A slide rail quality inspection method and system based on artificial intelligence

Abstract

The application discloses a kind of based on artificial intelligence's slide rail quality inspection method and system, it is related to slide rail quality inspection technical field.The based on artificial intelligence's slide rail quality inspection method, by obtaining the sliding acoustic wave signal data of several windows of setting slide rail, and analyze its corresponding window slide rail voiceprint defect aggregate characteristic value, and identify several abnormal time windows of setting slide rail, obtain the slide rail surface topography image data of each abnormal window, and combine pre-trained slide rail surface topography detection model, analyze its corresponding abnormal window slide rail topography damage characteristic value, read the slide rail voiceprint defect aggregate characteristic value of each abnormal window, and combine its corresponding abnormal window slide rail topography damage characteristic value, analyze the slide rail sound shape collaborative defect characteristic value of setting slide rail, the quality inspection processing of setting slide rail is carried out by slide rail sound shape collaborative defect characteristic value, to effectively reduce the misjudgment phenomenon caused by slight defect, and improve the accuracy of inspection.

Description

Technical Field

[0001] This invention relates to the field of slide rail quality inspection technology, specifically to a slide rail quality inspection method and system based on artificial intelligence. Background Technology

[0002] As a key component in precision mechanical equipment, electronic devices, furniture, and rail transportation, the smoothness and durability of slide rails directly affect the overall service life and safety. During the processing and assembly of slide rails, defects such as scratches and residual assembly contamination are prone to occur. These defects can cause fluctuations in frictional resistance, abnormal noises, and jamming during operation, seriously affecting the quality and reliability of the product. Existing slide rail quality inspection methods mainly rely on manual visual inspection. Manual inspection is subject to strong subjectivity, low efficiency, and difficulty in detecting hidden and subtle defects, which can easily lead to missed inspections and misjudgments. With the development of artificial intelligence technology, deep learning-based inspection methods are gradually being applied to inspection scenarios.

[0003] The limitations of existing technologies include at least the following problems: Existing technologies lack detailed analysis of the degree of defects in slide rails. Damage to slide rails during operational testing exhibits significant diversity and progressiveness, such as minor scratches. These defects initially manifest as slight abnormalities in friction and do not necessarily affect overall function; they can be restored to use through repair or maintenance. However, existing technologies often directly classify such minor problems as non-conforming, leading to the excessive discarding of repairable components, resulting in resource waste and increased maintenance costs. Simultaneously, when defects gradually evolve into severe damage, such as deep scratches or severe contamination, it is difficult to accurately characterize their severity, easily leading to the failure to fully identify potential hazards. This results in a disconnect between test conclusions and actual usage requirements, thereby weakening the reliability of the test results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for inspecting the quality of slide rails. This solves the problem that existing technologies struggle to distinguish the degree of slide rail defects, easily leading to erroneous rejection due to minor defects, and thus lacking sufficient detection accuracy.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based method for inspecting the quality of a slide rail, comprising the following steps: acquiring sliding acoustic signal data of several windows of a set slide rail, and analyzing the aggregated feature values ​​of slide rail acoustic defects in the corresponding windows; identifying several abnormal time windows of the set slide rail based on the aggregated feature values ​​of slide rail acoustic defects in each window of the set slide rail; acquiring slide rail surface morphology image data of each abnormal window of the set slide rail, and analyzing the slide rail morphology damage feature values ​​of the corresponding abnormal window in conjunction with a pre-trained slide rail surface morphology detection model; reading the aggregated feature values ​​of slide rail acoustic defects in each abnormal window of the set slide rail, and analyzing the slide rail acoustic-morphological co-defect feature values ​​of the set slide rail in conjunction with the slide rail morphology damage feature values ​​of the corresponding abnormal window, and performing quality inspection processing.

[0006] Furthermore, the specific steps for analyzing the aggregated feature values ​​of the acoustic defects of each window of the set slide rail are as follows: input the sliding acoustic wave signal data of each window of the set slide rail into the pre-trained acoustic defect detection model, analyze the acoustic defect mapping feature set of the corresponding window, including the feature value of cumulative friction bias intensity, the feature value of spectral imbalance fluctuation, and the feature value of stick-slip fluctuation amplitude; based on the acoustic defect mapping feature set of each window of the set slide rail, analyze the aggregated feature value of the acoustic defects of the corresponding window of the set slide rail.

[0007] Furthermore, the sliding acoustic wave signal data specifically refers to the sound amplitude at each time point, and the acoustic defect detection model includes an input layer, a voiceprint analysis layer, a voiceprint evolution layer, and a mapping output layer.

[0008] Furthermore, the specific steps for analyzing the acoustic defect mapping feature set of each window of the set slide rail are as follows: In the input layer of the acoustic defect detection model, the sliding acoustic wave signal data of each window of the set slide rail is received and preprocessed; in the acoustic pattern analysis layer of the acoustic defect detection model, based on the preprocessed sliding acoustic wave signal data of each window of the set slide rail, the acoustic wave defect mapping feature vector set of its corresponding window is extracted; in the acoustic pattern evolution layer of the acoustic defect detection model, based on the acoustic wave defect mapping feature vector set of each window of the set slide rail, the temporal feature vector of its corresponding window is analyzed; in the mapping output layer of the acoustic defect detection model, based on the temporal feature vector of the set slide rail, the acoustic pattern defect mapping feature set of its corresponding window is output.

[0009] Furthermore, the slide rail surface morphology image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the slide rail surface morphology image, and the slide rail surface morphology detection model includes a morphology input layer, a defect feature parsing layer, and a fusion output layer.

[0010] Furthermore, the specific steps for analyzing the slide rail morphology damage feature values ​​of each window of the slide rail are as follows: Input the slide rail surface morphology image data of each abnormal window of the slide rail into the pre-trained slide rail surface morphology detection model, and analyze the morphology defect feature set of the corresponding abnormal window, including scratch energy attenuation feature value along the path, assembly contamination abnormal feature value, and slide rail edge meshing disturbance feature value; Based on the morphology defect feature set of each abnormal window of the slide rail, analyze the slide rail morphology damage feature value of the corresponding abnormal window.

[0011] Furthermore, the specific steps for analyzing the shape defect feature set of each window of the set slide rail are as follows: In the shape input layer of the slide rail surface shape detection model, the slide rail surface shape image data of each abnormal window of the set slide rail is received and image preprocessed; in the defect feature parsing layer of the slide rail surface shape detection model, based on the slide rail surface shape image data of each abnormal window of the set slide rail after image preprocessing, the slide rail surface defect feature vector of the corresponding abnormal window is extracted; in the fusion output layer of the slide rail surface shape detection model, based on the slide rail surface defect feature vector of each abnormal window of the set slide rail, the shape defect feature set of the corresponding abnormal window is output.

[0012] Furthermore, the specific formula for calculating the characteristic value of the acoustic-visual cooperative defect of the set slide rail is as follows: ;in, To set the characteristic value of the acoustic-visual coordination defect of the slide rail, , The sequence is the first step of setting the slide rail. The aggregated characteristic values ​​of the sound signature defects of the slide rail and the characteristic values ​​of the slide rail morphology damage in each abnormal window. , , , , The following are, in order, the adjustment coefficients for voiceprint defects, morphological damage, smoothness, difference, and compensation stored in the database: , This represents the total number of abnormal windows.

[0013] Furthermore, the specific steps of the quality inspection process are as follows: the characteristic value of the slide rail's acoustic and visual coordination defect is judged against the preset threshold value of the slide rail's acoustic and visual coordination defect; based on the judgment result, corresponding quality handling measures are taken.

[0014] An artificial intelligence-based slide rail quality inspection system includes: an acoustic wave acquisition and analysis module for acquiring sliding acoustic wave signal data of several windows of a set slide rail and analyzing the aggregated feature values ​​of slide rail acoustic defects in the corresponding windows; an abnormal window identification module for identifying several abnormal time windows of the set slide rail based on the aggregated feature values ​​of slide rail acoustic defects in each window; a morphological feature analysis module for acquiring slide rail surface morphological image data of each abnormal window of the set slide rail and, in conjunction with a pre-trained slide rail surface morphological detection model, analyzing the slide rail morphological damage feature values ​​of the corresponding abnormal windows; and an acoustic-morphological collaborative inspection module for reading the aggregated feature values ​​of slide rail acoustic defects in each abnormal window of the set slide rail and, in conjunction with the slide rail morphological damage feature values ​​of the corresponding abnormal windows, analyzing the acoustic-morphological collaborative defect feature values ​​of the set slide rail and performing quality inspection processing.

[0015] The present invention has the following beneficial effects:

[0016] (1) The slide rail quality inspection method based on artificial intelligence analyzes the acoustic signals of multiple windows of the slide rail to generate the corresponding acoustic defect aggregated feature value of the slide rail in the window, and uses it to identify abnormal windows. At the same time, it triggers the slide rail surface morphology image in the abnormal window, thereby comprehensively generating the slide rail acoustic-morphological collaborative defect feature value. It can not only reflect whether the defect exists, but also characterize the severity of the defect, and then perform fine classification of the slide rail. It also effectively reduces the false detection phenomenon caused by minor defects, thereby improving the accuracy of inspection and meeting the needs of high-precision quality control of slide rail in practical applications.

[0017] (2) The artificial intelligence-based slide rail quality inspection method constructs an acoustic defect detection model and performs in-depth analysis of the sliding acoustic wave signal data of the slide rail, thereby effectively extracting the acoustic defect mapping feature set to accurately characterize the friction state of different windows. It can not only highlight the key acoustic defect features related to defects in the acoustic signal, but also generate the slide rail acoustic defect aggregate feature value on this basis, thereby ensuring that the output results can fully reflect the acoustic detection results of the slide rail in the corresponding window, thus ensuring the reliability of the acoustic detection results.

[0018] (3) The artificial intelligence-based slide rail quality inspection method introduces a slide rail surface morphology detection model and performs targeted analysis on the slide rail surface morphology image data during the abnormal window, thereby generating a morphology defect feature set corresponding to the acoustic anomaly. It can not only obtain the full-view image of the slide rail surface at the moment the acoustic detection detects the anomaly, but also refine the local state of the abnormal window, thereby realizing the accurate description of the slide rail surface damage. It can reflect the extension of the scratch, the aggregation of the contamination residue, and the non-uniformity of the edge meshing, thus making the detection results more targeted and interpretable, effectively supporting the final acoustic-morphological collaborative defect judgment, and ensuring the credibility of the inspection results.

[0019] (4) The artificial intelligence-based slide rail quality inspection system adopts a modular architecture for collaborative analysis to achieve the continuity of detection. The acoustic wave acquisition and analysis module is responsible for real-time processing of the sliding acoustic wave signal. The abnormal window identification module uses this result to filter out abnormal windows. The morphology feature analysis module builds a slide rail surface morphology detection model and completes damage feature extraction. Finally, the acoustic and morphology collaborative inspection module comprehensively analyzes the acoustic and morphological information of the slide rail and outputs a unified slide rail acoustic and morphology collaborative defect feature value, which is used for quality inspection. This effectively avoids false detection and improves the confidence of the detection conclusion.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This is a flowchart of a slide rail quality inspection method based on artificial intelligence according to the present invention.

[0022] Figure 2 This is a schematic diagram of the window sequence data of the acoustic defect mapping feature set of the slide rail in the slide rail quality inspection method based on artificial intelligence of the present invention.

[0023] Figure 3 This is a flowchart illustrating the specific steps involved in analyzing the acoustic defect mapping feature set of each window of a slide rail in an artificial intelligence-based slide rail quality inspection method of the present invention.

[0024] Figure 4 This is a block diagram of an artificial intelligence-based slide rail quality inspection system according to the present invention. Detailed Implementation

[0025] Please see Figure 1This invention provides a technical solution: an artificial intelligence-based method for inspecting the quality of a slide rail, comprising the following steps: during the slide rail operation test (the operation test process is a reciprocating motion, such as the slider sliding from the starting end to the ending end on the slide rail, and then returning from the ending end to the starting end, forming a complete reciprocating cycle), within this cycle, continuously acquire sliding acoustic wave signal data of several windows (such as 200ms as a window) of the set slide rail, and analyze the aggregated characteristic value of the slide rail acoustic defects in the corresponding window;

[0026] Based on the aggregated feature value of the acoustic fingerprint of each window of the set slide rail, several abnormal time windows of the set slide rail are identified. Specifically, it is determined whether the aggregated feature value of the acoustic fingerprint of each window of the set slide rail is higher than the set threshold value. If it is higher than the set threshold value, the window corresponding to the acoustic fingerprint of the set slide rail is marked as an abnormal window. (It should be noted that if there are no abnormal windows, the slide rail is considered to have passed the quality inspection, because the acoustic detection covers the round-trip cycle of the slide rail operation and can capture dynamic abnormal signals such as sudden changes in friction energy, imbalance in spectrum distribution, and alternating stick and slip in real time. If the aggregated feature value of the acoustic fingerprint of all windows is lower than the set threshold value, it indicates that the friction state of the slide rail is stable and the force is balanced throughout the entire operation process, and there are no functional defects such as whistling or jamming. At this time, no further image detection is required to determine that it is qualified, because the role of image detection is to verify the morphology of the abnormal position indicated by acoustic detection. When there is no acoustic abnormality, even if surface defects exist, they are minor and have no substantial impact on the function of the slide rail, and do not affect the quality judgment result.)

[0027] (And when marked as an abnormal window, trigger the acquisition of the slide rail surface morphology image at the corresponding position of the abnormal window, and it is a full view image of the slide rail) Obtain the slide rail surface morphology image data of each abnormal window of the set slide rail, and combine it with the pre-trained slide rail surface morphology detection model to analyze the slide rail morphology damage feature value of the corresponding abnormal window; Read the slide rail acoustic texture defect aggregation feature value of each abnormal window of the set slide rail, and combine it with the slide rail morphology damage feature value of the corresponding abnormal window to analyze the slide rail acoustic shape collaborative defect feature value of the set slide rail, and perform quality inspection processing.

[0028] The specific formula for calculating the characteristic value of the acoustic-visual coordination defect of the set slide rail is as follows: ;in, To set the characteristic value of the acoustic-visual coordination defect of the slide rail, To set the first of the slide rails Aggregated feature values ​​of slide rail acoustic defects in anomaly windows The adjustment coefficients for voiceprint defects are stored in the database. To set the first of the slide rails Damage characteristic values ​​of the slide rail morphology in each abnormal window. The morphological damage adjustment coefficients are stored in the database. The smoothing adjustment coefficients are stored in the database. The difference adjustment coefficients are stored in the database. These are the compensation adjustment coefficients stored in the database. , This represents the total number of abnormal windows, and in this implementation example, the database stores the voiceprint defect adjustment coefficient. Morphological damage adjustment coefficient Smoothing adjustment coefficient Difference adjustment coefficient Compensation adjustment coefficient The values ​​were 0.387, 0.543, 2.000, 0.764, and 0.001, respectively.

[0029] The specific steps of the quality inspection process are as follows: The characteristic value of the slide rail's acoustic-visual coordination defect is compared with the preset threshold value. Based on the judgment result, corresponding quality control measures are taken. Specifically: if the characteristic value of the slide rail's acoustic-visual coordination defect is lower than the preset threshold value, the slide rail is marked as repairable, a prompt message is issued to the relevant operator, and the slide rail undergoes visual re-inspection and manual repair; if the characteristic value of the slide rail's acoustic-visual coordination defect is higher than or equal to the preset threshold value, the slide rail is marked as unqualified.

[0030] Specifically, the steps for analyzing the aggregated feature values ​​of the acoustic signature defects of each window of the sliding rail are as follows: Input the sliding acoustic wave signal data of each window of the sliding rail into a pre-trained acoustic defect detection model, and analyze the acoustic signature defect mapping feature set of the corresponding window, including the cumulative intensity feature value of friction bias, the spectral imbalance fluctuation feature value, and the stick-slip fluctuation amplitude feature value; Based on the acoustic signature defect mapping feature set of each window of the sliding rail, analyze the aggregated feature values ​​of the acoustic signature defects of the corresponding window, specifically: based on the cumulative intensity feature value of friction bias, the spectral imbalance fluctuation feature value, the spike density abrupt change feature value, and the stick-slip fluctuation... The amplitude feature values ​​are extracted for each window, including the maximum value (i.e., the maximum value among the cumulative intensity feature values ​​of friction offset, spectral imbalance fluctuation, and stick-slip fluctuation amplitude), the minimum value (i.e., the minimum value among the cumulative intensity feature values ​​of friction offset, spectral imbalance fluctuation, and stick-slip fluctuation amplitude), and the mean value (i.e., the mean value among the cumulative intensity feature values ​​of friction offset, spectral imbalance fluctuation, and stick-slip fluctuation amplitude). These values ​​are then combined with the cumulative intensity feature values ​​of friction offset, spectral imbalance fluctuation, and stick-slip fluctuation amplitude of the corresponding window to analyze and define the aggregated feature value of the slide rail's acoustic defects for each window.

[0031] The specific formula for calculating the aggregated characteristic value of the sound signature defect of a certain window of the slide rail is as follows: ;in, To set the aggregated characteristic value of the soundprint defects of a certain window of the slide rail, To set the cumulative intensity characteristic value of friction offset for a certain window of the slide rail, The friction bias adjustment coefficient is stored in the database. To set the spectral imbalance fluctuation characteristic value of a certain window of the slide rail, The spectral imbalance adjustment coefficients are stored in the database. To set the stick-slip fluctuation amplitude characteristic value for a certain window of the slide rail, The viscosity-slip fluctuation adjustment coefficients are stored in the database. These are the coordination coefficients stored in the database. To set the maximum value of voiceprint defects for a specific window of the slider, To set the minimum soundprint defect value for a specific window of the slide rail, To set the average soundprint defect value for a specific window of the slide rail, The fluctuation adjustment coefficient is stored in the database, and in this embodiment, the friction bias adjustment coefficient is stored in the database. Spectrum imbalance adjustment coefficient Viscosity fluctuation adjustment coefficient Coordination adjustment coefficient Fluctuation adjustment coefficient The values ​​were 0.438, 0.462, 0.767, 0.333, and 0.512, respectively.

[0032] The following is a specific implementation example for calculating the aggregated characteristic value of the acoustic texture defect of a set slide rail in a certain window. The available data includes the cumulative intensity characteristic value of friction bias, the characteristic value of spectral imbalance fluctuation, and the characteristic value of stick-slip fluctuation amplitude for three windows (randomly selected) of the set slide rail, as detailed in Table 1 and... Figure 2 As shown:

[0033] Table 1. Example of window sequence data for setting the acoustic defect mapping feature set of the slide rail.

[0034] Friction bias cumulative strength characteristic value Spectral Imbalance Fluctuation Characteristics Stick-slip fluctuation amplitude characteristic value Window 1 0.346 0.412 0.286 Window 2 0.413 0.371 0.362 Window 3 0.327 0.341 0.314

[0035] Based on the data in Table 1, extract the maximum, minimum, and average voiceprint defects for the corresponding window, i.e.:

[0036] The maximum, minimum, and average values ​​of voiceprint defects in the first window are 0.412, 0.286, and 0.348, respectively.

[0037] The maximum, minimum, and average values ​​of voiceprint defects in the second window are 0.413, 0.362, and 0.382, respectively.

[0038] The maximum, minimum, and average values ​​of voiceprint defects in the third window are 0.341, 0.314, and 0.327, respectively.

[0039] Friction bias adjustment coefficient stored in the database The value is: 0.438;

[0040] Spectral imbalance adjustment coefficients stored in the database The value is: 0.462;

[0041] Viscosity fluctuation adjustment coefficients stored in the database The value is: 0.767;

[0042] Coordination coefficients stored in the database The value is: 0.333;

[0043] Fluctuation adjustment coefficients stored in the database The value is: 0.512;

[0044] Substituting the data from Table 1 and the data above into the specific formula for calculating the aggregated characteristic value of the slide rail acoustic texture defect in a certain window of the set slide rail, we get:

[0045] The feature value of the slide rail acoustic defect aggregation for the first window of the slide rail is set as follows: (ln(1+((0.346^0.438)×(0.412^0.462)×(0.286^0.767))^0.333))×(1+0.512×((0.412-0.286) / 0.348))≈0.514;

[0046] The feature value of the track acoustic defect aggregation of the first window of the track is set as follows: (ln(1+((0.413^0.438)×(0.371^0.462)×(0.362^0.767))^0.333))×(1+0.512×((0.413-0.362) / 0.382))≈0.491;

[0047] Set the feature value of the slide rail soundprint defect aggregation for the first window of the slide rail = (ln(1+((0.327^0.438)×(0.341^0.462)×(0.314^0.767))^0.333))×(1+0.512×((0.341-0.314) / 0.327))≈0.447).

[0048] The sliding acoustic wave signal data is specifically the sound amplitude at each time point (the sound amplitude at each time point corresponds to the instantaneous sound pressure relative to the environmental static pressure reference, where a positive sound amplitude indicates that the sound pressure is higher than the reference, and a negative sound amplitude indicates that the sound pressure is lower than the reference). The acoustic defect detection model includes an input layer, a voiceprint analysis layer, a voiceprint evolution layer, and a mapping output layer.

[0049] like Figure 3 As shown, the specific steps for analyzing the acoustic defect mapping feature set of each window of the set slide rail are as follows: In the input layer of the acoustic defect detection model, the sliding acoustic wave signal data of each window of the set slide rail is received and preprocessed. Specifically, the sliding acoustic wave signal data is denoised, such as using a bandpass filtering algorithm to remove environmental background noise and irrelevant low-frequency interference, and at the same time using an adaptive noise suppression method to weaken random burst noise, thereby retaining the main acoustic components related to slide rail friction. The denoised sliding acoustic wave signal is divided according to a preset time window, such as using a time window of 20ms to 30ms for segmentation, and allowing 25% to 50% overlap between adjacent windows to ensure the continuity of the time sequence.

[0050] In the acoustic signature analysis layer of the acoustic defect detection model, based on the preprocessed sliding acoustic signal data of each window of the set sliding rail, the acoustic defect mapping feature vector set of the corresponding window is extracted. In the acoustic signature evolution layer of the acoustic defect detection model, based on the acoustic defect mapping feature vector set of each window of the set sliding rail, the temporal feature vector of the corresponding window is analyzed. Specifically, this layer is an LSTM. The LSTM selectively memorizes and forgets the input data through the mechanism of gated recurrent units, that is, it controls the flow of information between time steps through input gates, forget gates, and output gates. At each time step, the network updates according to the current input and the hidden state of the previous time step, thereby retaining the memory information of historical time segments in the hidden state. Through this process, the LSTM can learn the dynamic pattern of the sliding rail acoustic signal evolution over time, capture temporal trends such as friction energy mutation and spectral distribution shift, and thus generate a temporal feature vector that reflects the evolution law of sliding rail friction anomalies over time, such as:

[0051] For the triboelectric acoustic energy bias characteristics of each time window in the acoustic defect mapping feature vector set, the triboelectric acoustic energy bias characteristics of any adjacent time window are processed by difference (average value) and root mean square processing to extract the cumulative intensity characteristics of friction bias, which are used to reflect the cumulative effect of the uneven distribution of positive and negative acoustic energy during the operation of the slide rail over time. When there are assembly deviations or uneven forces on the slide rail, this characteristic value continues to increase, indicating that the degree of friction bias is aggravated.

[0052] For the friction spectrum energy imbalance features of each time window in the acoustic defect mapping feature vector set, standard deviation processing is performed to extract the spectrum imbalance fluctuation features, which are used to reflect the instability of the energy distribution of the slide rail friction sound in different frequency bands; when insufficient lubrication or howling occurs, the frequency band energy distribution fluctuation is significantly enhanced, and this feature is significantly increased.

[0053] For the micro-viscosity-slip ratio features of each time window in the acoustic defect mapping feature vector set, the maximum, minimum, and mean micro-viscosity-slip ratio features are extracted, and then a ratio is applied: |maximum micro-viscosity-slip ratio feature - minimum micro-viscosity-slip ratio feature| / mean micro-viscosity-slip ratio feature to extract the viscosity-slip fluctuation amplitude features, which reflect the intensity of the alternation between viscosity and slippage during the slide rail friction process. When the slide rail lubrication deteriorates or wear is uneven, the viscosity-slip switching is frequent and the amplitude difference is large, and this feature increases significantly. The cumulative intensity features of friction bias, the spectral imbalance fluctuation features, and the viscosity-slip fluctuation amplitude features are concatenated into a time-series feature vector.

[0054] In the mapping output layer of the acoustic defect detection model, based on the temporal feature vector of the set slide rail, the acoustic defect mapping feature set of its corresponding window is output. Specifically, the friction bias cumulative intensity feature, spectral imbalance fluctuation feature, and stick-slip fluctuation amplitude feature in the temporal feature vector are processed by the Sigmoid function, and their results are mapped between 0 and 1 to obtain the friction bias cumulative intensity feature value, spectral imbalance fluctuation feature value, and stick-slip fluctuation amplitude feature value.

[0055] The specific steps for extracting the acoustic defect mapping feature vector of the set slide rail are as follows: For each time window of the preprocessed sliding acoustic signal data, the time point when the sound amplitude in the acoustic signal changes from positive to negative or from negative to positive is identified as the zero-crossing point. The waveform between two adjacent zero-crossing points is taken as a complete cycle (since the boundary of the time window may not coincide with the zero-crossing point, there may be incomplete cycles at the beginning or end of some time windows. For such incomplete cycles, the waveforms at the beginning or end that are less than a complete cycle are discarded, and only the complete cycle is retained). The sound amplitude is... A waveform with a positive value is considered a positive half-cycle, and vice versa for a negative half-cycle. For each complete cycle, the total energy of the positive half-cycle and the negative half-axis is calculated separately (e.g., the sum of the sound amplitudes in the positive half-cycle, and the absolute value of the total energy for the negative half-axis), and the difference is processed (the absolute value is taken). Based on the result, the root mean square is processed to extract the frictional sound energy bias characteristics of each window, which are used to reflect the degree of structural abnormality of the slide rail during operation. When there are assembly deviations, insufficient lubrication, or defects on the track surface, the sound energy of the positive and negative half-cycles will show significant differences, thereby increasing this characteristic.

[0056] For each time window of the preprocessed sliding acoustic wave signal data, a Fast Fourier Transform (FFT) is performed on the sound amplitude at each time point within that time window to obtain the corresponding spectral amplitude sequence (including several frequency components and their corresponding amplitudes). In the spectral amplitude sequence, the frequency range is divided into low-frequency band (e.g., 0–1 kHz), mid-frequency band (e.g., 1–5 kHz), and high-frequency band (e.g., above 5 kHz). The energy values ​​in each frequency band are statistically analyzed (i.e., the sum of the squares of the amplitudes of all frequency points within that frequency band). The energy values ​​of the above three frequency bands are normalized to obtain the energy proportion of each frequency band. The energy proportions of any two frequency bands are then compared (and the absolute value is taken), such as |energy proportion of low-frequency band - energy proportion of mid-frequency band|. The results of the comparison are then weighted to extract the friction spectrum energy imbalance feature of each time window. This feature is used to reflect the degree of abnormal distribution of friction energy in different frequency bands during the operation of the slide rail. When the slide rail has abnormal conditions such as insufficient lubrication, jamming, or abnormal noise, the spectral energy distribution will show a significant imbalance, and this feature will increase significantly accordingly.

[0057] For each window of the preprocessed sliding acoustic signal data, it is divided into several segments (each segment includes the sound amplitude at several time points, and the absolute value is taken). For the sound amplitude at two adjacent time points in any segment, the rate of change of the sound amplitude is analyzed, such as the absolute value of the difference between the sound amplitude at the first and second time points / the sound amplitude at the second time point, and the mean is taken to obtain the rate of change of the amplitude of the segment. It is then compared with a preset threshold for the rate of change of amplitude. Segments with a rate of change of amplitude higher than the preset threshold are marked as sliding friction segments, and those with a rate of change lower than the threshold are marked as viscous friction segments. The sliding energy value (i.e., the sum of the sound amplitudes in all sliding friction segments) and the viscous energy value (i.e., the sum of the sound amplitudes in all viscous friction segments) are statistically analyzed and their ratios are calculated as |sliding energy value - viscous energy value| / (sliding energy value + viscous energy value) to extract the micro-viscous-slip ratio feature. When the slide rail has insufficient lubrication, jamming, or surface defects, the sliding energy or viscous energy is significantly larger, resulting in an increase in the difference, which in turn significantly increases the feature value.

[0058] The triboacoustic energy bias characteristics, triboacoustic spectrum imbalance characteristics, triboacoustic burst peak density characteristics, and micro-viscosity-slip ratio characteristics of each window are then concatenated into an acoustic defect mapping feature vector.

[0059] Furthermore, the pre-training steps for the acoustic defect detection model are as follows:

[0060] An acoustic annotation dataset is obtained, which consists of sliding acoustic signals collected during the reciprocating operation test of the slide rail. The data is annotated by the inspection engineer in combination with the actual inspection results. Each sample in the acoustic annotation dataset includes an acoustic signal sequence within a continuous time window and a corresponding defect label. The defect labels specifically include assembly deviation, insufficient lubrication, surface scratches, edge jamming, etc. The original acoustic signals are preprocessed by using bandpass filtering to remove low-frequency environmental noise and adaptive noise suppression to weaken random interference. The processed signals are then divided into fixed time windows (e.g., 20ms to 30ms, with 25% to 50% overlap allowed between windows) to ensure temporal continuity.

[0061] During training, the acoustic defect mapping feature vectors for each time window are first extracted in the acoustic signature analysis layer, including frictional acoustic energy bias features, frictional spectral energy imbalance features, and viscosity-slip ratio features. Then, the time series features are input into the LSTM network in the acoustic signature evolution layer, and the time dependence is captured by its input gate, forget gate, and output gate gating mechanism to learn the dynamic evolution pattern of the rail friction signal over time. Finally, in the mapping output layer, the Sigmoid function is used to map each feature to the 0-1 interval and output the corresponding standardized feature values.

[0062] During model training, a weighted combination of cross-entropy loss and mean squared error loss is used as the optimization objective function. The Adam optimizer is used for iterative updates, and hyperparameters such as learning rate and batch size are gradually adjusted. The model learns the mapping relationship between features and defect types through the training set, monitors the model convergence through the validation set and prevents overfitting, and finally uses the test set to evaluate the model's generalization ability to ensure that the model can accurately identify acoustic defect features under different operating conditions. After training, the acoustic defect detection model with the optimal parameters is saved for real-time detection and judgment in the subsequent sliding rail quality inspection stage.

[0063] In this implementation scheme, the accuracy of defect detection is significantly improved by analyzing and aggregating the sliding acoustic signal data of the slide rail. Secondly, denoising and time windowing are used to purify and temporally sequence the acoustic data before it enters the model, thus ensuring the effectiveness of feature extraction. In the acoustic pattern analysis layer, key acoustic defect features are extracted from a single time window, and the acoustic pattern evolution layer uses an LSTM structure to learn its evolution over time. Finally, by analyzing the maximum, minimum, and average values ​​of acoustic pattern defects within a window and introducing multiple adjustment coefficients, the contribution of different features to the aggregation results can be effectively balanced, avoiding overall misjudgment due to the bias of a single feature. This ensures that the final aggregated feature value of the slide rail acoustic pattern defect not only reflects the instantaneous friction state during operation but also comprehensively presents the fluctuation amplitude and cumulative effect of the defect, thereby achieving reliable defect detection.

[0064] Specifically, the slide rail surface morphology image data consists of the pixel value and two-dimensional coordinates of each pixel in the slide rail surface morphology image, and the slide rail surface morphology detection model includes a morphology input layer, a defect feature analysis layer, and a fusion output layer.

[0065] The specific steps for analyzing the slide rail morphology damage feature values ​​of each window of the set slide rail are as follows: Input the slide rail surface morphology image data of each abnormal window of the set slide rail into the pre-trained slide rail surface morphology detection model, and analyze the morphology defect feature set of the corresponding abnormal window, including scratch energy attenuation feature value along the friction, assembly contamination abnormal feature value, and slide rail edge meshing disturbance feature value; Based on the morphology defect feature set of each abnormal window of the set slide rail, analyze the slide rail morphology damage feature value of the corresponding abnormal window, specifically: perform weighted processing on the scratch energy attenuation feature value along the friction, assembly contamination abnormal feature value, and slide rail edge meshing disturbance feature value of each abnormal window to obtain the slide rail morphology damage feature value of the corresponding abnormal window, which is used to characterize the overall damage degree of the slide rail surface morphology at the abnormal window position.

[0066] The specific steps for analyzing the feature set of morphological defects for each window of the slide rail are as follows: In the morphological input layer of the slide rail surface morphology detection model, the slide rail surface morphology image data of each abnormal window of the slide rail is received and image preprocessing is performed. Specifically, noise suppression processing is performed on the slide rail surface morphology image data, which can be done using methods such as median filtering, bilateral filtering, or Gaussian filtering to reduce random noise introduced by illumination, environment, or sensors during the acquisition process, thereby preserving key detail features; and region recognition processing is performed on the noise-suppressed slide rail surface morphology image data. Region recognition processing can call existing Mask... The R-CNN image segmentation network consists of the following components: a feature extraction network, which performs multi-layer convolution on the input slide rail surface topography image to obtain feature maps that can represent features at different scales. It can use ResNet and Feature Pyramid Network (FPN) as its backbone structure to simultaneously extract the overall contour of the slide rail and subtle crack textures; a region candidate network (RPN), which generates multiple candidate regions on the feature map. Each candidate region may contain the slide rail body. The RPN predicts the probability of a region on the feature map using a sliding window approach and outputs the position of the candidate box; a region feature alignment layer (RoI Align), which maps candidate regions of different sizes to a feature representation of a uniform scale, avoiding boundary shifts caused by downsampling and thus improving the accuracy of slide rail region localization; and a classification and regression branch, which determines whether a candidate region belongs to the slide rail body and performs regression correction on the bounding box position of the candidate region to obtain a more accurate slide rail bounding box. Pixels within the slide rail bounding box of the candidate region determined to be the slide rail body are marked as the slide rail pixel set.

[0067] In the defect feature analysis layer of the slide rail surface morphology detection model, based on the slide rail surface morphology image data of each abnormal window of the set slide rail after image preprocessing, the slide rail surface defect feature vector of the corresponding abnormal window is extracted.

[0068] In the fusion output layer of the slide rail surface morphology detection model, based on the slide rail surface defect feature vector of each abnormal window of the slide rail, the morphology defect feature set of the corresponding abnormal window is output. Specifically, the scratch energy attenuation feature, assembly contamination anomaly feature, and slide rail edge meshing disturbance feature in the slide rail surface defect feature vector are processed by the Sigmoid function, and the results are mapped between 0 and 1 to obtain the scratch energy attenuation feature value, assembly contamination anomaly feature value, and slide rail edge meshing disturbance feature value.

[0069] The specific steps for extracting the surface defect feature vector of each abnormal window of the set slide rail are as follows: For each pixel in the slide rail pixel set, perform connectivity processing (using 8-neighborhood or 4-neighborhood) to obtain several regions. For the two-dimensional coordinates of each pixel in any region, construct the covariance matrix and perform eigenvalue decomposition. Take the direction vector corresponding to the largest eigenvalue as the principal axis direction of that region. Project all pixels in that region onto the principal axis direction to obtain a one-dimensional projection coordinate axis. On this projection coordinate axis, sequentially calculate the average gray value of each projection position (i.e., the average gray value of all pixels in that region). The pixel values ​​of each point are processed into grayscale values ​​to obtain the grayscale pixel values ​​of the corresponding pixels. Then, the grayscale pixel values ​​of all pixels at each projection position are counted and the average value is taken. The average grayscale values ​​of any two adjacent projection positions are processed by first-order difference processing (taking the absolute value). The results are weighted to obtain the grayscale difference sequence (including several grayscale difference values) along the main axis of the region. Values ​​higher than the preset grayscale difference threshold are marked as grayscale abrupt change values. If the value is higher than the preset continuous grayscale abrupt change threshold (i.e., the total number of consecutive grayscale difference values), the region is marked as a scratch region to obtain several scratch regions.

[0070] For any scratch area, the main axis direction of the scratch area is read. Based on the main axis direction, the scratch area is divided into several profile positions at equal intervals along the main axis direction (e.g., sampled once every 1 or several pixels). At each profile position, a profile band perpendicular to the main axis direction is extracted. For any profile band, the gray values ​​of all pixels in it are counted, and these gray values ​​are squared and summed to obtain the energy value of the profile band. The same operation is performed on all profile bands in sequence to form an energy distribution sequence arranged along the main axis direction. Then, the profile energy of the scratch starting position (i.e., the starting point of the scratch area in the main axis direction, the profile band corresponding to the first profile position) is compared with the profile energy of the ending position (i.e., the ending point of the scratch area in the main axis direction, the profile band corresponding to the last profile position) to extract the scratch energy attenuation feature along the path, which is used to characterize the extensibility of the scratch on the slide rail surface. When the scratch extends for a long time, the profile energy attenuates more slowly along the path, resulting in an increase in this feature.

[0071] For each pixel in the slider pixel set, grayscale processing is performed to obtain its corresponding grayscale pixel value. Based on the local binary mode, the LBP value corresponding to each pixel is extracted, along with the LBP mean and LBP standard deviation. An LBP threshold is then set, such as LBP mean + 2 × LBP standard deviation. All pixels exceeding the set LBP threshold are identified and marked as granular bright pixels. Connectivity processing (e.g., using 8-neighborhood) is then performed to obtain several granular regions. The two-dimensional coordinates of each granular bright pixel in each granular region are read and averaged to extract the center two-dimensional coordinates of each granular region. Finally, Euclidean distance is calculated between the center two-dimensional coordinates of adjacent granular regions. The analysis results are analyzed, and the mean and standard deviation of Euclidean distance are extracted and then processed into a ratio, i.e., standard deviation of Euclidean distance / mean of Euclidean distance, to extract the pollution aggregation characteristics. At the same time, the particle area (i.e., the total number of granular bright pixel points) of each particle region is extracted, and the mean, maximum and minimum particle area are extracted and then comprehensively analyzed, i.e. (maximum particle area - minimum particle area) / mean particle area, to extract the particle area dispersion characteristics. This is then weighted with the pollution aggregation characteristics to extract the assembly pollution anomaly characteristics, which are used to characterize the spatial distribution of residual pollution on the slide rail surface. When pollution or residual particles are obviously aggregated and the size difference of the region is large, this characteristic increases.

[0072] For each pixel in the slider pixel set, an edge detection algorithm (such as the Canny operator or the Sobel operator) is used to extract the slider edge pixel set (including several slider edge pixels). Taking the vertical direction of the image (which can be set as the y-axis direction) as the reference, the image is scanned line by line. For each y-coordinate value, starting from the minimum value of the x-axis, the search is performed pixel by pixel to the right. The first slider edge pixel encountered is recorded as the leftmost boundary point of the row, and the last slider edge pixel encountered is recorded as the rightmost boundary point of the row. The leftmost boundary points of all rows are connected sequentially to obtain the left boundary curve of the slider. Similarly, the right boundary curve of the slider can be obtained. Based on the left and right boundary curves, a fixed pixel width (such as 5 to 10 pixels) is extended into the slider body. For example, for each slider edge pixel of the left boundary curve, its two-dimensional coordinates are read, and 5 to 10 pixels are continuously taken into the slider body along the positive x-axis. The area where these pixels are combined with the slider edge pixels in the left boundary curve is marked as the left edge analysis area. Similarly, the right edge analysis area can be obtained.

[0073] Within the left and right edge analysis areas, grayscale profile values ​​are calculated row by row along the y-axis: For any y-coordinate value, the grayscale values ​​of all pixels in the edge analysis area of ​​that row are extracted, and the average value is taken to obtain the corresponding profile grayscale value of that row. The above operation is performed on all rows in sequence to form a complete grayscale profile curve. The grayscale profile values ​​corresponding to two adjacent y-coordinates are differentially calculated (absolute value is taken) to obtain a grayscale difference sequence. Differential points that are greater than the preset disturbance threshold are identified and used as disturbance abrupt change points. The differential amplitude of all disturbance abrupt change points is accumulated by a sliding window to generate a disturbance amplitude envelope curve. On this envelope curve, the ratio of the local maximum value to the global average value is calculated as the disturbance amplitude envelope index of that edge. The disturbance amplitude envelope indices obtained from the left and right edge analysis areas are weighted to extract the sliding rail edge meshing disturbance features, which are used to characterize the overall disturbance level of the two sides of the sliding rail. When this feature is large, it indicates that there is local interference or jamming on both sides of the edge, indicating that there is poor meshing or abnormal overall force in the sliding rail during the assembly process.

[0074] The scratch energy attenuation characteristics along the path, assembly contamination anomaly characteristics, and slide rail edge meshing disturbance characteristics are spliced ​​together to form a slide rail surface defect feature vector.

[0075] Furthermore, the pre-training steps for the slide rail surface morphology detection model are as follows:

[0076] A topography annotation dataset is obtained, which consists of surface topography images of several slide rails collected during the operation and testing phase. The data is annotated by inspection engineers based on actual inspection records and manual review results. Each image in the annotation dataset contains pixel coordinates, gray values, and ground truth labels of corresponding defects for the slide rail surface area. The defect labels specifically include scratch areas, contamination residue areas, and edge meshing disturbance areas. The dataset is divided into training sets, validation sets, and test sets. For example, 80% is used for training, 10% for validation, and 10% for testing to ensure a balanced distribution of samples of different defect types.

[0077] During training, a deep convolutional neural network structure is adopted, with ResNet and Feature Pyramid Network (FPN) as the backbone for feature extraction, used to capture image features of different scales such as scratches, fine lines and contamination particles. The Region Proposal Network (RPN) of Mask R-CNN is combined to generate candidate regions. The features are aligned through the RoI Align layer, and then the classification and regression branches determine whether the candidate regions belong to the slide rail body and output the bounding box of the defect region. Subsequently, in the defect feature parsing layer, the corresponding feature vectors of the detected scratch regions, contamination regions and edge regions are extracted as the supervised training target of the model.

[0078] A weighted combination of cross-entropy loss and bounding box regression loss is used as the optimization objective function. The Adam optimizer is used to minimize the training error, and hyperparameters (such as learning rate and batch size) are adjusted step by step. The model performance is evaluated in real time using a validation set to prevent overfitting. After training, the generalization ability of the model is evaluated using a test set to ensure that the model can accurately segment different types of defect regions on the slide rail surface and extract reliable defect features. Finally, the trained slide rail surface morphology detection model is saved for use in subsequent slide rail quality inspection processes.

[0079] In this implementation scheme, a surface morphology detection model for the slide rail is constructed to perform in-depth analysis of the slide rail surface morphology image, thereby achieving accurate identification of different types of surface defects. Secondly, noise suppression and region segmentation are achieved using methods such as filtering and Mask R-CNN, ensuring accurate extraction of the slide rail body from the image and avoiding the influence of background interference on feature extraction. In the defect feature analysis stage, differentiated extraction mechanisms are designed for different defect categories: for scratches, an energy distribution sequence is constructed along the principal axis and the energy attenuation is analyzed to quantitatively reflect the scratch's extensibility; for assembly contamination, the degree of contamination aggregation and area dispersion are extracted using LBP grayscale mode and particle region statistics to reveal the anomalies in surface contamination distribution; and for edge meshing disturbances, the disturbance envelope index is calculated based on the grayscale difference curves of the left and right edge profiles to reflect the stability and force balance of edge meshing. Finally, by weighted fusion of the above features, slide rail morphology damage feature values ​​are generated, ensuring comprehensive detection and detailed characterization capabilities, while also improving the intuitiveness of judgment through numerical results, thereby enhancing detection accuracy.

[0080] Please see Figure 4 This invention provides a technical solution: an artificial intelligence-based slide rail quality inspection system, comprising: an acoustic wave acquisition and analysis module, used to acquire sliding acoustic wave signal data of several windows of a set slide rail, and analyze the aggregated feature values ​​of slide rail acoustic defects in the corresponding windows; an abnormal window identification module, used to identify several abnormal time windows of the set slide rail based on the aggregated feature values ​​of slide rail acoustic defects in each window; a morphological feature analysis module, used to acquire slide rail surface morphological image data of each abnormal window of the set slide rail, and analyze the slide rail morphological damage feature values ​​of the corresponding abnormal window in combination with a pre-trained slide rail surface morphological detection model; and an acoustic-morphological collaborative inspection module, used to read the aggregated feature values ​​of slide rail acoustic defects in each abnormal window of the set slide rail, and analyze the slide rail acoustic-morphological collaborative defect feature values ​​of the set slide rail in combination with the slide rail morphological damage feature values ​​of the corresponding abnormal window, and perform quality inspection processing.

[0081] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0082] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for inspecting the quality of slide rails based on artificial intelligence, characterized in that, Includes the following steps: Acquire the sliding acoustic wave signal data of several windows of the set slide rail, and analyze the aggregated characteristic values ​​of the slide rail acoustic defects in the corresponding windows. Specifically: The sliding acoustic wave signal data of each window of the set slide rail is input into the pre-trained acoustic defect detection model, and the acoustic defect mapping feature set of the corresponding window is analyzed, including the feature value of cumulative friction bias intensity, the feature value of spectral imbalance fluctuation, and the feature value of stick-slip fluctuation amplitude. Based on the acoustic defect mapping feature set of each window of the set slide rail, analyze the aggregated feature value of the slide rail acoustic defect of the corresponding window. Based on the aggregated feature value of the acoustic defect of the slide rail in each window of the set slide rail, several abnormal time windows of the set slide rail are identified. The surface topography image data of each abnormal window of the designated slide rail is obtained, and combined with the pre-trained slide rail surface topography detection model, the slide rail topography damage feature values ​​of the corresponding abnormal windows are analyzed, specifically as follows: The slide rail surface morphology image data of each abnormal window of the slide rail is input into the pre-trained slide rail surface morphology detection model, and the morphology defect feature set of the corresponding abnormal window is analyzed, including scratch energy attenuation feature value, assembly contamination abnormal feature value, and slide rail edge meshing disturbance feature value. Based on the feature set of morphological defects of each abnormal window of the set slide rail, the feature value of the morphological damage of the corresponding abnormal window slide rail is analyzed. Read the aggregated feature value of the acoustic texture defect of each abnormal window of the set slide rail, and combine it with the slide rail morphological damage feature value of the corresponding abnormal window to analyze the slide rail acoustic-morphological collaborative defect feature value of the set slide rail, and perform quality inspection processing. The specific formula for calculating the characteristic value of the acoustic-visual coordination defect of the set slide rail is as follows: ； in, To set the characteristic value of the acoustic-visual coordination defect of the slide rail, , The sequence is the first step of setting the slide rail. The aggregated characteristic values ​​of the sound signature defects of the slide rail and the characteristic values ​​of the slide rail morphology damage in each abnormal window. , , , , The following are, in order, the adjustment coefficients for voiceprint defects, morphological damage, smoothness, difference, and compensation stored in the database: , This represents the total number of abnormal windows.

2. The method for inspecting the quality of slide rails based on artificial intelligence according to claim 1, characterized in that, The sliding acoustic wave signal data specifically refers to the sound amplitude at each time point, and the acoustic defect detection model includes an input layer, a voiceprint analysis layer, a voiceprint evolution layer, and a mapping output layer.

3. The method for inspecting the quality of slide rails based on artificial intelligence according to claim 2, characterized in that, The specific steps for analyzing the acoustic defect mapping feature set of each window of the slide rail are as follows: In the input layer of the acoustic defect detection model, the sliding acoustic wave signal data of each window of the set slide rail is received and preprocessed. In the acoustic signature analysis layer of the acoustic defect detection model, based on the preprocessed sliding acoustic signal data of each window of the set slide rail, the acoustic defect mapping feature vector set of the corresponding window is extracted. In the acoustic pattern evolution layer of the acoustic defect detection model, the temporal feature vector of the corresponding window is analyzed based on the acoustic defect mapping feature vector set of each window of the set slide rail. In the mapping output layer of the acoustic defect detection model, based on the temporal feature vector of the set slide rail, the acoustic defect mapping feature set of its corresponding window is output.

4. The slide rail quality inspection method based on artificial intelligence according to claim 1, characterized in that, The slide rail surface morphology image data specifically refers to the pixel value and two-dimensional coordinates of each pixel point in the slide rail surface morphology image, and the slide rail surface morphology detection model includes a morphology input layer, a defect feature analysis layer, and a fusion output layer.

5. The slide rail quality inspection method based on artificial intelligence according to claim 4, characterized in that, The specific steps for analyzing the feature set of morphological defects for each window of the sliding rail are as follows: In the topography input layer of the slide rail surface topography detection model, the slide rail surface topography image data of each abnormal window of the slide rail is received and image preprocessing is performed. In the defect feature analysis layer of the slide rail surface morphology detection model, based on the slide rail surface morphology image data of each abnormal window of the set slide rail after image preprocessing, the slide rail surface defect feature vector of the corresponding abnormal window is extracted. In the fusion output layer of the slide rail surface morphology detection model, based on the slide rail surface defect feature vector of each abnormal window of the slide rail, the morphology defect feature set of the corresponding abnormal window is output.

6. The slide rail quality inspection method based on artificial intelligence according to claim 1, characterized in that, The specific steps for quality inspection and processing are as follows: The characteristic value of the sliding rail acoustic-visual coordination defect is set and compared with the preset threshold value of the sliding rail acoustic-visual coordination defect. Based on the judgment and processing results, corresponding quality handling measures will be taken.

7. An artificial intelligence-based slide rail quality inspection system, employing the artificial intelligence-based slide rail quality inspection method according to any one of claims 1-6, characterized in that, include: The acoustic wave acquisition and analysis module is used to acquire the sliding acoustic wave signal data of several windows of the set slide rail, and analyze the aggregated characteristic value of the slide rail acoustic texture defect in the corresponding window. The abnormal window identification module is used to identify several abnormal time windows of a set slide rail based on the aggregated feature value of the slide rail acoustic defect of each window. The morphology feature analysis module is used to acquire the surface morphology image data of each abnormal window of the set slide rail, and combine it with the pre-trained slide rail surface morphology detection model to analyze the slide rail morphology damage feature value of the corresponding abnormal window. The acoustic-morphological collaborative inspection module is used to read the aggregated characteristic value of the acoustic texture defect of each abnormal window of the set slide rail, and combine it with the slide rail morphological damage characteristic value of the corresponding abnormal window to analyze the acoustic-morphological collaborative defect characteristic value of the set slide rail, and perform quality inspection processing.

Images