(t) = a·e^(b·t) + c ; in y(t) for t A comprehensive degradation index over time. a , b , c The model parameters to be fitted can be obtained by performing nonlinear least squares fitting on the historical multidimensional state feature sequence.
[0061] As an example, based on the measured data of ten consecutive maintenance cycles given in the above embodiments, the parameters a=60.07, b=0.0801, and c=-55.54 were fitted. The corresponding curves showed a clear upward convex growth pattern as the maintenance cycle number increased, which is consistent with the general rule that the degradation process of components is slow at first and then rapid.
[0062] After obtaining the degradation curve, this embodiment further performs time derivative on the curve to obtain the degradation rate and degradation acceleration, which are used to determine whether the component is currently in a state of uniform degradation or accelerated deterioration. Degradation rate: v(t) = dy(t) / dt = a·b·e^(b·t) ; Degradation acceleration: a(t) = dv(t) / dt = a·b²·e^(b·t) ; when a(t) Continuously positive and v(t) As time increases, it indicates that the component degradation process is accelerating; conversely, if... v(t) If the trend is to stabilize, it indicates that the degradation of the component is in a relatively uniform stage.
[0063] Because this embodiment uses a time-series prediction model that can characterize nonlinear evolution, and further obtains the dynamic indicators of degradation rate and acceleration by differentiation, the system can not only give the current degree of degradation of the component, but also determine whether its degradation process is intensifying, thus providing maintenance personnel with richer decision-making basis than simple threshold judgment.
[0064] The specific method for outputting graded early warning information based on the remaining lifespan in step S6 above is as follows: In this embodiment, a multi-level early warning mechanism based on the number of remaining maintenance cycles is pre-set. The calculated remaining lifespan (expressed in the form of time difference) is converted into the number of remaining maintenance cycles and then compared with the preset multi-level thresholds. Based on the comparison results, the corresponding level of early warning information is triggered.
[0065] As an example, the following three-level warning thresholds can be set: When the remaining maintenance cycle is less than 5 cycles, a yellow alert is triggered, prompting maintenance personnel to pay close attention to the component. When the remaining maintenance cycle is less than 3 cycles, an orange alert is triggered. It is recommended to arrange manual verification or preventative replacement of the component when the train enters the depot next time. When the remaining maintenance cycle is less than 1 cycle, a red alert is triggered, requiring immediate mandatory maintenance to prevent functional failure of components in the next operating cycle.
[0066] It should be noted that the number of warning levels and the corresponding remaining maintenance cycle thresholds for each level can be flexibly adjusted based on the importance of the component, the scheduling cycle of maintenance resources, and the statistical results of historical failure data. For example, more levels or more lenient lead times can be set for components with higher safety levels, and this application does not limit this. The output format of warning information is not limited to software interface display; it can also be achieved through SMS, instant messaging push, automatic generation of maintenance work orders, etc., as long as the warning results can be promptly conveyed to relevant maintenance personnel.
[0067] Because this embodiment adopts a multi-level early warning mechanism based on the number of remaining maintenance cycles, maintenance personnel can take differentiated measures according to the urgency of component degradation. This avoids unnecessary over-maintenance of components that are still in the early stages of degradation, as well as delays in handling components that are close to failure. This improves the utilization efficiency of maintenance resources while ensuring driving safety.
[0068] In summary, the train component condition degradation early warning method provided by this invention solves the problem of inaccurate feature comparison across time periods caused by perspective and scale deviations in dynamic train capture scenarios by acquiring historical image sequences of the target component and the current image to be detected, and performing spatiotemporal alignment processing under a unified coordinate system. By extracting state features of three dimensions—geometric displacement, texture change, and contour wear—from key areas of the component and combining them into a multidimensional state feature sequence, it achieves sensitive capture of early progressive degradation of the component. By inputting the multidimensional state feature sequence into a time-series prediction model to fit the condition degradation curve, and combining the derivative analysis of degradation rate and acceleration, it achieves quantitative characterization of the component degradation trend. Furthermore, by extrapolating the intersection of the degradation curve and the limit state threshold to calculate the remaining lifespan, and triggering multi-level early warnings based on the remaining lifespan, it ultimately realizes the transformation from post-fault alarm to pre-fault predictive early warning.
[0069] The present invention also discloses a train component condition degradation early warning system, which includes an image sensor for acquiring images of target components of a target train and a data processor communicatively connected to the image sensor. The data processor performs condition degradation early warning of the target component based on the train component condition degradation early warning method in the above embodiments.
[0070] This invention also discloses another train component condition degradation early warning system, which includes one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors. The programs include instructions for executing the train component condition degradation early warning method as described above. The processor may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, used to execute relevant programs to implement the functions required by the modules in the train component condition degradation early warning system of this application embodiment, or to execute the train component condition degradation early warning method of this application method embodiment.
[0071] This invention also discloses a computer-readable storage medium comprising a computer program executable by a processor to perform the train component condition degradation early warning method described above. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating one or more available media. The available medium can be read-only memory (ROM), random access memory (RAM), or magnetic media, such as floppy disks, hard disks, magnetic tapes, magnetic disks, or optical media, such as digital versatile discs (DVDs), or semiconductor media, such as solid-state disks (SSDs).
[0072] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned train component condition degradation early warning method.
[0073] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
Claims
1. A method of train component condition degradation early warning, characterized in that, include: Acquire a historical image sequence consisting of historical images of the target components of the target train at multiple consecutive historical moments, as well as the image to be detected at the current moment; The historical image sequence and the image to be detected are spatiotemporally aligned to obtain an aligned image sequence mapped to the same reference coordinate system. Based on the aligned image sequence, multidimensional state features at multiple time points are extracted for the key regions of the target component, and the multidimensional state features are combined into a multidimensional state feature sequence according to the time sequence. Based on the multidimensional state feature sequence, a state degradation curve characterizing the relationship between the overall degradation degree of the target component and time is fitted; Determine the predicted failure time when the state degradation curve intersects with a preset limit state threshold; and use the current time corresponding to the image to be detected as the time reference to calculate the time difference between the predicted failure time and the current time to obtain the remaining lifespan of the target component. When the remaining lifespan meets the preset warning conditions, a warning message for the state degradation of the target component is output.
2. The train component condition degradation early warning method of claim 1, wherein, The method for spatiotemporally aligning the historical image sequence with the image to be detected includes: Evaluate and select one image from the historical image sequence as a reference image; Structural feature points are extracted from the reference image, and feature point matching is performed between the remaining images in the historical image sequence and the image to be detected and the reference image, respectively. Based on the feature point matching results, the homography transformation matrix or affine transformation matrix is calculated, and the corresponding images are geometrically corrected using the matrix so that the pixel coordinates of the same physical point in each corrected image are consistent, thereby obtaining the aligned image sequence.
3. The train component condition degradation early warning method of claim 2, wherein, In the process of calculating the homography transformation matrix or the affine transformation matrix, the random sampling consensus algorithm is used to minimize the reprojection error in order to obtain the optimal transformation matrix.
4. The method of claim 1, wherein, The multidimensional state features include at least a combination of the following dimensional indicators: Geometric displacement representing the degree of loosening or deformation of the target component; The rate of change of texture characterizing the surface condition of the target component, such as rust, oil stains, or fatigue cracks; The profile wear index characterizes the degree of physical wear of the target component.
5. The train component condition degradation early warning method of claim 4, wherein, For any image in the aligned image sequence, the Euclidean distance between the feature points of the target component in that image and the corresponding feature points in the reference image is calculated to obtain the geometric displacement at the corresponding time. The reference image is the selected image that serves as the starting point for state comparison. The contrast features of the target component in the corresponding image region are extracted using the gray-level co-occurrence matrix, and the texture change rate at the corresponding time is calculated. The loss ratio of the edge area of the target component in the corresponding image to the reference image is quantified, and the contour wear index at the corresponding time is calculated.
6. The method of claim 1, wherein, The fitting method for the state degradation curve includes: Using the multidimensional state feature sequence as input, the state degradation curve is fitted using a long short-term memory network model or an exponential smoothing time series prediction model.
7. The method of claim 1, wherein, Methods for outputting early warning information on the state degradation of the target component include: Pre-set a multi-level early warning mechanism based on the number of remaining maintenance cycles; The calculated remaining lifespan is converted into the number of remaining maintenance cycles and compared with the thresholds at each level in the multi-level early warning mechanism; The corresponding level of warning information is triggered based on the comparison results.
8. A train component condition degradation early warning system, characterized in that, The system includes an image sensor for acquiring images of target components of a target train and a data processor communicatively connected to the image sensor. The data processor performs a condition degradation warning for the target component based on the train component condition degradation warning method according to any one of claims 1 to 7.
9. A train component condition degradation early warning system, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the train component condition degradation early warning method as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It includes a computer program that can be executed by a processor to perform the train component condition degradation early warning method as described in any one of claims 1 to 7.
