A method for performance testing of a cvt belt

Abstract

The present application relates to the technical field of CVT belt test, and discloses a performance test method for CVT belt. The method comprises the following steps: acquiring tension data and vibration data of the CVT belt during operation; then, dividing the CVT belt into multiple test sub-regions according to the tension data and the vibration data; then, calculating performance indexes of each test sub-region based on the test sub-regions, wherein the performance indexes comprise a tension change index and a vibration amplitude index; constructing a correlation graph between the multiple test sub-regions, wherein the test sub-regions are taken as vertices, and the performance difference values between adjacent test sub-regions are taken as edges; determining a test order according to the correlation graph, and driving a test device to perform performance test on the multiple test sub-regions according to the test order. The method can accurately capture the performance difference of different regions of the CVT belt, optimize the test process, improve the accuracy and efficiency of the test, solve the limitations of traditional overall test, and meet the performance test requirements of the CVT belt.

Description

Technical Field

[0001] The present invention relates to the technical field of CVT belt testing, and specifically to a performance testing method for CVT belts. Background Art

[0002] With the continuous development of the automotive industry, continuously variable transmissions (CVTs) are increasingly widely used in the field of passenger cars due to their smooth shifting experience and good fuel economy. As the core transmission component in a CVT, the performance of the CVT belt directly affects the operation stability and service life of the entire transmission. During the production and later maintenance of CVT belts, performance testing is an important link to ensure their quality. Through effective performance testing, potential defects in the belt can be detected in a timely manner, avoiding transmission failures caused by belt failures and thus ensuring vehicle driving safety.

[0003] In the industry, the performance testing methods for CVT belts mainly focus on overall testing, that is, uniformly detecting the performance of the entire CVT belt to obtain overall data such as the tension and vibration of the belt, and judging whether the belt performance meets the standards based on these overall data. However, during the actual operation of the CVT belt, due to differences in structural design, force distribution, and wear conditions, the performance of different regions varies significantly. For example, the area where the belt contacts the pulley bears a large pressure for a long time, and abnormal tension fluctuations are likely to occur; while the vibration of the non-contact area in the middle of the belt is more affected by the dynamic load during the transmission process. Using the overall testing method, the performance differences of different regions of the belt cannot be accurately captured, and there will often be a situation where "the overall is qualified but there are local defects". These local defects will gradually expand during long-term use, ultimately leading to the failure of the entire CVT belt and posing a hidden danger to the operation of the transmission.

[0004] There are also deficiencies in data processing and test process planning in existing testing methods. After obtaining the tension and vibration data, traditional methods only perform simple statistical analysis on the data, lacking in-depth data mining and regional correlation analysis, and it is difficult to establish the relationship between the performances of different regions. In terms of test sequence arrangement, most use random testing or testing in a fixed order, without considering the impact of performance differences in different regions on test efficiency, resulting in problems such as repeated testing or missing key regions during the test process. This not only increases the test time and cost, but may also affect the accuracy of test data due to unreasonable test sequence. In addition, although some testing methods attempt to perform zonal testing on the belt, the zonal method is relatively rough, mostly based on simple division according to the physical structure of the belt, without dynamically dividing the zones in combination with the data characteristics during actual operation. The divided zones cannot accurately reflect the true performance state of different parts of the belt, and it is still difficult to achieve the goal of accurate testing.

[0005] As CVT technology develops towards higher speeds and higher loads, the performance requirements for CVT belts are becoming increasingly stringent. The limitations of existing testing methods are becoming more and more apparent, failing to meet the industry's demand for precise and efficient performance testing of CVT belts. Therefore, there is an urgent need for a CVT belt performance testing method that can accurately partition and efficiently plan the testing process to solve the problems existing in the current technology. Summary of the Invention

[0006] The purpose of this invention is to provide a performance testing method for CVT belts to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a performance testing method for CVT belts, the method comprising:

[0008] Acquire tension and vibration data of the CVT belt during operation;

[0009] Based on the tension data and the vibration data, the CVT belt is divided into multiple test sub-regions;

[0010] Based on the multiple test sub-regions, the performance index of each test sub-region is calculated, and the performance index includes the tension change index and the vibration amplitude index.

[0011] Construct an association graph among the multiple test sub-regions, where the multiple test sub-regions are vertices and the performance difference between adjacent test sub-regions is the edge;

[0012] The test order is determined based on the association diagram, and the test equipment is driven to perform performance tests on the multiple test sub-regions according to the test order.

[0013] Preferably, acquiring tension and vibration data of the CVT belt during operation includes:

[0014] The tension and vibration sequences of the CVT belt are collected by sensors within a preset test duration.

[0015] The tension sequence and the vibration sequence are preprocessed to generate standardized tension vectors and vibration vectors;

[0016] Based on the standardized tension vector and the standardized vibration vector, the potential defect locations of the CVT belt are identified.

[0017] Preferably, dividing the CVT belt into multiple test sub-regions includes:

[0018] Based on the location of the potential defects, the high-concern area and low-concern area of ​​the CVT belt are determined;

[0019] Based on the high-interest area and the low-interest area, the CVT belt is divided into a main test sub-area and a secondary test sub-area.

[0020] In areas not covered by the main test sub-region and the secondary test sub-region, an auxiliary test sub-region is determined. The main test sub-region, the secondary test sub-region, and the auxiliary test sub-region constitute a test region set.

[0021] Preferably, the calculation of performance metrics for each test sub-region includes:

[0022] For each test sub-region in the set of test regions, extract the tension change characteristics and vibration change characteristics of the corresponding region;

[0023] Calculate the tension change index based on the described tension change characteristics;

[0024] Calculate the vibration amplitude index based on the vibration change characteristics;

[0025] By combining the tension change index and the vibration amplitude index, a comprehensive performance score is generated for each test sub-region.

[0026] Preferably, constructing the association graph between the plurality of test sub-regions includes:

[0027] For adjacent first and second test sub-regions in the test region set, identify the region type of the first and second test sub-regions. The region type includes one of the main test sub-region, secondary test sub-region, and auxiliary test sub-region.

[0028] Based on the identified region type, determine the first performance reference value of the first test sub-region and the second performance reference value of the second test sub-region;

[0029] Calculate the difference between the first performance reference value and the second performance reference value, and use the difference as the performance difference value between the first test sub-region and the second test sub-region.

[0030] Preferably, determining the test order based on the correlation diagram includes:

[0031] Analyze the topology of the association graph to generate a minimum path sequence;

[0032] Based on the minimum path sequence, the test priorities of each test sub-region in the test region set are arranged.

[0033] The test order is output based on the test priority.

[0034] Preferably, driving the testing equipment to perform performance tests on the plurality of test sub-regions according to the test sequence includes:

[0035] According to the test order, the real-time test parameters of each test sub-region are obtained sequentially;

[0036] The real-time test parameters are input into the performance analysis model, and the test output results are generated through the performance analysis model.

[0037] Based on the test output results, adjust the operating parameters of the test equipment.

[0038] Preferably, adjusting the operating parameters of the testing equipment includes:

[0039] Monitor the deviation between the test output results and the preset benchmark value;

[0040] When the deviation exceeds the threshold, the parameter update mechanism is triggered;

[0041] The test intensity or test frequency of the test equipment is corrected through the parameter update mechanism.

[0042] Preferably, the trigger parameter update mechanism includes:

[0043] Collect historical test output results and construct residual data sequences;

[0044] The residual data sequence is input into an adaptive network, and adjustment parameters are generated through the adaptive network.

[0045] The operating parameters of the test equipment are updated using the adjusted parameters.

[0046] Preferably, the method further includes:

[0047] After completing the performance tests of all test sub-regions, integrate the test output results;

[0048] Based on the integrated test output results, an overall performance report of the CVT belt is generated;

[0049] The overall performance report is stored for subsequent analysis.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This performance testing method for CVT belts overcomes the limitations of traditional overall testing, which cannot accurately capture local performance differences, by acquiring tension and vibration data during CVT belt operation and dividing the belt into multiple test sub-regions based on this data. Because the division of test sub-regions is based on key performance data from actual belt operation, rather than solely relying on physical structure, each sub-region accurately corresponds to different parts of the belt exhibiting different performance characteristics during operation. This more realistically reflects the actual performance status of different areas of the belt, avoiding situations where "overall performance is acceptable but local defects exist." It allows testers to clearly understand the performance of each sub-region, thereby achieving precise testing of CVT belt performance.

[0052] After calculating the tension change and vibration amplitude indices for each test sub-region, specific quantitative performance results can be generated for each sub-region, rather than the vague overall performance judgment found in traditional testing. These specific performance indicators can intuitively show the performance of each sub-region in terms of tension stability and vibration control, enabling testers to quickly locate sub-regions with abnormal performance and clarify the specific type of abnormality. Whether it is excessive tension fluctuation or excessive vibration amplitude, it can be accurately identified, providing a clear direction for subsequent defect analysis and maintenance, and avoiding performance misjudgments caused by overall data masking local problems.

[0053] By constructing a correlation graph between multiple test sub-regions, with sub-regions as vertices and performance differences between adjacent sub-regions as edges, the relationships between the performance of different sub-regions can be established, enabling a holistic correlation analysis of belt performance. The correlation graph clearly shows the performance connection between adjacent sub-regions; for example, whether an abnormal tension change in one sub-region affects the vibration amplitude in adjacent regions. This correlation analysis helps to understand the changing patterns of belt performance from a holistic perspective, discovering performance interactions between regions that are difficult to detect in traditional testing, and thus more comprehensively evaluating the overall performance of CVT belts, avoiding the one-sided evaluation caused by focusing only on a single region while ignoring the correlation between regions.

[0054] Determining the test sequence based on the correlation diagram and driving the testing equipment to perform performance tests according to this sequence can significantly improve the efficiency and accuracy of the tests. Since the test sequence is determined based on the performance difference values ​​of adjacent sub-regions, rather than a random or fixed order, the testing process can prioritize areas with significant performance differences. These areas are often key parts of the belt with weak performance or potential defects. Prioritizing testing can quickly identify core problems, reduce unnecessary testing steps, and save testing time and costs. At the same time, the test sequence based on the correlation diagram can avoid interference between regions caused by an unreasonable test order. For example, when testing a sub-region with high vibration amplitude, the accuracy of the vibration data in that region will not be affected by testing adjacent high-tension regions first, ensuring the reliability of the test data and allowing the final test results to more accurately reflect the performance status of each sub-region of the belt.

[0055] The entire testing methodology, from data acquisition, region division, and index calculation to correlation graph construction and test sequence planning, forms a complete closed-loop testing process. Each step is closely linked, and the core data-driven approach is consistently applied, making the testing process more scientific and systematic. Compared to traditional testing methods, this method not only improves the accuracy of CVT belt performance testing but also optimizes the testing process and increases testing efficiency. It can better adapt to the higher requirements for belt performance testing brought about by the development of CVT technology, providing more effective technical support for the quality control and subsequent maintenance of CVT belts, and helping to ensure the stable operation of continuously variable transmissions. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the working principle of the performance testing method for CVT belts described in this invention.

[0057] Figure 2 A flowchart for obtaining CVT belt tension and vibration data;

[0058] Figure 3 A flowchart for calculating the performance metrics for each test sub-region;

[0059] Figure 4 This is a flowchart for determining the test order based on the association diagram. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Please see Figure 1This invention provides a performance testing method for CVT belts, the method comprising:

[0062] During belt operation, tension and vibration data are collected in real time. Based on the feature analysis of the collected tension and vibration data, the continuously running CVT belt is dynamically divided into multiple test sub-regions with independent characteristics. For each sub-region, key performance indicators are calculated, including tension change indicators that quantify tension fluctuation characteristics and vibration amplitude indicators that characterize vibration intensity. A correlation graph model is constructed with all test sub-regions as vertices and performance differences between adjacent sub-regions as edges. Based on the topological properties of this correlation graph, the optimal test path sequence is calculated, and the testing equipment is driven to perform performance tests on each sub-region sequentially according to the determined test order.

[0063] Example 1: See Figure 2 Under preset test conditions and durations, high-precision tension sensors are used to collect tension data in real time along the belt length at equal intervals or specific key points. Tension sensors typically employ resistance strain gauge or piezoelectric principles, and their installation must ensure stable contact with the belt surface without significantly interfering with its normal operation. The acquired raw tension signal is a continuously varying analog voltage or current signal over time. This signal is converted into a discrete digital time series, i.e., a tension sequence, by an analog-to-digital converter. This sequence contains dynamic tension fluctuations caused by load changes, friction, and belt deformation during operation. Simultaneously, vibration sensors, such as ICP-type accelerometers, are used to collect vibration signals generated by the belt during operation at the same or adjacent locations corresponding to the tension measurement points. The vibration sensors must be securely installed to accurately capture belt vibrations in three-dimensional space, and their sensitive axis direction should be adjusted according to the main vibration modes. The vibration signals are also converted from analog to digital to generate a corresponding vibration sequence, which reflects the belt's structural vibration, impact, and potential abnormal shaking characteristics.

[0064] The raw tension and vibration sequences typically contain various noise and interference components. These noises may originate from electromagnetic interference, sensor background noise, minor impacts from the mechanical transmission system, or environmental vibrations. The first step in preprocessing is to smooth the tension sequence using a low-pass digital filter. The filter's cutoff frequency is set based on the highest characteristic frequency of the CVT belt during normal operation, aiming to filter out high-frequency noise components far exceeding this characteristic frequency while retaining the low- and mid-frequency useful signals reflecting the true tension changes of the belt. After filtering, the waveform of the tension sequence becomes smoother, and abrupt noise spikes are effectively suppressed. Subsequently, linear normalization is performed on the filtered tension sequence, mapping its amplitude range to between zero and one, eliminating the differences in data dimensions caused by variations in sensor sensitivity or gain settings in the original signal, making subsequent processing more stable. For transient data gaps that may exist in the sequence due to instantaneous signal loss or abnormal interference, linear interpolation or spline interpolation methods are used to fill in the gaps, ensuring the continuity of the sequence.

[0065] The preprocessing procedure for vibration sequences is similar to that for tension sequences, but the parameters differ. Since vibration signals typically contain richer frequency components, a bandpass digital filter is first applied. The passband frequency range of this filter is set based on the typical vibration spectrum of a CVT belt, aiming to retain characteristic frequency bands closely related to belt performance while suppressing low-frequency drift and high-frequency noise outside the passband. The filtered vibration sequence then undergoes amplitude normalization, typically adjusting its root mean square value to a standard reference level to eliminate the influence of differences in test batches or sensor sensitivity, ensuring comparability of vibration data acquired at different times and under different conditions.

[0066] After preprocessing, two sets of standardized data vectors are obtained: a standardized tension vector and a standardized vibration vector. The number of data points in each vector is consistent with the total number of sampling time points, forming the data foundation for subsequent analysis. Based on these standardized vectors, anomaly detection algorithms are executed to identify potential defect locations on the belt. A common method is based on statistical thresholds: calculating statistics such as the mean and standard deviation of the tension or vibration vector within a sliding time window, and setting a threshold range (e.g., mean plus or minus three standard deviations). The spatial locations of belt points that consistently fall outside this range are marked as potential anomalies. Another more advanced method uses unsupervised machine learning models, such as the Isolation Forest algorithm or an autoencoder, to learn and reconstruct features from the entire vector, classifying sample points with reconstruction errors significantly higher than normal as anomalies. These marked anomalies are often not isolated in space; they tend to form clusters.

[0067] Based on the spatial distribution of identified potential defect locations, the area is divided. First, the spatial density of these anomalies is calculated, for example, using kernel density estimation. On the belt length axis, continuous segments with a density exceeding a preset threshold are identified as high-concern areas; these areas have a higher probability of defects and require focused monitoring. Segments with low density or no anomalies are considered low-concern areas. The boundaries between high-concern and low-concern areas are determined, and based on this, the CVT belt is divided into main test sub-regions and secondary test sub-regions. The main test sub-region precisely covers all identified high-concern areas, and its extent is typically extended outwards by a small buffer zone to cover possible transition areas. The secondary test sub-regions cover belt segments identified as low-concern areas. Segments that are neither classified as high-concern nor low-concern areas, or small transitional areas between clearly defined boundaries, are further divided into auxiliary test sub-regions. Finally, the entire CVT belt is completely and non-overlappingly covered by these main test sub-regions, secondary test sub-regions, and auxiliary test sub-regions, forming a complete set of test areas.

[0068] Example 2: See Figure 3 For each test sub-region in the test area set, a performance index calculation process is executed. Each test sub-region corresponds to its own standardized tension vector and standardized vibration vector, which are generated in the preprocessing stage. First, feature extraction is performed. For the standardized tension vector, its time-varying pattern is analyzed to extract key features that quantify tension fluctuation characteristics. These tension variation features include, but are not limited to: a measure of the dispersion of the tension sequence within a specific time window, reflecting the degree to which the sequence deviates from the average level; the numerical span between the highest and lowest points in the sequence, used to capture the extreme range of tension changes; the number of peaks exceeding a set threshold and their average intensity, used to identify sudden tension impacts; the steepness of the rising or falling edges in the sequence waveform, reflecting the rate characteristics of tension change; the location of the frequency point where the energy distribution is most concentrated after frequency domain transformation, revealing the main frequency components of tension fluctuations; and the estimated repetition frequency of the sequence's periodic fluctuations. These tension variation features characterize the dynamic behavior of tension within the test sub-region from different perspectives.

[0069] For standardized vibration vectors, feature extraction is also performed to obtain key features characterizing vibration amplitude and energy properties. These vibration variation features include, but are not limited to: the root mean square of the sum of squares of all data points in the vibration sequence, reflecting the overall level of vibration energy; the vertical distance between the highest peak and lowest trough in the sequence, visually indicating the magnitude of vibration amplitude; the average of the absolute values ​​of all data points in the sequence, providing an average intensity reference for vibration amplitude; the energy value obtained by integrating the sequence within a specific preset frequency band (which is preset according to the belt material or structural characteristics) after frequency domain transformation, used to capture the vibration energy of specific frequency components; the number of positive or negative pulses exceeding a set threshold in the sequence and their cumulative energy, used to identify abnormal impact events; and the number of times the sequence waveform crosses zero per unit time, used to roughly estimate the main frequency components of vibration. These vibration variation features collectively depict the intensity, energy distribution, and abnormal patterns of vibration within the test sub-region.

[0070] After successfully extracting tension and vibration change features, the next step is index calculation. For the tension change index calculation, a predefined tension change index calculation function is used. This function receives the extracted tension change features as input parameters. This function can be a pre-configured mathematical expression, such as a weighted summation of features, where the weight coefficients are pre-set based on the importance of each feature to the sensitivity of tension fluctuations; it can also be a rule-based mapping table that maps combinations of different feature values ​​to specific index value ranges; or it can be a trained lightweight machine learning model (such as a decision tree or linear regression model) that learns the non-linear relationship between tension change features and the final index value from historical data. Regardless of the specific form used, the core function is to synthesize multiple tension change features into a single, dimensionless tension change index value, which is used to quantify the overall degree or abnormal level of tension fluctuations in the test sub-region.

[0071] For the calculation of the vibration amplitude index, a predefined vibration amplitude index calculation function is used. This function receives the extracted vibration change features as input parameters. Its implementation is similar to that of the tension change index calculation function, and can be a weighted combination, a regular mapping, or a simple model. The task of this function is to integrate multiple vibration change features into a single, dimensionless vibration amplitude index value, which is used to quantify the overall level or degree of abnormality of the vibration intensity in the test sub-region.

[0072] After obtaining the tension variation index and vibration amplitude index values ​​separately, they need to be integrated into a comprehensive evaluation. For this purpose, a pre-defined comprehensive performance score generation model is invoked. This model receives the tension variation index and vibration amplitude index as its main inputs. The model is designed to balance the impact of tension fluctuation and vibration intensity on the overall performance of the belt. A common implementation method is weighted fusion: assigning a weight coefficient (with a total weight of 1) to the tension variation index and vibration amplitude index respectively, and then summing the two index values ​​to obtain the comprehensive performance score. The weight coefficients may be set based on experience or dynamically adjusted according to the belt type and test objectives. Another implementation method may employ a fuzzy logic rule system: defining fuzzy membership functions for the input indices (such as "high tension fluctuation" and "medium vibration intensity") and a set of "IF-THEN" rule bases, and deriving a clear comprehensive performance score value through fuzzy inference. Alternatively, a simple nonlinear function (such as the sigmoid function) can be used to transform and combine the two indices. Regardless of the method used, the output of the model is a comprehensive performance score value representing the overall performance status of the test sub-region. A higher score generally indicates that the performance of that sub-region is closer to ideal or normal; a lower score may indicate a greater risk of potential problems or performance degradation. This process is performed independently on each test sub-region within the test region set, ultimately yielding a comprehensive performance score for each sub-region.

[0073] After calculating the performance metrics of all test sub-regions and obtaining their respective comprehensive performance scores, the association graph is constructed. The association graph is an undirected weighted graph structure. The set of vertices in the graph consists of all test sub-regions in the test region set, with each vertex uniquely representing a test sub-region. The edges of the graph represent the spatial adjacency relationships between test sub-regions. The process of constructing edges is as follows: traverse the entire test region set and identify all pairs of spatially adjacent sub-regions. For each identified pair of adjacent sub-regions, such as the first and second test sub-regions spatially connected, calculate the performance difference between them as the weight of the connecting edge.

[0074] Before calculating the performance difference value, it is necessary to identify the region type attribute of each adjacent sub-region. The region type was determined in the division stage of Example 1, including three types: main test sub-region, secondary test sub-region, or auxiliary test sub-region. Each region type is associated with a preset performance reference value. Generally, the main test sub-region has the highest preset performance reference value because it covers high-concern areas, reflecting a higher expected problem risk; the secondary test sub-region covers low-concern areas, so its preset performance reference value is the lowest, reflecting a lower expected problem risk; the preset performance reference value of the auxiliary test sub-region is between the two. These preset values ​​can be a specific numerical value or a relative level (such as high, medium, low). In the latter case, they need to be mapped to numerical values ​​for calculation.

[0075] For adjacent first and second test sub-regions, a preset performance reference value mapping table is queried according to their respective region types to obtain the first performance reference value corresponding to the first test sub-region and the second performance reference value corresponding to the second test sub-region. Then, the difference between these two performance reference values ​​is calculated. The difference can be calculated as a simple absolute difference: |first performance reference value - second performance reference value|; or as a relative difference rate: (|first performance reference value - second performance reference value|) / ((first performance reference value + second performance reference value) / 2)*100%. The value obtained by choosing one of these methods is the performance difference value between the adjacent sub-region pair. Finally, in the association graph, an undirected edge is added between the vertex representing the first test sub-region and the vertex representing the second test sub-region, and the calculated performance difference value is used as the weight value of this edge. This process is repeated on all identified adjacent sub-region pairs until a complete association graph structure is constructed. The vertex set of this graph covers all test sub-regions, and the edge set fully expresses their spatial adjacency relationships and the preset reference performance difference degree based on the region type.

[0076] Example 3: See Figure 4 After the association graph is constructed, it uses each test sub-region in the test region set as a vertex and the performance difference between adjacent sub-regions as the edge weight. Analyzing the topology of the association graph is a key step in determining the test order. The association graph is an undirected weighted graph where the total number of vertices equals the number of test sub-regions, and the edge weights represent the degree of difference in preset performance reference values ​​between adjacent regions. The goal of topology analysis is to generate a path sequence that covers all vertices, which must minimize the sum of edge weights accumulated during traversal, thereby optimizing testing efficiency. The generation of the minimum path sequence uses a graph theory-based shortest path algorithm, specifically an improved Dijkstra's algorithm. This algorithm initializes from a selected starting vertex, which is usually specified as a vertex of the region type that is the main test sub-region, to prioritize the coverage of potentially high-interest regions.

[0077] During the algorithm initialization phase, a distance estimate is assigned to each vertex. The distance estimate of the starting vertex is set to zero, and the distance estimates of all other vertices are set to infinity. Simultaneously, a priority queue is maintained to store vertices and their current distance estimates. Initially, the priority queue contains only the starting vertex. Then, the iterative processing begins: the vertex with the smallest distance estimate is extracted from the priority queue and designated as the current vertex. For each adjacent vertex of the current vertex (i.e., vertices directly connected by edges), the potential distance from the starting vertex to that adjacent vertex via the current vertex is calculated. This step involves a distance update operation, the core formula of which is as follows:

[0078]

[0079] In this formula: This represents the current estimate from the starting vertex to the next vertex. The shortest cumulative distance, Indicates the currently processed vertex. Distance estimate, Representing an edge The weight of the vertex With vertex The min function is used to compare and update the performance difference values ​​between them. A smaller value ensures path optimization.

[0080] If the calculated potential distance is less than the existing distance estimate of an adjacent vertex, the distance estimate of that adjacent vertex is updated, and its position in the priority queue is added or updated. This iterative process continues until the priority queue is empty or all vertices have been visited. Ultimately, the shortest path distance from the starting point to each vertex is determined. Based on these distance values, a sequence that visits all vertices is generated by backtracking. This sequence satisfies the minimization of the total weight sum and is called the minimum path sequence. The minimum path sequence directly specifies the vertex visiting order; the vertices visited first in the sequence correspond to the earlier tested sub-regions.

[0081] Based on the minimum path sequence, the test priorities of each test sub-region in the test region set are arranged. The positional order in the sequence is directly mapped to a priority value, with higher-priority test sub-regions appearing earlier. For example, the sub-region corresponding to the first vertex in the sequence has the highest priority, decreasing sequentially. This mapping ensures that high-priority sub-regions (such as the main test sub-region) are processed first. The priority arrangement process is accomplished through a simple index assignment, converting the sequence index into a priority score, with higher scores indicating earlier test order. Finally, a test order list sorted in descending order of priority is output. This list is structured data, containing unique identifiers for the test sub-regions and their priority scores, facilitating reading and execution by the test equipment.

[0082] Following the output test sequence, the testing equipment initiates performance testing operations for each test sub-region. The testing equipment includes tension sensors, vibration sensors, a data acquisition module, and adjustable actuators such as motors or vibrators. The equipment drive process follows a sequential order, activating the test programs for each target sub-region in turn. During the testing of each target sub-region, test parameters for that region are acquired in real time. These parameters are dynamically acquired through sensors, reflecting the current operating status, including instantaneous tension values ​​(in Newtons), vibration acceleration values ​​(in meters per second squared), ambient temperature (in degrees Celsius), and belt speed (in meters per second). Parameter acquisition is performed within a preset time window, for example, sampling 100 data points per second to form a real-time parameter vector. All acquisition operations are synchronized through the data acquisition module to ensure timestamp alignment and data integrity.

[0083] The acquired real-time test parameter set is immediately fed into a pre-trained performance analysis model. This model, deployed on an embedded system or edge computing device, employs a convolutional neural network architecture, and its training data comes from historical CVT belt test datasets. The model's input layer receives a parameter vector, including tension, vibration, temperature, and velocity values. Hidden layers perform feature extraction and nonlinear transformations, and the output layer generates the test output results. The test output results are a multi-dimensional vector containing key performance indicators such as a health score (ranging from 0 to 1, with 1 being optimal), a failure probability estimate (in percentage form), and a remaining life prediction (in hours). The model inference process is executed in real time, and the output results are generated within milliseconds. This output is used to quantify the immediate performance status of the current test sub-region.

[0084] Based on the test output generated by the performance analysis model, it is determined whether adjustments to the operating parameters of the test equipment are needed. The decision is based on a comparison of the output results with preset benchmark values. These benchmark values ​​are stored in a configuration file and include pass / fail thresholds for various indicators, such as a health score below 0.6 or a failure probability exceeding 20%. The deviation between the current output result and the corresponding benchmark value is calculated, such as an absolute difference or percentage offset. If the deviation exceeds a set threshold (e.g., a health score deviation greater than 0.2), a parameter adjustment mechanism is triggered. The adjustment process involves sending control commands to the execution unit of the test equipment to dynamically modify the operating parameters. Adjustable parameters include test intensity (e.g., the force amplitude output by the vibrator, in Newtons) and test frequency (e.g., the data sampling rate, in Hertz). Modifications are made according to the direction of the output deviation; for example, increasing the test intensity when the health score is low to obtain more detailed data. After adjustment, the equipment continues to test subsequent sub-regions sequentially. The entire implementation process emphasizes real-time feedback and adaptive control to ensure efficient test workflow. Operational details such as model parameter loading and sensor calibration are handled automatically in the background without manual intervention.

[0085] Example 4: During the performance testing of the test sub-regions by the test equipment according to the test sequence, the monitoring system continuously tracks the test output results generated by the performance analysis model. This output result is typically in the form of structured data, such as records containing fields like health score and fault probability prediction. Preset benchmark values ​​are stored in the configuration database, containing reference ranges for each indicator under different sub-region types (e.g., the lower limit of the health score benchmark for the main test sub-region is 0.7). The real-time monitoring module performs a comparison every second: extracting the output result of the current test sub-region, retrieving the corresponding benchmark value, and calculating the absolute deviation of key indicators. The deviation is defined as the absolute value of the difference between the output result value and the benchmark value. When the deviation of any key indicator (such as the health score) exceeds a set threshold (e.g., 0.15), the system automatically activates the parameter update mechanism. Refer to Table 1, which shows the monitoring and triggering process at five consecutive time points (assuming the current test sub-region is the main test sub-region and the health score benchmark value is 0.75).

[0086] Table 1: Test Output Result Deviation Monitoring Record Table

[0087]

[0088] After triggering the parameter update mechanism, the system first collects historical test data. Taking time T3 as an example, it collects all test output results (including data from times T1, T2, and T3) for the current sub-region within the last 10 seconds, calculates the residual between its health score and the corresponding baseline value, and forms a residual sequence sorted by time: [0.03, 0.07, 0.17]. This sequence represents the evolution trend of the bias. The residual sequence is input into a preset adaptive network. This network adopts a time series prediction architecture and includes a long short-term memory (LSTM) layer. The network internally extracts features from the residual sequence: identifying the mean level, slope of change, fluctuation amplitude, and recent abrupt changes. Based on these features, the network outputs a set of adjustment parameter vectors. The vector contains two core elements: a test intensity correction coefficient and a test frequency adjustment amount.

[0089] For example, when the residual sequence shows an upward trend and the latest residual is large (such as the T3 sequence), the adaptive network may output the following adjustment parameters:

[0090] Test intensity correction factor: 1.25 (indicating increased test stimulus intensity);

[0091] Test frequency adjustment: +5Hz (indicates an increase in data sampling rate);

[0092] The adjustment parameters are passed to the device control layer. The device control layer performs specific operations based on the parameter type:

[0093] Test intensity correction: This is achieved through a proportional controller. The current output intensity setting of the test equipment (e.g., exciter force amplitude of 50N) is read, multiplied by a correction factor (1.25), and the updated setting is 62.5N. A control signal is then sent in real-time to the exciter drive module, adjusting its output force amplitude according to the new setting.

[0094] Test frequency adjustment: This is achieved by reconfiguring the data acquisition module. The current sampling rate (e.g., 100Hz) is read, and an adjustment amount (+5Hz) is added to update the sampling rate to 105Hz. The acquisition module synchronously adjusts the clock divider parameters to ensure the sampling interval is precisely shortened.

[0095] During the secondary trigger at time T5, the system collects a new residual sequence (e.g., residuals at times T3, T4, and T5: [0.17, 0.14, 0.20]). The adaptive network identifies persistent high-level fluctuations in the residuals and may output a larger correction coefficient (e.g., 1.4) and frequency adjustment (e.g., +10Hz) to further intensify the test conditions. The device control layer applies the new parameters to subsequent test procedures in real time. The entire process forms a closed loop: when significant deviations are detected, the adaptive network is used to generate device parameter adjustments based on recent historical residuals, updating the test intensity or frequency through the hardware interface, thereby optimizing the quality of subsequent test data acquisition. All parameter update records are stored in association with the original data for subsequent auditing and model optimization.

[0096] Example 5: After the testing equipment completes performance testing on all test sub-regions within the test area set, the system initiates the result integration process. Each test sub-region generates an independent test output result record during the testing process. This record contains multi-dimensional data fields: a unique spatial identifier for the test sub-region, a test timestamp, a health score output by the performance analysis model, a fault probability prediction value, a remaining life estimate value, and original feature vectors (such as extracted tension change features and vibration change features). The integration operation first performs spatial alignment: based on the unique spatial identifier of the test sub-region, all records are arranged in order of their actual physical positions on the CVT belt, forming a spatially continuous sequence. The arrangement is based on the position coordinates in the belt running direction, ensuring that the data order is consistent with the actual belt spatial distribution. After alignment, data continuity is checked. For cases where individual sub-region records are missing due to equipment failure or signal interference, a data filling operation is performed. Filling adopts a neighboring region interpolation strategy: the position of the missing record is located, the record data of the two adjacent test sub-regions are extracted, the average value of the corresponding fields is calculated, and a filling record is generated and inserted at the missing position.

[0097] After spatial alignment and data imputation are completed, the multi-source data fusion stage begins. This stage integrates the independent records of each test sub-region into a unified data view. The fusion operation includes three levels: First, all health score values ​​are aggregated to form a health score distribution sequence covering the entire belt; second, fault probability prediction values ​​are aggregated to construct a fault probability spatial distribution sequence; third, all original feature vectors are integrated to form a tension change feature matrix and a vibration change feature matrix across the entire belt dimension. The row index of the matrix corresponds to the test sub-region location, and the column index corresponds to the feature type. The fusion process uses weighted superposition: data from the main test sub-region is assigned a higher weight (e.g., 1.2 times); data from the secondary test sub-region uses a standard weight (1.0 times); and data from the auxiliary test sub-region is assigned a lower weight (e.g., 0.8 times). The weight difference reflects the relative importance of different region types. The weighted data is processed by a spatial smoothing filter to eliminate abrupt noise caused by boundary division, generating a continuous and smooth performance dataset.

[0098] Based on the integrated performance dataset, the report generation module is invoked to create an overall performance report for the CVT belt. The report content is defined by a structured template and includes the following core sections:

[0099] Overall performance level determination: Calculate the weighted average of the health scores of the entire belt and map it to the performance level (e.g., excellent / good / medium / poor) according to the preset threshold range.

[0100] Spatial distribution of key indicators: The health score sequence and the failure probability sequence are mapped to a spatial distribution curve according to the location coordinates. The horizontal axis represents the belt length coordinate, and the vertical axis has two scales corresponding to the score and probability value respectively.

[0101] Performance Heatmap: Extract the tension change feature matrix and vibration change feature matrix, and calculate the Euclidean norm of the feature vector for each spatial location as a heatmap value. A visualization engine is then used to generate a two-dimensional heatmap, with belt length as the horizontal axis and feature intensity as the vertical axis. A color map reflects the intensity level (red for high, blue for low).

[0102] Defect location marking: Compared with the potential defect locations identified in Example 1, the locations where the actual detected health scores are below the threshold or the failure probability exceeds the limit are marked on the spatial distribution map, and their coordinate values ​​are associated.

[0103] Historical comparative analysis: retrieve historical test reports for the same type of belt, extract performance data for the corresponding region, and generate a line graph showing the deviation between the current score and the historical average score.

[0104] The report generation utilizes an automated document building engine. The engine first loads a pre-defined XML template file, parsing its content structure and style definitions. Then, it injects elements from the integrated dataset according to template placeholders: numerical data directly populates tables, distribution curves are rendered as PNG images using the Matplotlib library and embedded in the document, and heatmaps are generated and inserted using OpenCV color mapping functions. The final output is a PDF document containing text descriptions, data tables, and image elements.

[0105] The generated PDF overall performance report is automatically transferred to the storage management system. The storage system employs a layered architecture: structured data (test output result records, integrated datasets) is written to a dedicated data table in a relational database via a JDBC interface. The table structure includes fields such as timestamp, device number, region type, and health score. Unstructured report documents are transferred to the object storage service via FTP and stored as PDF files according to the "year / month / device ID" path rule. A two-way index is established between database records and storage files: a report file path field is added to the database table, and the corresponding database record ID is inserted into the file metadata of the object storage. All storage operations are logged in an audit log, including storage time, operator identifier (automatically marked as "AUTO"), and data verification code. After storage is complete, the system updates the test task status to "completed" and releases related computing resources. The stored datasets and reports can be retrieved and accessed in subsequent maintenance cycles using a unique task ID for trend analysis or comparative diagnostics.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A performance testing method for CVT belts, characterized in that, include: Acquire tension and vibration data of the CVT belt during operation; Based on the tension data and the vibration data, the CVT belt is divided into multiple test sub-regions; Based on the multiple test sub-regions, the performance index of each test sub-region is calculated, and the performance index includes the tension change index and the vibration amplitude index. Construct an association graph among the multiple test sub-regions, where the multiple test sub-regions are vertices and the performance difference between adjacent test sub-regions is the edge; The test order is determined according to the association diagram, and the test equipment is driven to perform performance tests on the multiple test sub-regions according to the test order; Dividing the CVT belt into multiple test sub-regions includes: Based on the location of potential defects, the high-concern and low-concern areas of the CVT belt are determined; Based on the high-interest area and the low-interest area, the CVT belt is divided into a main test sub-area and a secondary test sub-area. In the areas not covered by the main test sub-region and the secondary test sub-region, an auxiliary test sub-region is determined. The main test sub-region, the secondary test sub-region, and the auxiliary test sub-region constitute a test region set. The performance metrics for each test sub-region include: For each test sub-region in the set of test regions, extract the tension change characteristics and vibration change characteristics of the corresponding region; Calculate the tension change index based on the described tension change characteristics; Calculate the vibration amplitude index based on the vibration change characteristics; By combining the tension change index and the vibration amplitude index, a comprehensive performance score is generated for each test sub-region; Constructing the association graph between the multiple test sub-regions includes: For adjacent first and second test sub-regions in the test region set, identify the region type of the first and second test sub-regions. The region type includes one of the main test sub-region, secondary test sub-region, and auxiliary test sub-region. Based on the identified region type, determine the first performance reference value of the first test sub-region and the second performance reference value of the second test sub-region; Calculate the difference between the first performance reference value and the second performance reference value, and use the difference as the performance difference value between the first test sub-region and the second test sub-region.

2. The performance testing method for CVT belts according to claim 1, characterized in that, Acquiring tension and vibration data of the CVT belt during operation includes: The tension and vibration sequences of the CVT belt are collected by sensors within a preset test duration. The tension sequence and the vibration sequence are preprocessed to generate standardized tension vectors and vibration vectors; Based on the standardized tension vector and the standardized vibration vector, the potential defect locations of the CVT belt are identified.

3. The performance testing method for CVT belts according to claim 1, characterized in that, Determining the test order based on the correlation diagram includes: Analyze the topology of the association graph to generate a minimum path sequence; Based on the minimum path sequence, the test priorities of each test sub-region in the test region set are arranged. The test order is output based on the test priority.

4. The performance testing method for CVT belts according to claim 3, characterized in that, Driving the testing equipment to perform performance tests on the multiple test sub-regions according to the aforementioned test sequence includes: According to the test order, the real-time test parameters of each test sub-region are obtained sequentially; The real-time test parameters are input into the performance analysis model, and the test output results are generated through the performance analysis model. Based on the test output results, adjust the operating parameters of the test equipment.

5. The performance testing method for CVT belts according to claim 4, characterized in that, Adjusting the operating parameters of the testing equipment includes: Monitor the deviation between the test output results and the preset benchmark value; When the deviation exceeds the threshold, the parameter update mechanism is triggered; The test intensity or test frequency of the test equipment is corrected through the parameter update mechanism.

6. The performance testing method for CVT belts according to claim 5, characterized in that, The trigger parameter update mechanism includes: Collect historical test output results and construct residual data sequences; The residual data sequence is input into an adaptive network, and adjustment parameters are generated through the adaptive network. The operating parameters of the test equipment are updated using the adjusted parameters.

7. The performance testing method for CVT belts according to claim 6, characterized in that, The method further includes: After completing the performance tests of all test sub-regions, integrate the test output results; Based on the integrated test output results, an overall performance report of the CVT belt is generated; The overall performance report is stored for subsequent analysis.

Images