Inspection algorithm of composite machine

Abstract

The invention relates to an inspection algorithm for a composite machine, and the method comprises the steps: obtaining the operation parameters of a function cooperation assembly of the composite machine, eliminating abnormal data through a density clustering model, eliminating false anomalies through the combination of assembly cooperation logic, and generating a standardized parameter set through an interval scaling method; constructing a double-branch feature extraction model, mining single-component local features and cross-component coupling features, and fusing through a feature interaction layer to obtain a multi-dimensional feature set; inputting the feature set into a dual-module detection model, optimizing a random forest to output an initial anomaly probability, performing correction in combination with a working condition and a historical fault rule, performing secondary verification through a component collaboration rule base, and outputting an anomaly parameter type, a time sequence node and a severity degree; abnormal core parameters and grades are positioned, an abnormal conduction path map is generated, a component physical connection relation and a historical maintenance case are matched, and an abnormal basic cause is locked; according to the invention, data reliability and identification accuracy are improved, efficient fault tracing is realized, and maintenance shutdown loss is reduced.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent technology, specifically relating to an inspection algorithm for composite machines. Background Technology

[0002] Composite machines consist of multiple functionally coordinating components with close physical connections and functional coupling. Their operational status directly affects industrial production efficiency and safety. Traditional inspection methods rely on manual operation, which suffers from low efficiency, high missed inspection rates, and strong subjectivity, making them unsuitable for real-time inspection needs under complex working conditions.

[0003] Existing intelligent inspection algorithms have several shortcomings: First, data preprocessing only removes outliers without considering environmental noise interference and false anomalies caused by component collaboration logic, affecting data reliability. Second, feature extraction often focuses on a single local or global dimension, ignoring the interaction between the two, resulting in incomplete feature representation. Third, the anomaly detection model does not combine real-time machine operating conditions and historical fault patterns, resulting in insufficient accuracy in anomaly judgment and a lack of component collaboration constraint verification. Fourth, it can only identify abnormal phenomena but cannot locate the anomaly propagation path and root cause, increasing maintenance difficulty and downtime. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a composite machine inspection algorithm. The objective of this invention can be achieved through the following technical solution: A composite machine inspection algorithm includes: S1: Obtain the operating status parameters of the functional collaboration components of the composite machine, construct a density clustering model, calculate the Euclidean distance between samples based on the preset neighborhood radius threshold and minimum sample number threshold, and obtain a standardized parameter set based on the component collaboration logic and the interval scaling method. S2: Based on the standardized parameter set, a dual-branch feature extraction model is constructed through the collaborative characteristics of composite machine components: the local feature extraction branch mines the local features of each collaborative component; the global feature extraction branch extracts cross-component parameter coupling features; a feature interaction layer is added to fuse the local features and parameter coupling features to generate a multi-dimensional feature set; S3: Input the multi-dimensional feature set into a preset dual-module fusion detection model, including optimizing a random forest classifier and a component collaborative rule base, outputting an initial anomaly probability value based on training samples, constructing a dynamic correction function by introducing real-time machine operating conditions and historical fault patterns, and dynamically correcting the initial anomaly probability value; perform a secondary verification on the corrected anomaly probability value according to the component's rated operating threshold and the inherent collaborative constraint relationship between components, and output the anomaly parameter type, anomaly time sequence node, and anomaly severity assessment results; S4: Based on the abnormal parameter type, locate the core abnormal parameters and their fluctuation characteristics at time-series nodes, and classify the abnormal level by associating them with the severity assessment results, and dynamically couple them with the component parameters. Combine the time-series nodes to analyze the transmission time sequence of the abnormal signal and generate a transmission path map. By matching the physical connection relationship of components in the machine structure principle and the root cause cases of similar abnormalities in historical maintenance data, generate the root cause of the abnormality.

[0005] Specifically, the density clustering model is constructed based on the physical characteristics of component parameters and the distribution characteristics of historical normal operation data. A neighborhood radius threshold and a minimum sample number threshold are preset for the parameter type. After preprocessing the original parameter sequence, the Euclidean distance between parameter samples is calculated, the density correlation of sample points is determined, and isolated points that do not belong to any cluster are marked as abnormal data and removed.

[0006] Specifically, the remaining data is verified based on the component collaboration logic. Two types of core verification rules are preset: mechanical linkage rules, which clarify the parameter linkage ratio based on the physical connection relationship of components; and function matching rules, which set the parameter adaptation range in combination with the operation process of composite machines. When the parameter of a single component exceeds the normal fluctuation range but the parameters of related components all meet the verification rules, it is judged as a false abnormal parameter and removed.

[0007] Specifically, the interval scaling method processes the valid parameters after anomaly handling, and uses the safe operating range of each parameter as a benchmark. Through linear transformation, all parameters are uniformly mapped to a preset range to generate a standardized parameter set.

[0008] Specifically, based on the standardized parameter set and the collaborative characteristics of composite machine components, a dual-branch feature extraction model is constructed: The local feature extraction branch extracts features based on the core operating parameters of each coordinating component through sliding window time-domain analysis: for the spindle operating parameter sequence, the peak deviation within the continuous sampling window is calculated to obtain the radial runout feature; for the conveyor belt tension parameter sequence, the standard deviation of the fluctuation within the window is statistically analyzed to obtain the tension fluctuation feature. The radial runout feature and the tension fluctuation feature constitute a local feature set for a single component. The global feature extraction branch calculates cross-component coupling features based on pre-set collaborative constraints between components: the speed ratio deviation is obtained by the absolute difference between the parameter ratio and the theoretical ratio; the timing synchronization error is obtained by the deviation between the time difference between the drive motor action time and the signal acquisition start time and the threshold, thus forming a cross-component coupling feature set.

[0009] Specifically, the added feature interaction layer, based on the physical connection strength and working conditions of the composite machine components, performs weighted concatenation of the local feature set of the single component and the cross-component coupling feature set, and at the same time suppresses weakly correlated features through a gating unit to generate the multi-dimensional feature set.

[0010] Specifically, the optimized random forest classifier expands the data volume by using synthetic sampling for scarce outlier samples, assigns differentiated weights to outlier samples, adopts a node splitting strategy based on feature importance when constructing each decision tree, and removes overfitting branches through pruning operations; during fitting calculation, the classifier performs path traversal on each feature vector in the multi-dimensional feature set, combines the output results of all decision trees to perform probability aggregation, and outputs the initial outlier probability value of the corresponding component.

[0011] Specifically, the process of constructing a dynamic correction function to calibrate and adjust the initial anomaly probability value is as follows: taking the initial anomaly probability value as the basic input, firstly, a correlation mapping relationship is established between machine operating parameters and historical fault pattern data; then, the influence weights of operating parameters and historical fault characteristics on the initial anomaly probability value are determined through statistical analysis, and corresponding correction coefficients are generated for the operating scenario; finally, the initial anomaly probability value and the corresponding correction coefficient are weighted and calculated, and the calculation results are dynamically adjusted in combination with the matching degree between the operating condition and historical data to obtain the calibrated anomaly probability value.

[0012] Specifically, the component collaboration rule base includes component rated operating threshold rules and inherent collaboration constraint rules between components; the local component collaboration rule base performs secondary verification on the corrected anomaly probability value: by comparing the actual value of the parameter with the dynamic threshold range through the rated threshold rule, the deviation is recorded to form a single component anomaly feature; the single component feature is matched with the parameter status of the associated component through the collaboration constraint rules; when the corrected anomaly probability value exceeds the preset threshold and the secondary verification passes, the identification result containing the anomaly parameter type, time sequence node, and severity assessment is output.

[0013] Specifically, the method for locating and classifying the abnormal core parameters is as follows: based on the abnormal parameter type, time series node, and severity assessment result, the standardized time series curve of the corresponding parameter is retrieved, and the fluctuation characteristics of the parameter at the time series node, including peak offset, mutation amplitude, and duration, are identified to obtain the abnormal core parameters; combined with the numerical range of the severity assessment result, the initial abnormal level is classified.

[0014] Specifically, the abnormal transmission path map generation process involves: calling the extracted dynamic coupling features between components, taking the abnormal core parameters and time sequence nodes as the starting point, analyzing the transmission timing of abnormal signals between different components, marking the abnormal response order and time difference of the components, and generating a transmission path map of "abnormal source - associated components - scope of influence - transmission time difference".

[0015] Specifically, the root cause identification process is as follows: matching the physical connection relationship of components in the machine structure principle, verifying the rationality of component association in the transmission path diagram; retrieving the historical maintenance database, filtering out similar abnormal cases that are consistent with the abnormal core parameter type, initial abnormal level, and transmission path, extracting the root cause in the case, performing causal verification in combination with the machine operating condition, and finally generating the root cause that matches the current abnormal scenario.

[0016] The beneficial effects of this invention are as follows: Breaking through the limitations of existing technologies that only remove outliers, this invention pioneers a dual-layer mechanism of "density clustering anomaly identification and component collaborative logic verification." This mechanism accurately removes isolated outliers from the original parameters and effectively filters out false anomalies caused by collaborative logic, achieving data purity and reliability that traditional preprocessing methods cannot match. Targeting the collaborative characteristics of multi-component composite machines, a dual-branch extraction architecture of "local details + cross-component coupling" plus a feature interaction layer is designed. This architecture covers both local features in the time / frequency domain of single components and captures coupling features such as rotational speed ratios and timing synchronization between components. After deep fusion, the comprehensiveness and accuracy of feature characterization of abnormal states far surpasses existing single-dimensional feature extraction technologies. Integrating the triple innovation of "optimized random forest + dynamic correction + secondary verification of collaborative rules," this approach enhances the sensitivity of severe anomaly identification through sample expansion and differentiated weights, dynamically adapts to complex scenarios by combining real-time operating conditions and historical fault patterns, and filters out false positives through cross-component collaborative verification. The accuracy and adaptability of anomaly detection are significantly better than existing single-detection models. It closely aligns with the core characteristics of physical connection and functional coupling of multiple components in composite machines. The entire process algorithm design revolves around the component collaboration logic, making it far more adaptable and practical than general inspection systems that do not consider collaboration characteristics. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart illustrating an inspection algorithm for a composite machine according to the present invention. Figure 2 This is a structural diagram of the dual-branch feature extraction and fusion module in this invention; Figure 3 This is a correlation diagram of anomaly detection, secondary verification, and fault tracing in this invention. Detailed Implementation

[0019] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of example embodiments to those skilled in the art. Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring aspects of this disclosure. The blocks shown in the drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the drawings are merely illustrative and do not necessarily include all contents and operations / steps, nor do they necessarily have to be performed in the order described. For example, some operations / steps can be broken down, while others can be combined or partially combined. Therefore, the actual execution order may change depending on the actual situation.

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0021] Please see Figure 1-3 A composite machine inspection algorithm, comprising: S1: Obtain the operating status parameters of the functional collaboration components of the composite machine, construct a density clustering model, calculate the Euclidean distance between samples based on the preset neighborhood radius threshold and minimum sample number threshold, and obtain a standardized parameter set based on the component collaboration logic and the interval scaling method. S2: Based on the standardized parameter set, a dual-branch feature extraction model is constructed through the collaborative characteristics of composite machine components: the local feature extraction branch mines the local features of each collaborative component; the global feature extraction branch extracts cross-component parameter coupling features; a feature interaction layer is added to fuse the local features and parameter coupling features to generate a multi-dimensional feature set; S3: Input the multi-dimensional feature set into a preset dual-module fusion detection model, including optimizing a random forest classifier and a component collaborative rule base, outputting an initial anomaly probability value based on training samples, constructing a dynamic correction function by introducing real-time machine operating conditions and historical fault patterns, and dynamically correcting the initial anomaly probability value; perform a secondary verification on the corrected anomaly probability value according to the component's rated operating threshold and the inherent collaborative constraint relationship between components, and output the anomaly parameter type, anomaly time sequence node, and anomaly severity assessment results; S4: Based on the abnormal parameter type, locate the core abnormal parameters and their fluctuation characteristics at time-series nodes, and classify the abnormal level by associating them with the severity assessment results, and dynamically couple them with the component parameters. Combine the time-series nodes to analyze the transmission time sequence of the abnormal signal and generate a transmission path map. By matching the physical connection relationship of components in the machine structure principle and the root cause cases of similar abnormalities in historical maintenance data, generate the root cause of the abnormality.

[0022] Specifically, the density clustering model is constructed based on the physical characteristics of component parameters and the distribution characteristics of historical normal operation data. It presets neighborhood radius thresholds and minimum sample number thresholds for different parameter types. After preprocessing the original parameter sequence, it calculates the Euclidean distance between each parameter sample, determines the density correlation of sample points, and marks isolated points that do not belong to any cluster as abnormal data and removes them.

[0023] Specifically, the remaining data is verified based on the component collaboration logic. Two types of core verification rules are preset: mechanical linkage rules, which clarify the parameter linkage ratio based on the physical connection relationship of components; and function matching rules, which set the parameter adaptation range in combination with the operation process of composite machines. When the parameter of a single component exceeds the normal fluctuation range but the parameters of related components all meet the verification rules, it is judged as a false abnormal parameter and removed.

[0024] Specifically, the interval scaling method processes the valid parameters after anomaly handling, and uses the safe operating range of each parameter as a benchmark. Through linear transformation, all parameters are uniformly mapped to a preset range to generate a standardized parameter set.

[0025] Specifically, based on the standardized parameter set and the collaborative characteristics of composite machine components, a dual-branch feature extraction model is constructed: The local feature extraction branch extracts features based on the core operating parameters of each coordinating component through sliding window time-domain analysis: for the spindle operating parameter sequence, the peak deviation within the continuous sampling window is calculated to obtain the radial runout feature; for the conveyor belt tension parameter sequence, the standard deviation of the fluctuation within the window is statistically analyzed to obtain the tension fluctuation feature. The radial runout feature and the tension fluctuation feature constitute a local feature set for a single component. The global feature extraction branch calculates cross-component coupling features based on pre-set collaborative constraints between components: the speed ratio deviation is obtained by the absolute difference between the parameter ratio and the theoretical ratio; the timing synchronization error is obtained by the deviation between the time difference between the drive motor action time and the signal acquisition start time and the threshold, thus forming a cross-component coupling feature set.

[0026] Specifically, the added feature interaction layer, based on the physical connection strength and working conditions of the composite machine components, performs weighted concatenation of the local feature set of the single component and the cross-component coupling feature set, and at the same time suppresses weakly correlated features through a gating unit to generate the multi-dimensional feature set.

[0027] In this embodiment, a multi-station composite machining center is used as the application object. This machine tool includes three major collaborative components: a machining component (the core component is the spindle), a transmission component (the core component is the conveyor belt), and a detection component (the core component is the signal acquisition module). Feature extraction and fusion are achieved based on the aforementioned method. The specific process is as follows: The machining component spindle, the transmission component conveyor belt, and the detection component signal acquisition module are selected as the core collaborative components. Based on a standardized parameter set (covering parameters such as spindle speed, conveyor belt tension, and signal acquisition frequency, which have been mapped to a preset range), features are extracted through a local feature extraction branch. Set the sliding window duration to T, the sampling frequency to f, and each window to contain n sampling points; Temporal feature extraction: Calculate the peak deviation of n consecutive sampling points of the main shaft to obtain the radial runout; statistically analyze k times the standard deviation of the conveyor belt tension sensor readings to obtain the tension fluctuation; Frequency domain feature extraction: Perform fast Fourier transform (frequency range 0-F) on the spindle vibration parameters and drive motor current parameters to extract the main frequency component of the spindle vibration (AB range) and the proportion of specific harmonics of the drive motor current (≤C). Integrate time-domain and frequency-domain features to form a single-component local feature set (each component contains several feature dimensions).

[0028] Based on the machine tool's pre-defined collaborative logic (theoretical spindle to conveyor belt speed ratio 1:D, and the drive motor current change rate and signal acquisition trigger delay must be positively correlated), calculate the cross-component coupling characteristics: Speed ​​ratio deviation: The absolute difference between the actual ratio of the spindle speed to the conveyor belt speed and the theoretical ratio 1 / D, calculated in real time. Current-delay correlation: The Pearson correlation coefficient between the rate of change of the drive motor current and the signal acquisition trigger delay is calculated by using a sliding window (window duration of 2T) to quantify the correlation between the two. Sampling-speed adaptation deviation: The relative deviation between the actual sampling frequency and the preset value is calculated based on the preset sampling frequency of the detection component according to the conveyor belt running speed; By integrating the above features, a cross-component coupled feature set (containing several feature dimensions) is formed.

[0029] An additional feature interaction layer is added to achieve fusion based on the physical connection relationship of machine tool components and real-time operating conditions (high-speed machining mode, spindle speed ≥ S): Set the physical connection strength weights: the spindle and conveyor belt are directly connected by transmission, so the weight is set to W1; the drive motor and detection components are indirectly connected by signals, so the weight is set to W2. Dynamically adjust operating condition weights: During high-speed machining, the weight of the timing coordination feature is increased to W3, and the weight of the component independent operation feature is adjusted to W4; Two types of features are weighted and stitched together: local features are weighted according to component weights (main axis w1, conveyor belt w2, detection component w3), and coupled features are weighted according to correlation strength weights, and then the features are stitched together. Gating unit suppression: Set the correlation coefficient threshold to θ, and remove weak correlation features (such as frequency domain harmonic features of the sampling frequency of the detection component) that have a correlation coefficient < θ with the core operating state of the machine tool after splicing. Finally, a multi-dimensional feature set containing m core feature dimensions is generated to provide input for subsequent anomaly detection.

[0030] Specifically, the optimized random forest classifier expands the data volume by using synthetic sampling for scarce outlier samples, and assigns differentiated weights to outlier samples of different severity. When constructing each decision tree, it adopts a node splitting strategy based on feature importance and removes overfitting branches through pruning operations. During fitting calculation, the classifier performs path traversal on each feature vector in the multi-dimensional feature set, combines the output results of all decision trees to perform probability aggregation, and outputs the initial outlier probability value of the corresponding component.

[0031] Specifically, the process of constructing a dynamic correction function to calibrate and adjust the initial anomaly probability value is as follows: taking the initial anomaly probability value as the basic input, firstly, a correlation mapping relationship is established between machine operating parameters and historical fault pattern data; then, the influence weights of operating parameters and historical fault characteristics on the initial anomaly probability value are determined through statistical analysis, and corresponding correction coefficients are generated for the operating scenario; finally, the initial anomaly probability value and the corresponding correction coefficient are weighted and calculated, and the calculation results are dynamically adjusted in combination with the matching degree between the operating condition and historical data to obtain the calibrated anomaly probability value.

[0032] Specifically, the component collaboration rule base includes component rated operating threshold rules and inherent collaboration constraint rules between components; the local component collaboration rule base performs secondary verification on the corrected anomaly probability value: by comparing the actual value of the parameter with the dynamic threshold range through the rated threshold rule, the deviation is recorded to form a single component anomaly feature; the single component feature is matched with the parameter status of the associated component through the collaboration constraint rules; when the corrected anomaly probability value exceeds the preset threshold and the secondary verification passes, the identification result containing the anomaly parameter type, time sequence node, and severity assessment is output.

[0033] In this embodiment, the anomaly detection of a multi-station composite machining tool is used as the application scenario. Based on the aforementioned method, a dual-module fusion detection model composed of an optimized random forest classifier and a component collaborative rule base is constructed. To address the scarcity of abnormal machine tool samples (such as spindle jamming and conveyor belt slippage), a synthetic sampling method is used to expand the original sample set: for severe abnormal samples (such as spindle radial runout exceeding the safety threshold), synthetic samples with the same distribution as the original samples are generated, increasing the number of severe abnormal samples to 1.5 times the original size; for minor abnormal samples (such as slight fluctuations in conveyor belt tension), synthetic samples are added according to the original distribution, increasing the sample size to 1.2 times the original size.

[0034] To enhance the accuracy of identifying severe anomalies, differentiated weights are assigned to the samples: severe anomaly samples are assigned a high weight, such as a1; slightly anomaly samples are assigned a medium weight, such as a2; and normal samples are assigned a low weight, such as a3, with a1>a2>a3.

[0035] When constructing the decision tree, a node splitting strategy based on feature importance is adopted: features that contribute highly to anomaly identification (such as spindle vibration frequency shift and rotation speed ratio deviation) are prioritized as splitting attributes; at the same time, branches with depths exceeding a preset value are removed through pruning operations (such as retaining only effective branches with depths ≤ d) to avoid overfitting.

[0036] During the fitting calculation stage, the classifier traverses all decision tree paths for each feature vector in the multi-dimensional feature set (such as features including spindle radial runout, speed ratio deviation, etc.), counts the anomaly judgment results output by each tree, and calculates and outputs the initial anomaly probability value of the component (such as spindle anomaly probability P1, conveyor belt anomaly probability P2) through probability aggregation (such as majority voting weighting).

[0037] Based on the initial anomaly probability value, real-time machine tool operating parameters (such as the current spindle speed s and the hardness of the processed material h) and historical fault pattern data (such as the spindle anomaly rate reaching 60% when processing high-hardness materials) are introduced to establish a correlation mapping relationship: the influence weight of real-time operating conditions and historical characteristics is determined through statistical analysis (such as the weight of speed s being α, the weight of material hardness h being β, and α+β=1).

[0038] Correction coefficients are generated for different working conditions: when the spindle speed is greater than or equal to the high-speed threshold s0, the correction coefficient is set to k1 (to enhance the sensitivity to potential anomalies during high-speed operation); when the hardness of the processed material is greater than or equal to the high hardness threshold h0, the correction coefficient is set to k2 (to match the abnormal patterns during historical high-hardness processing).

[0039] The initial anomaly probability value is weighted and calculated with the corresponding correction coefficient (e.g., calibrated probability = P1×k1×α + P1×k2×β). At the same time, the result is dynamically adjusted by combining the matching degree between real-time operating conditions and historical data (e.g., the correction amplitude is enhanced when the matching degree is ≥80%), and finally the calibrated anomaly probability value is obtained.

[0040] Secondary validation of component collaboration rule base The component collaboration rule base contains two core types of rules: Rated operating threshold rules: preset component dynamic threshold range (e.g., the safe range for spindle radial runout is [r_min, r_max], which is dynamically adjusted with the rotational speed s); Collaborative constraint rules: Clearly define the inter-component constraints (e.g., the spindle speed s and the conveyor belt speed v must satisfy s / v∈[k_min, k_max]).

[0041] During the secondary verification, the actual value of the parameter is first compared with the dynamic threshold (such as the deviation between the actual value r of the spindle radial runout and [r_min, r_max]) using the nominal threshold rule to form the abnormal characteristics of a single component; then, the characteristics of the single component are matched with the state of the associated component using the collaborative constraint rule (such as checking whether v deviates synchronously from the constraint range of s / v when r exceeds the limit).

[0042] When the abnormal probability value after calibration exceeds the preset threshold and the secondary verification passes (single component feature abnormality and the state of associated components does not meet the collaborative constraint), the following identification results are output: abnormal parameter type (main shaft radial runout abnormality), abnormal time sequence node (time t), severity assessment (moderate abnormality, because the deviation does not exceed the threshold by 1.5 times).

[0043] Specifically, the method for locating and classifying the abnormal core parameters is as follows: based on the abnormal parameter type, time series node, and severity assessment result, the standardized time series curve of the corresponding parameter is retrieved, and the fluctuation characteristics of the parameter at the time series node, including peak offset, mutation amplitude, and duration, are identified to obtain the abnormal core parameters; combined with the numerical range of the severity assessment result, the initial abnormal level is classified.

[0044] Specifically, the abnormal transmission path map generation process involves: calling the extracted dynamic coupling features between components, taking the abnormal core parameters and time sequence nodes as the starting point, analyzing the transmission timing of abnormal signals between different components, marking the abnormal response order and time difference of the components, and generating a transmission path map including "abnormal source - associated components - scope of influence - transmission time difference".

[0045] Specifically, the root cause identification process is as follows: matching the physical connection relationship of components in the machine structure principle, verifying the rationality of component association in the transmission path diagram; retrieving the historical maintenance database, filtering out similar abnormal cases that are consistent with the abnormal core parameter type, initial abnormal level, and transmission path, extracting the root cause in the case, combining it with the current real-time operating condition of the machine to perform causal verification, and finally generating the root cause that matches the current abnormal scenario.

[0046] In this embodiment, taking the "spindle vibration anomaly" detection scenario of a multi-station composite machining center as an example, the entire anomaly analysis process is completed based on the aforementioned method, as follows: The known anomaly detection results are as follows: the anomaly parameter type is "spindle vibration anomaly", the timing node is time t0 of the machining stage, and the severity assessment result is in the value range of [L1, L2) (moderate anomaly range).

[0047] Locating the core parameters of the anomaly: retrieve the standardized time-series curve of the spindle vibration (horizontal axis is time, vertical axis is the standardized vibration value of the preset interval), and analyze the fluctuation characteristics at time t0: Peak offset: The deviation of the vibration peak value at time t0 from the steady-state peak value under normal operating conditions by Δa standardized units; Abrupt change amplitude: From t0-Δt to t0, the vibration value suddenly increases from the reference value V0 to V1, with an abrupt change amplitude of Δb; Duration: The abnormal vibration state tends to stabilize after about m sampling cycles.

[0048] Based on the fluctuation characteristics, the "main shaft vibration amplitude" was confirmed as the core abnormal parameter; according to the numerical range [L1, L2) of the severity assessment results, the initial abnormality level was classified as "moderate abnormality".

[0049] By invoking the dynamic coupling characteristics between components (such as the rotational speed coupling coefficient between the spindle and the conveyor belt, and the signal synchronization characteristics between the conveyor belt and the detection component), and taking "abnormal spindle vibration (time t0)" as the starting point, the timing of abnormal signal propagation is analyzed: At time t0: The spindle vibrates abnormally (abnormal source), and the vibration amplitude exceeds the steady-state threshold; At time t0+Δt1: The speed of the conveyor belt, which is directly connected to the main shaft via gears, fluctuates due to vibration transmission, and the rotational speed deviates from the theoretical value by Δc standardized units (associated component 1, transmission time difference Δt1). At time t0+Δt1+Δt2: the fluctuation of the conveyor belt speed causes the sampling trigger signal of the detection component to be delayed, and the mismatch between the sampling frequency and the conveyor belt running speed increases to Δd (associated component 2, conduction time difference Δt2).

[0050] The abnormal response sequence of the components (main shaft → conveyor belt → detection component) and the time difference (Δt1, Δt2) are marked, and a transmission path map is generated: abnormal source (main shaft, t0) → associated component 1 (conveyor belt, t0+Δt1) → associated component 2 (detection component, t0+Δt1+Δt2), and the scope of influence covers the entire process of processing-transmission-detection.

[0051] Verify the rationality of the transmission path: Matching the machine tool structure principle, the spindle and the conveyor belt are directly driven by a reduction gear set (high physical connection strength). The operating state of the conveyor belt directly affects the sampling triggering logic of the detection component (closely related functions). The transmission path conforms to the physical connection and functional coordination relationship of the components, and the logic is valid.

[0052] Retrieve historical maintenance database: Filter out similar cases with "abnormal spindle vibration + moderate level + transmission path 'spindle → conveyor belt → detection component'", and find that the root cause of n historical cases is "excessive gear meshing clearance between the spindle and the conveyor belt".

[0053] Causal verification: Based on the current real-time operating conditions of the machine tool (continuous machining of high-hardness materials for T hours, with continuously high gear load), the gear meshing clearance is prone to increase due to increased wear under high load, which leads to vibration transmission to downstream components. This is completely consistent with the causes of historical cases and the current abnormal characteristics.

[0054] The root cause of the current anomaly was finally identified as "excessive clearance in the spindle-conveyor belt connecting gears (due to accelerated wear caused by high-load machining)".

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A composite machine inspection algorithm, characterized in that, Includes the following steps: S1: Obtain the operating status parameters of the functional collaboration components of the composite machine, construct a density clustering model, calculate the Euclidean distance between samples based on the preset neighborhood radius threshold and minimum sample number threshold, and obtain a standardized parameter set based on the component collaboration logic and the interval scaling method. S2: Based on the standardized parameter set, a dual-branch feature extraction model is constructed through the collaborative characteristics of composite machine components: the local feature extraction branch mines the local features of each collaborative component; the global feature extraction branch extracts cross-component parameter coupling features; a feature interaction layer is added to fuse the local features and parameter coupling features to generate a multi-dimensional feature set; S3: Input the multi-dimensional feature set into a preset dual-module fusion detection model, including an optimized random forest classifier and a component collaborative rule base, and output an initial anomaly probability value based on the training samples. A dynamic correction function is constructed by introducing machine operating conditions and historical fault patterns to dynamically correct the initial abnormal probability value; based on the rated operating threshold of the components and the inherent collaborative constraint relationship between the components, the corrected abnormal probability value is verified, and the abnormal parameter type, abnormal timing node and abnormal severity assessment result are output. S4: Based on the abnormal parameter type, locate the core abnormal parameters and their fluctuation characteristics, and classify the abnormal level based on the severity assessment results; combine the dynamic coupling relationship of component parameters and abnormal timing nodes to analyze the transmission timing of abnormal signals and generate an abnormal transmission path map; by matching the physical connection relationship of components in the machine structure principle and similar abnormal root cause cases in historical maintenance data, obtain the root cause of the abnormality.

2. The method according to claim 1, characterized in that, In S1, the density clustering model is constructed based on the physical characteristics of component parameters and the distribution characteristics of historical normal operation data. The neighborhood radius threshold and minimum sample number threshold of parameter types are preset. After preprocessing the original parameter sequence, the Euclidean distance between parameter samples is calculated to obtain the density correlation of sample points. Isolated points that do not belong to any cluster are marked as abnormal data and removed.

3. The method according to claim 1, characterized in that, In S1, the remaining data is verified based on the component collaboration logic. Two types of core verification rules are preset: mechanical linkage rules, which clarify the parameter linkage ratio based on the physical connection relationship of the components; and function matching rules, which set the parameter range in combination with the operation process of the composite machine. When a single component parameter exceeds the preset normal fluctuation range, but the parameters of related components all meet the verification rules, it is judged as a false abnormal parameter and removed.

4. The method according to claim 1, characterized in that, In S1, the interval scaling method processes the valid parameters after anomaly handling. Based on the safe operating range of each parameter, it uses linear transformation to uniformly map all parameters to a preset range, generating a standardized parameter set.

5. The method according to claim 1, characterized in that, In S2, a dual-branch feature extraction model is constructed based on the standardized parameter set and the collaborative characteristics of composite machine components: The local feature extraction branch extracts features based on the core operating parameters of each collaborative component through sliding window time-domain analysis: for the spindle operating parameter sequence, the peak deviation within the continuous sampling window is calculated to obtain the radial runout feature; For the tension parameter sequence of the conveyor belt, the standard deviation of the fluctuation within the statistical window is used to obtain the tension fluctuation characteristics. The radial runout characteristics and the tension fluctuation characteristics constitute a local feature set of a single component. The global feature extraction branch calculates cross-component coupling features based on pre-set collaborative constraints between components: the speed ratio deviation is obtained by the absolute difference between the parameter ratio and the theoretical ratio; the timing synchronization error is obtained by the deviation between the time difference between the drive motor action time and the signal acquisition start time and the threshold, thus forming a cross-component coupling feature set.

6. The method according to claim 1, characterized in that, In S2, the added feature interaction layer, based on the physical connection strength and working conditions of the composite machine components, performs weighted concatenation of the local feature set of the single component and the cross-component coupling feature set, and at the same time suppresses weakly correlated features through the gating unit to generate the multi-dimensional feature set.

7. The method according to claim 1, characterized in that, In S3, the optimized random forest classifier expands the data volume by using synthetic sampling for scarce outlier samples, assigns differential weights to outlier samples, adopts a node splitting strategy based on feature importance when constructing each decision tree, and removes overfitting branches through pruning operations. During the fitting calculation, the classifier performs path traversal on each feature vector in the multi-dimensional feature set, combines the output results of all decision trees to perform probability aggregation, and outputs the initial anomaly probability value of the corresponding component.

8. The method according to claim 1, characterized in that, In S3, the specific process of constructing a dynamic correction function to calibrate and adjust the initial anomaly probability value is as follows: taking the initial anomaly probability value as the basic input, firstly, a correlation mapping relationship is established between the machine's real-time operating parameters and historical fault pattern data; then, through statistical analysis, the influence weights of each operating parameter and historical fault characteristics on the initial anomaly probability value are determined, and corresponding correction coefficients are generated; finally, the initial anomaly probability value and the corresponding correction coefficients are weighted and calculated, and the calculation results are dynamically adjusted in combination with the matching degree between the operating conditions and historical data to obtain the calibrated anomaly probability value.

9. The method according to claim 1, characterized in that, In S3, the component collaboration rule base includes component rated operating threshold rules and inherent collaboration constraint rules between components; the component collaboration rule base performs secondary verification on the corrected anomaly probability value: by comparing the actual value of the parameter with the dynamic threshold range through the rated threshold rule, the deviation is recorded to form a single component anomaly feature; the single component feature is matched with the parameter status of the associated component through the collaboration constraint rule; when the corrected anomaly probability value exceeds the preset threshold and the secondary verification passes, the identification result including the anomaly parameter type, time sequence node, and severity assessment is output.

10. The method according to claim 1, characterized in that, In S4, the method for locating and classifying the abnormal core parameters is as follows: based on the abnormal parameter type, time series node, and severity assessment result, the standardized time series curve of the corresponding parameter is retrieved, and the fluctuation characteristics of the parameter at the time series node, including peak offset, mutation amplitude, and duration, are identified to obtain the abnormal core parameters; combined with the numerical range of the severity assessment result, the initial abnormal level is classified.

11. The method according to claim 1, characterized in that, In S4, the abnormal transmission path map generation process involves: calling the extracted dynamic coupling features between components, taking the abnormal core parameters and time sequence nodes as the starting point, analyzing the transmission timing of abnormal signals between components, marking the abnormal response sequence and time difference of components, and generating a transmission path map of "abnormal source - associated component - scope of influence - transmission time difference".

12. The method according to claim 1, characterized in that, In S4, the root cause locking process is as follows: matching the physical connection relationship of components in the machine structure principle, verifying the rationality of component association in the transmission path map; retrieving the historical maintenance database, filtering out similar abnormal cases that are consistent with the abnormal core parameter type, initial abnormal level, and transmission path, extracting the root cause in the case, performing causal verification in combination with the current machine, and finally generating the root cause that matches the current abnormal scenario.

Images