A three-coordinate measurement action recognition system based on semiconductor processing equipment

Abstract

The present application relates to the technical field of semiconductor processing, in particular to a three-coordinate measurement action recognition system based on semiconductor processing equipment. It comprises a motion data acquisition module, an abnormal measurement action recognition module, a detection action generation module, a proactive detection module and an abnormal feature comparison module. The system determines single-axis abnormality or coupled abnormality through motion feature correlation analysis, generates a detection sequence containing single-axis detection path and multi-axis linkage detection path for high-frequency abnormalities, and actively executes during the idle period of the equipment. The motion feature difference between single-axis and multi-axis linkage detection is compared to determine the source of the fault, the abnormal source is distinguished through the comparison of the characteristics of the empty load and the load state and the difference degree gradient between load groups, and the fault type is identified based on the time-frequency-time sequence form multi-dimensional feature fingerprint library, forming a complete abnormal detection and diagnosis system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor processing, in particular to a three-coordinate measurement action recognition system based on a semiconductor processing device. BACKGROUND

[0002] The three-coordinate measurement device in the semiconductor processing device is used for precise measurement of the size of a workpiece, and the measurement accuracy directly affects the product quality. With the shrinking of the semiconductor process node, the reliability requirement of the measurement device is higher and higher, and early identification of abnormal measurement action and accurate fault diagnosis become the key to ensuring production continuity.

[0003] The prior art mainly adopts a passive monitoring method to detect abnormalities during production measurement, which has the following technical limitations: first, there is a lack of active detection mechanism, only abnormalities are found during production measurement, potential faults cannot be detected in advance during idle periods of the device, resulting in late fault discovery. Second, the abnormal source determination capability is insufficient, it is difficult to systematically distinguish whether the abnormality is caused by device degradation, process factors or environmental factors, there is a lack of effective traceability analysis method, and maintenance decisions rely on experience. Third, there is a lack of multi-axis coupling abnormality diagnosis capability, it is difficult to determine whether the fault root cause is a mechanical coupling structure problem or a multi-axis synchronization problem of the control system, and it is impossible to provide clear fault location. Fourth, the fault feature recognition dimension is single, mainly based on time domain features for abnormal detection, and the fault type recognition capability for features with frequency domain or motion trajectory shape characteristics is limited. SUMMARY

[0004] The present application provides a three-coordinate measurement action recognition system based on a semiconductor processing device, which realizes active detection of abnormal measurement action and fault traceability.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] The three-coordinate measurement action recognition system based on a semiconductor processing device provided by the present application comprises:

[0007] A motion data acquisition module acquires position, speed and acceleration time series data of each motion axis of the three-coordinate measurement device;

[0008] An abnormal measurement action recognition module extracts motion features of the measurement action according to the time series data and compares them with a normal measurement action feature library, identifies an abnormal measurement action and outputs an abnormal type and motion features when the deviation exceeds a preset threshold, and calculates the correlation of motion features of each motion axis when multiple motion axes simultaneously appear abnormal, and determines single-axis abnormality or coupling abnormality;

[0009] The probe action generation module, the abnormal measurement action frequency is counted, the motion parameter interval of the high-frequency abnormal measurement action is extracted, and the targeted probe action sequence is generated based on the motion parameter interval. The probe action sequence includes a single-axis probe path and a multi-axis linkage probe path.

[0010] The active probe module controls the three-coordinate measuring equipment to execute the probe action sequence during the equipment idle period, collects the motion response data in the no-load state, identifies whether the abnormal measurement action occurs, and determines the fault source by comparing the motion characteristic differences of single-axis probe and multi-axis linkage probe for coupling abnormality.

[0011] The abnormal feature comparison module calculates the similarity between the motion characteristics in the active probe state and the motion characteristics of the same type of abnormality in the production measurement state, determines whether the abnormality is caused by equipment degradation or process environmental factors according to the similarity, and identifies the specific fault type through the abnormal fingerprint library matching. The abnormal fingerprint library stores multi-dimensional feature fingerprints of multiple known fault types.

[0012] As a preferred technical solution of the present application, the motion characteristics include the position change range of each motion axis, the speed extreme value, the acceleration peak and valley value, the speed direction change frequency, and the acceleration sign change frequency.

[0013] As a preferred technical solution of the present application, the normal measurement action feature library is established by the following method:

[0014] Collecting the timing data when the standard measurement program is executed during the equipment acceptance stage, and extracting the motion characteristics;

[0015] Calculating the mean and standard deviation of each feature component for multiple groups of motion characteristics of the same type of standard measurement action;

[0016] The mean value is used as the feature reference value, and three times the standard deviation is used as the normal fluctuation range.

[0017] As a preferred technical solution of the present application, the correlation of each motion axis motion characteristic includes:

[0018] When at least two motion axes simultaneously appear abnormal measurement action, the correlation coefficient between any two motion axes is calculated for each feature component;

[0019] When the correlation coefficient of at least one feature component exceeds the preset correlation threshold, it is determined as a coupling abnormality.

[0020] As a preferred technical solution of the present application, the generation of the targeted probe action sequence includes:

[0021] Extract the position change range, speed extreme value and acceleration peak and valley value of the high-frequency abnormal measurement action to determine the position parameter interval, speed parameter interval and acceleration parameter interval;

[0022] For single-axis abnormality, three different motion speed return motion paths are planned within the position parameter interval;

[0023] For coupled abnormality, multi-axis linkage detection paths are planned within the position parameter interval of each motion axis, and the motion speed set for each motion axis is proportionally distributed according to the midpoint value of its speed parameter interval.

[0024] As a preferred technical solution of the present application, the detection action generation module further comprises a detection coverage optimization function:

[0025] The number of abnormalities identified by active detection and the total number of abnormalities identified by production measurement in a preset period are counted to calculate the detection coverage;

[0026] When the detection coverage is lower than the preset threshold, the motion parameter interval of the abnormal measurement action that has not been detected is extracted and merged into the original motion parameter interval, and the detection action sequence is regenerated.

[0027] As a preferred technical solution of the present application, the determination of the fault source comprises:

[0028] The motion response data when the single-axis detection path and the multi-axis linkage detection path are executed are collected respectively, the motion characteristics are extracted and the deviation value is calculated;

[0029] When the motion characteristic deviation value when multi-axis linkage is greater than a preset multiple threshold value when single-axis detection is performed, it is determined that the abnormality is caused by degradation of the mechanical coupling structure;

[0030] When the motion characteristic deviation value when multi-axis linkage is less than a preset difference threshold value when single-axis detection is performed, it is determined that the abnormality is caused by the decline of the multi-axis synchronization performance of the control system.

[0031] As a preferred technical solution of the present application, the determination of the abnormality source according to the similarity comprises:

[0032] The motion characteristics in the active detection state are extracted as the unloaded motion characteristics;

[0033] The motion characteristics of the same type of abnormality in the production measurement state are extracted, and according to the weight of the workpiece, they are divided into light load group, medium load group and heavy load group, and the average value of the motion characteristics of each load group is calculated;

[0034] The difference degree between the unloaded motion characteristics and the average value of the motion characteristics of each load group is calculated, and the difference degree gradient between different load groups is calculated;

[0035] When the difference degree of the empty-load motion feature and all load groups is less than a preset threshold, it is determined that the type of abnormality is caused by device self-degradation;

[0036] When the difference degree of the empty-load motion feature and at least one load group is greater than a preset threshold, and the difference degree gradient between different load groups is greater than a preset gradient threshold, it is determined that the type of abnormality is caused by load-related process factors.

[0037] When the difference degree of the empty-load motion feature and at least one load group is greater than a preset threshold, and the difference degree gradient between different load groups is less than a preset gradient threshold, it is determined that the type of abnormality is caused by environmental factors.

[0038] As a preferred technical solution of the present application, the multi-dimensional feature fingerprint includes a time-domain feature vector, a frequency-domain feature vector, and a time sequence pattern code.

[0039] The time-domain feature vector is composed of feature components of the motion feature;

[0040] The frequency-domain feature vector is extracted by performing fast Fourier transform on acceleration time sequence data in the motion response data, and extracting power spectrum density peak frequencies and corresponding amplitudes in a preset frequency band range.

[0041] The time sequence pattern code is extracted by performing piecewise linear fitting on a position-time curve in the motion response data, and extracting slope signs and fitting residuals of each segment to form a code sequence.

[0042] As a preferred technical solution of the present application, the matching and identifying of a specific fault type through the abnormality fingerprint library includes:

[0043] For an abnormal measurement action identified in the active detection state, a time-domain feature vector, a frequency-domain feature vector, and a time sequence pattern code are extracted to form a to-be-matched fingerprint;

[0044] The comprehensive similarity between the to-be-matched fingerprint and the abnormality fingerprint of each known fault type in the abnormality fingerprint library is calculated, and the comprehensive similarity is weighted and fused by a time-domain feature vector similarity, a frequency-domain feature vector similarity, and a time sequence pattern similarity.

[0045] When the highest value of the comprehensive similarity is greater than a preset matching threshold, the corresponding fault type is output.

[0046] The present application has the following beneficial effects:

[0047] 1. This invention utilizes a high-frequency anomaly statistics-driven automatic generation mechanism for detection sequences, combined with closed-loop optimization of detection coverage, to achieve a shift from passively discovering anomalies during production to actively reproducing anomalies during idle periods. This targeted proactive detection strategy overcomes the limitations of traditional passive monitoring, enabling the early exposure of potential faults when equipment is idle, significantly improving early warning capabilities.

[0048] 2. This invention constructs a differentiated comparison system for single-axis detection and multi-axis linkage detection. Through a dual-parameter quantitative criterion of deviation ratio and difference value, it systematically solves the technical problem of tracing the source of coupling anomalies. This dual-mode detection mechanism can accurately distinguish between two fundamentally different fault causes: degradation of the mechanical coupling structure and decline in the synchronization performance of the control system, providing a clear basis for targeted maintenance.

[0049] 3. This invention establishes a hierarchical diagnostic architecture that combines no-load and load feature comparison with multi-dimensional fingerprint matching. By analyzing the gradient of differences between load groups, it systematically determines whether the anomaly originates from equipment degradation, process factors, or environmental factors. Combined with multi-dimensional feature fingerprints in the time domain, frequency domain, and time series, it achieves a complete diagnostic closed loop from anomaly tracing to specific fault type identification, significantly improving the comprehensiveness and accuracy of diagnosis. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a schematic diagram of the structure of a three-coordinate measurement motion recognition system based on semiconductor processing equipment according to the present invention;

[0052] Figure 2 This is a schematic diagram of the implementation process of a three-coordinate measurement motion recognition system based on semiconductor processing equipment according to the present invention. Detailed Implementation

[0053] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0054] Example 1: As Figure 1 As shown, a coordinate measuring and motion recognition system based on semiconductor processing equipment includes:

[0055] The motion data acquisition module collects the position, velocity, and acceleration time-series data of each motion axis of the coordinate measuring machine.

[0056] Specifically, the motion data acquisition module connects to the control system of the coordinate measuring machine via a data interface, and acquires data at a fixed sampling frequency. positions of three movement axes , velocities and accelerations Timing data. The collected data are stored by timestamp and axis identification, each record contains timestamp, axis identification, position value, velocity value, acceleration value. The context information of the measurement task is recorded at the same time, including measurement program identification, workpiece type, workpiece weight, for subsequent correlation analysis of abnormal source.

[0057] Abnormal measurement action recognition module, according to the timing data, extract the movement characteristics of the measurement action and compare with the normal measurement action feature library, when the deviation exceeds the preset threshold, identify as abnormal measurement action and output abnormal type and movement characteristics, when detecting multiple movement axes simultaneously appear abnormal, calculate the correlation of each movement axis movement characteristics, determine as single axis abnormal or coupling abnormal;

[0058] Further, the movement characteristics include the position change range of each movement axis, the speed extreme value, the acceleration peak and valley value, the speed direction change frequency, and the acceleration sign change frequency.

[0059] Further, the normal measurement action feature library is established by the following method:

[0060] Collect timing data when performing standard measurement program in equipment acceptance stage, extract movement characteristics;

[0061] Calculate the mean and standard deviation of each feature component for multiple sets of movement characteristics of the same type of standard measurement action;

[0062] Take the mean as the feature reference value, and take three times the standard deviation as the normal fluctuation range.

[0063] Specifically, the normal measurement action feature library is established in the equipment acceptance stage. In the acceptance stage, a standard measurement program is executed, which contains various typical measurement actions (point measurement, line scanning, surface scanning, etc.) of the equipment in normal production. Each measurement action is repeated m times (m is preferably not less than 30), and the movement data acquisition module collects the timing data of each movement axis and segments it according to the measurement action.

[0064] For each segment of timing data, extract the movement feature vector , where is the position change range, is the speed extreme value, and is the acceleration peak and valley value, is the speed direction change frequency, is the acceleration sign change frequency. For m sets of feature vectors of the same type of measurement action, calculate the mean and standard deviation .

[0065] Will As a feature benchmark value, As within the normal fluctuation range. The principle is because, under the assumption of normal distribution, The range covers 99.7% of normal data. The normal measurement motion feature library stores the feature components of each measurement motion type on each motion axis. and .

[0066] Furthermore, the calculation of the correlation of motion characteristics of each motion axis includes:

[0067] When at least two motion axes simultaneously exhibit abnormal measurement actions, calculate the correlation coefficient between any two motion axes for each feature component.

[0068] When the correlation coefficient of at least one feature component exceeds a preset correlation threshold, it is determined to be an abnormal coupling.

[0069] Specifically, after the equipment is put into production, the abnormal measurement action identification module monitors each measurement action in real time. It identifies the type of measurement action according to the measurement program instructions, extracts motion feature vectors from the time-series data, and compares them with the feature benchmarks of the corresponding type in the normal measurement action feature library.

[0070] For a certain characteristic component of a certain motion axis Calculate its relative deviation:

[0071]

[0072] in As the characteristic reference value, is the standard deviation. When an anomaly is detected on a motion axis, the motion axis identifier, the type of the out-of-limit feature component, and the complete motion feature vector are recorded. If multiple feature components exceed limits or multiple motion axes malfunction simultaneously during a measurement action, further investigation is needed to determine whether it is a single-axis anomaly or a coupling anomaly.

[0073] When at least two motion axes exhibit abnormalities simultaneously, the correlation of motion features is calculated to determine whether the abnormality is a single axis or a coupled abnormality. A time window with a length equal to the length of the k most recent measurement actions (k ranging from 20 to 100) is selected, and motion feature data of all abnormal measurement actions within the window are extracted.

[0074] For any two motion axes that exhibit anomalies simultaneously, calculate the Pearson correlation coefficient for each of the six feature components:

[0075] ;

[0076] wherein and are the characteristic component values of the two motion axes in the second measurement action, and are the mean values of the respective sequences.

[0077] When the absolute value of the correlation coefficient of at least one characteristic component exceeds a preset correlation threshold (preferably 0.7), it is determined to be a coupling anomaly. This indicates a strong correlation, indicating that the two motion axes exhibit significant synchronization or reverse synchronization changes in this characteristic component, indicating mechanical coupling or mutual influence of the control system. If the correlation coefficient of all characteristic components is less than the preset correlation threshold, it is determined to be a single-axis anomaly.

[0078] For a three-axis device, the correlation of three pairs of axes needs to be calculated. As long as any one pair of axes exhibits strong correlation in any one characteristic component, it is determined to be a coupling anomaly. The determination result is output together with the anomaly type and the motion characteristic vector.

[0079] The detection action generation module counts the frequency of occurrence of abnormal measurement actions, extracts the motion parameter interval of high-frequency abnormal measurement actions, and generates a targeted detection action sequence based on the motion parameter interval. The detection action sequence includes a single-axis detection path and a multi-axis linkage detection path.

[0080] Specifically, the detection action generation module receives the abnormal information output by the abnormal measurement action recognition module, including the abnormal type (single-axis anomaly or coupling anomaly), the involved motion axes, the characteristic component exceeding the limit, and the motion characteristic vector.

[0081] The system classifies and counts abnormal measurement actions according to the motion axis group to which the anomaly belongs, the type of characteristic component exceeding the limit, and the determination result of the anomaly. For example, "X-axis single-axis anomaly - speed extreme value exceeding the limit" is one category, and "X-Y-axis coupling anomaly - acceleration peak value exceeding the limit" is another category.

[0082] The number of occurrences of each category of anomaly is recorded within a preset statistical period (preferably 1 to 7 days). After the statistical period ends, the frequencies are sorted from high to low, and the top N (N ranges from 3 to 10) anomaly categories are selected as high-frequency anomalies. High-frequency anomalies represent the most prominent performance problems of the device and are the focus of active detection.

[0083] Further, the generation of a targeted detection action sequence includes:

[0084] Extract the range of position changes, velocity extremes, and acceleration peaks and valleys of high-frequency abnormal measurement actions to determine the intervals of position parameters, velocity parameters, and acceleration parameters.

[0085] For single-axis anomalies, three reciprocating motion paths with different motion speeds are planned within the range of the position parameters;

[0086] To address coupling anomalies, a multi-axis linkage detection path is planned within the position parameter range of each motion axis, and the motion speed set for each motion axis is proportionally allocated according to the midpoint value of its speed parameter range.

[0087] Specifically, for each high-frequency anomaly category, motion feature data of all abnormal measurement actions under that category are extracted, and the distribution characteristics of their motion parameters are analyzed.

[0088] For each motion axis involved, extract its position variation range. Extreme speed , acceleration peak and valley values ​​( and A sequence of numerical values. Assume this anomaly category contains... The corresponding abnormal measurement action will obtain The range of values ​​for each position change A velocity extreme value, The maximum acceleration and The minimum acceleration value.

[0089] Determine the parameter intervals for each of these numerical sequences: position parameter intervals Take the minimum and maximum values ​​of all position variation ranges under this anomaly category; velocity parameter range. ,in The maximum value of all velocity extrema under this anomaly category; acceleration parameter range. ,in It is the minimum of all minimum acceleration values. It is the maximum value among all acceleration values.

[0090] These parameter ranges characterize the distribution range of this anomaly category in the motion space and serve as the basis for generating targeted detection actions. This is based on the type of high-frequency anomaly (uniaxial anomaly or coupled anomaly). Based on the range of motion parameters, a corresponding detection action sequence is generated.

[0091] For categories classified as single-axis anomalies, the position parameter range of that motion axis... The internal plan includes three round-trip paths with different movement speeds.

[0092] The first path uses low-speed motion, with a set speed of [speed value missing]. ; The second path adopts medium-speed motion, and the set speed is ; The third path adopts high-speed motion, and the set speed is . The three speeds respectively correspond to the low, medium, and high sections of the abnormal occurrence speed range to comprehensively cover possible abnormal working conditions.

[0093] The motion process of each path is as follows: starting from the starting point of the position parameter interval accelerating to the set speed, moving at a constant speed to the end point , decelerating to a stop, and then moving in the reverse direction back to the starting point. One round trip constitutes a complete detection action.

[0094] The acceleration parameter is determined according to the acceleration parameter interval. It is preferably taken as 80% of as the acceleration value during the acceleration process, and 80% of

[0095]

[0096] For the category determined as a coupling anomaly, it involves the coordinated motion of multiple motion axes. Plan multi-axis linkage detection paths within the position parameter intervals of each motion axis. The multi-axis linkage detection path adopts a synchronous motion strategy, that is, multiple motion axes start simultaneously and reach their respective target positions simultaneously. To achieve synchronization, it is necessary to allocate motion speeds to each motion axis.

[0097] The speed allocation method is as follows: First, determine the midpoint value of the speed parameter interval of each motion axis. For the motion axis , its speed midpoint value is . Then calculate the ratio of the position travel of each motion axis to the speed midpoint value:

[0098] <( ;

[0099] Select the maximum value among all motion axes as the unified motion time for the linkage motion. The actual motion speed of each motion axis is determined according to the ratio of its position travel to the unified motion time: <(

[0100] ;

[0101] Through this proportional allocation, the motion speeds of each motion axis are maintained near the midpoint value of their speed parameter intervals, and they can synchronously complete their respective position travels, simulating the multi-axis coordinated motion state when a coupling anomaly occurs.

[0102] The multi-axis linkage detection path also adopts a round-trip motion form. Each motion axis synchronously moves from the starting point of its respective position parameter interval to the end point and then synchronously returns to the starting point.

[0103] For each high-frequency abnormality category, the generated detection actions constitute a detection action sequence. Single-axis abnormality categories generate sequences containing three single-axis detection paths, and coupling abnormality categories generate sequences containing one multi-axis linkage detection path (which can be repeated multiple times to enhance detection confidence .

[0104] All high-frequency abnormality category corresponding detection action sequences are summarized to form a complete active detection scheme for the active detection module to execute during the device idle period.

[0105] Further, the detection action generation module further includes a detection coverage optimization function:

[0106] Statistical active detection identified abnormalities in a preset period and the total number of abnormal production measurements identified, calculate the detection coverage;

[0107] When the detection coverage is lower than the preset threshold, extract the motion parameter interval of the abnormal measurement action that has not been detected, merge it into the original motion parameter interval, and regenerate the detection action sequence.

[0108] Specifically, the detection action generation module further includes a detection coverage optimization function for evaluating the effectiveness of the detection action sequence and dynamically adjusting the detection scheme according to the evaluation results.

[0109] The detection coverage is defined as the ratio of the number of abnormalities identified by the active detection module when executing the detection action sequence in a preset evaluation period to the total number of abnormalities identified in the production measurement process in the evaluation period. The calculation formula is:

[0110] ;

[0111] Where is the number of abnormalities identified by active detection, is the total number of abnormalities identified by production measurement. The evaluation period is preferably 7 to 30 days.

[0112] When the detection coverage is lower than the preset threshold (preferably 60% , it means that the current detection action sequence has not fully covered the abnormal state of the device, and the detection scheme needs to be optimized.

[0113] The optimization method is: statistical active detection has not identified the abnormal measurement action in the evaluation period, that is, the abnormality identified in the production measurement but not reproduced in the active detection. Group these undetected abnormalities by category, and extract the motion parameter interval of each category.

[0114] The motion parameter interval of the undetected abnormality is merged with the motion parameter interval of the current detection action sequence corresponding category. The merging method is: the position parameter interval takes the union of the two, the speed parameter interval takes the maximum of the upper limit, and the acceleration parameter interval takes the maximum and minimum of the upper and lower limits.

[0115] Based on the merged motion parameter interval, the detection action sequence is regenerated. The new detection action sequence expands the motion space range of the detection and can cover the previously missed abnormal working conditions.

[0116] The detection coverage optimization function adopts an iterative improvement mechanism, and the evaluation and optimization process is automatically triggered once at the end of each evaluation period, ensuring that the detection scheme continuously adapts to changes in device performance.

[0117] The active detection module controls the three-coordinate measuring equipment to execute the detection action sequence during the idle period of the equipment, collects the motion response data under the no-load state, identifies whether abnormal measurement actions occur, and determines the fault source by comparing the motion characteristic differences between single-axis detection and multi-axis linkage detection for coupling abnormalities.

[0118] Specifically, the active detection module identifies the idle period by monitoring the task queue state of the equipment control system. When the measurement task queue is empty and no new task arrives within a predetermined time window (preferably 5 to 15 minutes), it is determined that the equipment is idle, and a detection task instruction is sent to the control system. The detection task has a lower priority than normal measurement tasks and is interrupted immediately when a production task arrives.

[0119] When executing the detection action sequence, the equipment is in a no-load state (the measuring table has no workpiece), excluding the influence of workpiece load, and the detection result reflects the motion performance of the equipment itself. The detection execution frequency is dynamically adjusted according to the equipment usage, preferably 1 to 3 times per day.

[0120] The active detection module controls each motion axis to execute the planned motion path according to the detection action sequence provided by the detection action generation module. For single-axis detection paths, the target motion axis is driven separately to execute three different speed back-and-forth movements; for multi-axis linkage detection paths, multiple motion axes are synchronously driven to execute coordinated movements.

[0121] During the execution of the detection action, the motion data acquisition module continuously collects the position, speed, and acceleration time series data of each motion axis to form the motion response data under the no-load state. The motion response data and the time series data under the production measurement state use the same sampling frequency and data format.

[0122] After the motion response data is collected, the active detection module calls the feature extraction and comparison function of the abnormal measurement action recognition module to determine whether an abnormality occurs during the execution of the detection action. The specific process is to extract the motion feature vector of each motion axis from the motion response data, compare it with the feature reference of the corresponding type of action in the normal measurement action feature library, and calculate the relative deviation. When the relative deviation exceeds the threshold value, it is determined that the detection action has an abnormality.

[0123] The type of abnormality identified by the detection, the motion axis involved, the characteristic component that exceeds the limit, and the complete motion feature vector are recorded as abnormal feature data in the empty load state.

[0124] Since the detection action is specifically designed for high-frequency abnormalities, if no abnormality is identified during the execution of the detection action, it means that the abnormality may be related to the workpiece load or the production environment factors, rather than the equipment itself degradation. This information provides a basis for subsequent abnormality source determination.

[0125] Further, the determining the fault source includes:

[0126] Collecting motion response data when the single-axis detection path and the multi-axis linkage detection path are executed, extracting motion features and calculating deviation values;

[0127] When the motion feature deviation value of multi-axis linkage is greater than the preset multiple threshold value of single-axis detection, it is determined that the abnormality is caused by the degradation of the mechanical coupling structure;

[0128] When the difference between the motion feature deviation value of multi-axis linkage and the single-axis detection is less than the preset difference threshold value, it is determined that the abnormality is caused by the decline of the multi-axis synchronization performance of the control system.

[0129] Specifically, for the category of coupling abnormality, the active detection module needs to determine whether the fault source is the degradation of the mechanical coupling structure or the decline of the multi-axis synchronization performance of the control system by comparing the motion feature difference between single-axis detection and multi-axis linkage detection.

[0130] The determination process first executes the single-axis detection path for each motion axis involved in the coupling abnormality. During single-axis detection, the target motion axis performs reciprocating motion within its position parameter interval, and other motion axes remain stationary. The motion data collection module collects the motion response data during single-axis detection, extracts the motion feature vector of the motion axis, and calculates the deviation value of each characteristic component relative to the normal feature reference, denoted as single-axis deviation value This single-axis detection process is performed for each motion axis involved in the coupling abnormality to obtain the single-axis deviation value of each axis.

[0131] After the single-axis detection is completed, a multi-axis linkage detection path is performed. During the multi-axis linkage detection, multiple motion axes involved in the coupling anomaly move in coordination according to a pre-planned synchronous motion strategy. The motion data acquisition module acquires motion response data during the multi-axis linkage detection, extracts the motion feature vectors of each motion axis, and calculates the deviation values of each feature component relative to the normal feature reference, denoted as linkage deviation values .

[0132] After obtaining the single-axis deviation values and the linkage deviation values, the source of the fault is determined by analyzing the relationship between the two. First, the ratio of the linkage deviation values to the single-axis deviation values is calculated . The ratio reflects the amplification multiple of the abnormality degree when multiple axes move in linkage compared to when the single-axis moves independently. At the same time, the difference value between the two is calculated , which reflects the additional deviation amount introduced by the linkage effect.

[0133] When the ratio is greater than a preset multiple threshold, it is determined that the abnormality originates from the degradation of the mechanical coupling structure. The preset multiple threshold is preferably 2.0 to 3.0. A ratio significantly greater than 1 indicates that the deviation when multiple axes move in linkage is much greater than when the single-axis moves independently, indicating that additional mutual interference is generated when multiple motion axes move in coordination. Such interference is usually caused by the performance degradation of the mechanical coupling structure, such as the decrease in the stiffness of the cross rail causing additional force transmission to another axis when one axis moves, or the increase in the gap of the connecting member causing cumulative errors when multiple axes move in linkage, or the deformation of the support structure being more obvious under multiple-axis load. These mechanical problems will be amplified when multiple axes move in coordination, resulting in a significant increase in the degree of abnormality.

[0134] When the ratio is close to 1.0 and the difference value is less than a preset difference threshold, it is determined that the abnormality originates from the decrease in the multi-axis synchronization performance of the control system. The preset difference threshold is preferably 20% to 30% of the single-axis deviation value. A small difference indicates that the deviation when multiple axes move in linkage is basically consistent with the deviation when the single-axis moves independently, indicating that the performance of the mechanical coupling structure itself is good and no significant additional mechanical interference is introduced by the multi-axis linkage. At this time, the anomaly is mainly caused by factors such as insufficient accuracy of the multi-axis coordination algorithm of the control system, failure of the synchronization error compensation function, or increase in the communication delay between axes, etc. in the software or electrical level. Such problems in the control system do not have essential differences between single-axis and multi-axis motion modes, so they show similar deviation levels in the two detection modes.

[0135] The specific determination criteria are:

[0136] When When the preset difference threshold is preferably 20% to 30% of the single-axis deviation value, it is determined that the abnormality is caused by the decline in the multi-axis synchronization performance of the control system. At this time, the single-axis and multi-axis deviations are basically consistent, indicating that the mechanical structure is normal but the control system has a problem.

[0137] When and When the preset difference threshold is preferably 20% to 30% of the single-axis deviation value, it is determined that the abnormality is caused by the decline in the multi-axis synchronization performance of the control system. At this time, the single-axis and multi-axis deviations are basically consistent, indicating that the mechanical structure is normal but the control system has a problem.

[0138] When neither of the above two conditions is met (for example, when or But Δ is also large), it is marked as a composite failure, which may involve the joint action of the mechanical and control systems, and further manual analysis is required.

[0139] The failure source determination result is output together with the abnormality type detected and identified, the single-axis deviation value, the linkage deviation value, and the complete motion feature data, for further analysis by the abnormality feature comparison module. By clearly distinguishing between mechanical failure and control system failure, maintenance personnel can take more targeted measures. For mechanical coupling structure degradation, mechanical parts can be repaired or replaced. For control system synchronization performance decline, control parameter optimization or software upgrade can be performed to avoid time and resource waste caused by blind maintenance.

[0140] The abnormality feature comparison module calculates the similarity between the motion features in the active detection state and the motion features of the same type of abnormality in the production measurement state, determines whether the abnormality is caused by equipment degradation or process environmental factors, and identifies the specific failure type through matching and identification of the abnormality fingerprint library, which stores multi-dimensional feature fingerprints of multiple known failure types.

[0141] Specifically, the abnormality feature comparison module receives the abnormality feature data in the empty state output by the active detection module, and the abnormality feature data in the production measurement state accumulated by the abnormality measurement action recognition module. For the same type of abnormality, the motion features in the empty state and the production measurement state are extracted and compared.

[0142] First, the motion feature vector of the abnormal measurement action identified in the active detection state is extracted as the empty motion feature. The empty motion feature reflects the motion performance state of the equipment under the condition of no workpiece load.

[0143] Then, the motion characteristics of the same type of anomalies under production measurement conditions are extracted. Since workpieces of varying weights are carried during production measurement, the anomaly data needs to be grouped according to workpiece weight. Based on workpiece weight, the anomaly measurement actions are divided into three load groups: light load, medium load, and heavy load. The light load group corresponds to measurement tasks where the workpiece weight is less than 30% of the equipment's rated load; the medium load group corresponds to measurement tasks where the workpiece weight is between 30% and 70% of the rated load; and the heavy load group corresponds to measurement tasks where the workpiece weight is greater than 70%.

[0144] For abnormal measurement actions within each load group, extract their motion feature vectors, calculate the average value of each feature component, and obtain the average motion feature value for that load group. For example, for the light load group, if it includes... The average value of the characteristic components of the position change range for the second abnormal measurement action is: Other feature components are calculated similarly. The average motion feature vectors for each of the three load groups are obtained.

[0145] Next, the difference between the no-load motion characteristics and the average motion characteristics of each load group is calculated. To eliminate the influence of characteristics with different dimensions, each characteristic component is normalized before calculating the difference by dividing each characteristic component value by its standard deviation in the normal characteristic library. This ensures that all feature components are of the same order of magnitude. The difference is measured using Euclidean distance for unloaded motion feature vectors. and the average value vector of motion characteristics of a certain load group The degree of difference is calculated as follows:

[0146] ;

[0147] in and The first under no-load and loaded conditions respectively The values ​​of each feature component For the first The weight coefficients of each feature component are normalized to have equal values, i.e. .

[0148] Calculate the difference between the unloaded motion characteristics and the average motion characteristics of the light-load, medium-load, and heavy-load groups, respectively, and denote it as . .

[0149] Simultaneously, the difference gradient between different load groups is calculated to determine the degree of influence of load changes on anomalous features. The difference gradient measures the sensitivity of anomalous features to load changes. The calculation method is as follows:

[0150] First, calculate the Euclidean distance between the average vectors of motion characteristics of adjacent load groups. The calculation formula is the same as that for the no-load-load difference:

[0151] ;

[0152] wherein and are the average values of the first characteristic components of the light load group and the medium load group, respectively. Similarly, the average values of the second characteristic components are calculated.

[0153] The difference degree gradient is the average value of the two:

[0154] ;

[0155] The greater the gradient value, the more obvious the abnormal characteristics change with the load.

[0156] Further, the determining the abnormal source according to the similarity comprises:

[0157] extracting the motion characteristics in the active detection state as the unloaded motion characteristics;

[0158] extracting the motion characteristics of the same type of abnormality in the production measurement state, dividing the workpiece weight into a light load group, a medium load group, and a heavy load group, and calculating the average values of the motion characteristics of each load group;

[0159] calculating the difference degree between the unloaded motion characteristics and the average values of the motion characteristics of each load group, and calculating the difference degree gradient between different load groups;

[0160] when the difference degree between the unloaded motion characteristics and all load groups is less than a preset threshold value, it is determined that the type of abnormality is caused by equipment degradation itself;

[0161] when the difference degree between the unloaded motion characteristics and at least one load group is greater than a preset threshold value, and the difference degree gradient between different load groups is greater than a preset gradient threshold value, it is determined that the type of abnormality is caused by load-related process factors;

[0162] when the difference degree between the unloaded motion characteristics and at least one load group is greater than a preset threshold value, and the difference degree gradient between different load groups is less than a preset gradient threshold value, it is determined that the type of abnormality is caused by environmental factors.

[0163] Specifically, the source of the type of anomaly is determined according to the difference degree of the no-load motion feature and each load group, and the difference degree gradient between the load groups. The determination principle is that the motion feature of the device itself degradation type anomaly (such as guide rail wear, transmission system aging) is not affected by the load, so the no-load and each load state are consistent; the anomaly caused by the load related process factor (such as workpiece clamping eccentricity, structural deformation under heavy load) changes systematically with the increase of the load, which is manifested as a large difference degree gradient between the load groups; the environmental factor (such as temperature fluctuation during production period, workshop vibration) exists universally in the production state but is irrelevant to the load size, which is manifested as a significant difference between the no-load and load states but a small difference degree gradient between the load groups. Based on the above principle, the determination criterion is:

[0164] When the difference degree of the no-load motion feature and all load groups is less than a preset threshold value , that is , and , and , it is determined that the type of anomaly is caused by the degradation of the device itself. The preset threshold value is determined according to the dimension and normal fluctuation range of the feature vector, and is preferably 0.5 times to 1.0 times the sum of the standard deviations of each component of the feature vector. In this case, the abnormal features of the no-load state and various load states all show high consistency, which indicates that the device shows the same abnormal behavior whether it is loaded with a workpiece or not. This indicates that the anomaly is caused by internal degradation factors such as mechanical wear, guide rail precision decline, transmission system aging, and is irrelevant to external process conditions.

[0165] When the difference degree of the no-load motion feature and at least one load group is greater than the preset threshold value, and the difference degree gradient between the load groups is greater than a preset gradient threshold value , it is determined that the type of anomaly is caused by the load related process factor. The preset gradient threshold value is preferably 0.3 times to 0.5 times the preset threshold value . In this case, the abnormal features of the no-load state and the load state are significantly different, and the abnormal features change systematically with the increase of the load. This indicates that the anomaly is strongly related to the workpiece load, and may be caused by uneven load distribution due to improper workpiece clamping, dynamic balance problem caused by workpiece gravity center offset, or elastic deformation of the mechanical structure under heavy load exceeds the compensation range, etc. load related process factors.

[0166] When the difference degree of the no-load motion feature and at least one load group is greater than the preset threshold value, but the difference degree gradient between the load groups is less than the preset gradient threshold value When the abnormality is detected in the idle state, it is determined that the abnormality is caused by the environment factor. In this case, the abnormal characteristics of the idle state and the load state are different, but the abnormal characteristics of different load groups are relatively close, indicating that the occurrence of the abnormality is related to whether the workpiece is loaded, but has no obvious correlation with the load size. This usually reflects the influence of production environment factors, such as the fluctuation of the ambient temperature during the production period causing thermal deformation, the vibration interference of the production workshop interfering with the equipment, or the airflow disturbance affecting the measurement accuracy. These environmental factors are different from the production period, but relatively stable during the production period, so they are similar under different loads.

[0167] The abnormality source determination result provides guidance for subsequent maintenance decisions. For device self-degradation, device maintenance and component replacement need to be arranged; for load-related process factors, workpiece clamping schemes and load control strategies need to be optimized; for environmental factors, production environment conditions need to be improved or the environmental adaptability of the equipment needs to be enhanced.

[0168] Further, the multi-dimensional feature fingerprint includes a time domain feature vector, a frequency domain feature vector, and a time sequence pattern code;

[0169] The time domain feature vector is composed of the feature components of the motion feature;

[0170] The frequency domain feature vector extracts the power spectral density peak frequency and the corresponding amplitude in the preset frequency range by performing fast Fourier transform on the acceleration time sequence data in the motion response data;

[0171] The time sequence pattern code extracts the slope sign and fitting residual of each segment by performing piecewise linear fitting on the position-time curve in the motion response data to form a code sequence.

[0172] Specifically, to achieve accurate identification of specific fault types, the system establishes an abnormality fingerprint library to store multi-dimensional feature fingerprints of multiple known fault types. The multi-dimensional feature fingerprint is composed of a time domain feature vector, a frequency domain feature vector, and a time sequence pattern code, which comprehensively characterizes the feature performance of the abnormality.

[0173] The time domain feature vector is directly composed of the six feature components of the aforementioned motion feature, that is, The time domain feature vector reflects the statistical characteristics and motion pattern features of the abnormality in the time domain.

[0174] The frequency domain feature vector is obtained by frequency domain analysis on the acceleration time sequence data in the motion response data. The acceleration time sequence data is subjected to fast Fourier transform (FFT , and the frequency domain power spectral density function ​The peak frequency of the power spectral density and its corresponding amplitude are searched within a preset frequency band. The preset frequency band is determined based on the mechanical characteristics of the equipment, typically covering the main resonant frequency range, preferably from 1Hz to 500Hz. The peak frequency of the power spectral density is extracted. Peak frequency and its corresponding amplitude Composed of frequency domain feature vectors Number of peaks Preferably, there are 3 to 5. Frequency domain eigenvectors can reveal abnormal vibration characteristics and frequency components, and have unique advantages in identifying faults such as resonance, wear, and loosening.

[0175] Temporal morphological encoding is obtained by piecewise linear fitting of the position-time curve. The piecewise method involves identifying the moment when the direction of motion changes, i.e., the velocity. The points where the sign changes (from positive to negative or from negative to positive) are used to segment the data. In this way, each linear segment corresponds to a continuous unidirectional motion process.

[0176] For each linear segment, perform least-squares line fitting, and then determine the slope of the fitted line. The sign of the slope determines the direction of motion; a positive slope is denoted as +, and a negative slope as -. Simultaneously, the fitting residual for this segment is calculated, which is the root mean square error between the actual position data and the fitted line.

[0177] ;

[0178] in For actual location, To fit the corresponding positions on the straight line This represents the number of data points in that segment. The slope sign of each segment is sequentially combined with the fitting residuals to form a time-series morphological code, for example... This indicates that the trajectory of motion has gone through four stages: forward-backward-forward-backward. The first term of the tuple is the slope sign, and the second term is the normalized fitting residual.

[0179] Temporal morphological similarity is calculated by comparing two encoded sequences. For the sequence to be matched and the fingerprint database sequence, the shorter sequence length is used as the benchmark, and the segments are aligned for comparison. For each corresponding segment, the following calculation is performed:

[0180] Slope sign consistency: 1 for the same sign, 0 for different signs;

[0181] Residual similarity: ;

[0182] Temporal morphological similarity is:

[0183] ;

[0184] wherein is preferably 0.6, indicating that the slope sign weight is higher.

[0185] In the abnormal fingerprint library, each known fault type corresponds to a set of multi-dimensional feature fingerprints . The known fault types include but are not limited to: guide rail wear, ball screw gap, motor encoder failure, controller parameter drift, mechanical resonance, bearing wear, drive belt slack, etc. The feature fingerprints of these fault types are obtained through data analysis of historical fault cases or extracted through fault simulation tests.

[0186] Further, the matching and identifying the specific fault type through the abnormal fingerprint library comprises:

[0187] For the abnormal measurement action identified in the active detection state, its time domain feature vector, frequency domain feature vector, and time sequence pattern code are extracted to form a matching fingerprint;

[0188] The comprehensive similarity between the matching fingerprint and the abnormal fingerprint of each known fault type in the abnormal fingerprint library is calculated, and the comprehensive similarity is weighted and fused by the time domain feature vector similarity, the frequency domain feature vector similarity, and the time sequence pattern similarity.

[0189] When the highest value of the comprehensive similarity is greater than a preset matching threshold, the corresponding fault type is output.

[0190] Specifically, for the abnormal measurement action identified in the active detection state, its multi-dimensional feature fingerprint is extracted for fault type identification.

[0191] First, the time domain feature vector of the abnormal measurement action is extracted, i.e., a vector composed of six motion feature components. Then, the acceleration time sequence data of the abnormal measurement action is subjected to fast Fourier transform to extract the power spectral density peak frequency and amplitude, forming a frequency domain feature vector. Next, the position-time curve is subjected to piecewise linear fitting to extract the slope sign and fitting residual of each segment, forming a time sequence pattern code. Combining these three parts, a matching fingerprint is formed .

[0192] Then, the matching fingerprint is matched with the abnormal fingerprint of each known fault type in the abnormal fingerprint library, and the comprehensive similarity is calculated. The comprehensive similarity is weighted and fused by the time domain feature vector similarity, the frequency domain feature vector similarity, and the time sequence pattern similarity.

[0193] The time domain feature vector similarity is calculated using the cosine similarity. For the time domain feature vector to be matched and the time domain feature vector of a fault type in the abnormal fingerprint library, the time domain similarity is:

[0194] ;

[0195] The frequency domain feature vector similarity calculation method is: extracting the peak frequency and amplitude sequence of the to-be-matched frequency domain vector , and the peak frequency and amplitude sequence of the fingerprint library frequency domain vector . For each peak value of the to-be-matched vector , search for the peak value with the smallest frequency deviation in the fingerprint library vector . The peak value matching adopts a combination of a relative threshold and an absolute threshold: if (the relative frequency deviation is less than 10%) or (the absolute frequency deviation is less than 5 Hz), if one of them is met, it is considered that the peak value matching is successful, and the amplitude similarity is calculated; if the peak value matching is unsuccessful, the amplitude similarity is 0. In this way, reasonable matching can be achieved in both the low frequency band (such as around 10 Hz) and the high frequency band (such as around 200 Hz). The frequency domain feature vector similarity is the average value of all peak value similarities. If the number of peak values of the two frequency domain vectors is different, the average value is calculated based on the one with more peak values, and the peak value similarity that is not matched is recorded as 0.

[0196] The comprehensive similarity is obtained by weighted fusion of the three similarities:

[0197] ;

[0198] wherein is a weight coefficient, and satisfies . The weight coefficient is determined according to the feature saliency of different fault types, and is preferably .

[0199] The comprehensive similarity of the to-be-matched fingerprint and all known fault types in the abnormal fingerprint library is calculated, and the fault type with the highest comprehensive similarity is selected as the candidate matching result. When the highest comprehensive similarity value is greater than the preset matching threshold (preferably 0.75 to 0.85 , it is determined that the abnormality corresponds to the fault type, and the fault type name, similarity value and typical maintenance suggestion of the fault type are output.

[0200] If the highest comprehensive similarity value is less than the preset matching threshold, it indicates that the abnormality does not belong to the known fault type, and it may be a new type of fault or a composite fault. At this time, the system marks the abnormality as a case to be analyzed, records its multi-dimensional feature fingerprint, and provides subsequent expert analysis and fingerprint library expansion for use.

[0201] Through the multi-dimensional feature matching of the abnormal fingerprint library, the system can realize a closed loop from abnormal detection to specific fault diagnosis, provide clear fault positioning and repair guidance for equipment maintenance, and significantly improve the maintenance efficiency and equipment availability.

[0202] Embodiment two:

[0203] A three-coordinate measuring device is configured in a wafer detection production line of a semiconductor manufacturing enterprise for key size detection after wafer processing. The measuring device is an important part of the semiconductor processing equipment, and the measuring accuracy directly affects the process control and product yield. The device appears the phenomenon of increased measurement data fluctuation after six months of continuous operation, resulting in misjudgment of some qualified wafers, but the conventional maintenance inspection fails to locate the fault cause. Therefore, a three-coordinate measurement action recognition system based on the semiconductor processing equipment is adopted, and the overall implementation process is as shown in Figure 2 .

[0204] After the system is deployed, first, the normal measurement action feature library is established on the basis of the device acceptance standard procedure. In the production operation process, the abnormal measurement action recognition module continuously monitors the motion characteristics of each motion shaft, and identifies Y-axis speed extreme value overrun and X-Y axis coupling abnormality as high-frequency abnormal types within two weeks.

[0205] The detection action generation module generates a detection sequence according to the motion parameter interval of the two types of high-frequency abnormalities, including a single-axis detection path of three speeds of the Y-axis and an X-Y axis linkage detection path. The active detection module executes the detection sequence during the idle period of the device at night, and successfully reproduces the Y-axis speed abnormality in the no-load state, but the X-Y axis coupling abnormality does not appear in the no-load detection.

[0206] For the single-axis abnormality of the Y-axis, through the comparison and analysis of single-axis detection and multi-axis linkage detection, it is found that the speed extreme value deviation of the Y-axis in multi-axis linkage is 2.8 times that in single-axis detection, and it is determined that the fault source is the degradation of the mechanical coupling structure. The maintenance personnel find that the fastening bolts at the connection between the Y-axis guide rail and the X-axis are loose, which causes additional vibration when multi-axis linkage is performed.

[0207] For the X-Y axis coupling abnormality, the abnormal feature comparison module analyzes and finds that (assuming that the rated load of the device is 2kg), the motion feature difference degree between the no-load state and the light load group 400g wafer carrier, about 20% of the rated load) is 0.15, and the motion feature difference degree between the medium load group 1000g wafer + carrier, about 50% of the rated load) is 0.52, and the motion feature difference degree between the heavy load group 1600g multi-wafer + carrier, about 80% of the rated load) is 0.78, and the difference degree gradient between the load groups is 0.42, and it is determined that the abnormality is caused by load-related process factors. Further inspection finds that the clamping method of the heavy load wafer carrier causes the center of gravity to deviate, causing dynamic load imbalance.

[0208] Through multi-dimensional feature fingerprint matching, the system identifies the Y-axis abnormality as the "loose rail connection" fault type and the X-Y axis coupling abnormality as the "improper carrier clamping" problem, thereby providing clear guidance for maintenance.

[0209] The equipment performance before and after the implementation of the system of the application is compared in the following table:

[0210] Table 1 Comparison of equipment performance

[0211]

[0212] Through the application of the system of the application, the enterprise has realized the transformation from passive maintenance after the occurrence of a fault to active early warning in the fault germination period, and the reliability and production stability of the semiconductor processing equipment have been significantly improved.

[0213] Finally, it should be noted that: the above only for the preferred embodiments of the application, and not for the purpose of limiting the application, although the application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, it still can be modified to the technical solution recorded in the foregoing embodiments, or equivalent replacement of some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A coordinate measuring and motion recognition system based on semiconductor processing equipment, characterized in that, include: The motion data acquisition module collects the position, velocity, and acceleration time-series data of each motion axis of the coordinate measuring machine. The abnormal measurement action recognition module extracts motion features of the measurement action based on the time series data and compares them with the normal measurement action feature library. When the deviation exceeds a preset threshold, it is identified as an abnormal measurement action and the abnormality type and motion features are output. When multiple motion axes are detected to be abnormal at the same time, the correlation of motion features of each motion axis is calculated and determined to be a single-axis abnormality or a coupling abnormality. The detection action generation module counts the frequency of abnormal measurement actions, extracts the motion parameter range of high-frequency abnormal measurement actions, and generates a targeted detection action sequence based on the motion parameter range. The detection action sequence includes a single-axis detection path and a multi-axis linkage detection path. The active detection module controls the coordinate measuring machine to execute the detection action sequence during the idle period of the equipment, collects motion response data under no-load conditions, identifies whether abnormal measurement actions occur, and determines the source of the fault by comparing the motion characteristics of single-axis detection and multi-axis linkage detection for coupling anomalies. The abnormal feature comparison module calculates the similarity between motion features under active detection and motion features of the same type of abnormality under production measurement. Based on the similarity, it determines whether the abnormality originates from equipment degradation or process environment factors, and identifies the specific fault type by matching with an abnormal fingerprint database. The abnormal fingerprint database stores multi-dimensional feature fingerprints of various known fault types.

2. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The motion characteristics include the range of position changes of each motion axis, the extreme values ​​of velocity, the peak and valley values ​​of acceleration, the number of times the velocity direction changes, and the number of times the acceleration sign changes.

3. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The normal measurement action feature library is established in the following way: Collect time-series data during the standard measurement procedure execution phase of equipment acceptance and extract motion features; For multiple sets of motion characteristics of the same type of standard measurement action, calculate the mean and standard deviation of each characteristic component; The mean is used as the characteristic benchmark value, and three times the standard deviation is used as the normal fluctuation range.

4. The coordinate measuring and motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The calculation of the correlation of motion characteristics of each motion axis includes: When at least two motion axes simultaneously exhibit abnormal measurement actions, calculate the correlation coefficient between any two motion axes for each feature component. When the correlation coefficient of at least one feature component exceeds a preset correlation threshold, it is determined to be an abnormal coupling.

5. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The generation of the targeted detection action sequence includes: Extract the range of position changes, velocity extremes, and acceleration peaks and valleys of high-frequency abnormal measurement actions to determine the intervals of position parameters, velocity parameters, and acceleration parameters. For single-axis anomalies, three reciprocating motion paths with different motion speeds are planned within the range of the position parameters; To address coupling anomalies, a multi-axis linkage detection path is planned within the position parameter range of each motion axis, and the motion speed set for each motion axis is proportionally allocated according to the midpoint value of its speed parameter range.

6. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The detection action generation module also includes a detection coverage optimization function: The number of anomalies actively detected and identified within a preset period is compared with the total number of anomalies identified by production measurement, and the detection coverage rate is calculated. When the detection coverage is lower than a preset threshold, the motion parameter range of the abnormal measurement actions that were not detected is extracted, merged into the original motion parameter range, and the detection action sequence is regenerated.

7. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The sources of the faults are determined as follows: Motion response data were collected during the execution of single-axis and multi-axis linkage detection paths, respectively, and motion features were extracted and deviation values ​​were calculated. When the motion characteristic deviation value during multi-axis linkage is greater than the preset multiple threshold during single-axis detection, the abnormality is determined to originate from the degradation of the mechanical coupling structure. When the difference between the motion characteristic deviation value during multi-axis linkage and that during single-axis detection is less than a preset difference threshold, the abnormality is determined to be due to a decrease in the multi-axis synchronization performance of the control system.

8. The coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The method of determining the source of anomalies based on similarity includes: Motion features under active detection conditions are extracted as no-load motion features; Extract the motion characteristics of the same type of anomaly under production measurement conditions, and divide them into light load group, medium load group and heavy load group according to the workpiece weight. Calculate the average value of the motion characteristics of each load group. Calculate the degree of difference between the no-load motion characteristics and the average value of the motion characteristics of each load group, and calculate the degree of difference gradient between different load groups; When the difference between the no-load motion characteristics and all load groups is less than a preset threshold, it is determined that this type of anomaly originates from the equipment itself. When the difference between the no-load motion characteristics and at least one load group is greater than a preset threshold, and the difference gradient between different load groups is greater than a preset gradient threshold, it is determined that this type of anomaly originates from load-related process factors. When the difference between the no-load motion characteristics and at least one load group is greater than a preset threshold, and the difference gradient between different load groups is less than a preset gradient threshold, it is determined that this type of anomaly originates from environmental factors.

9. A coordinate measuring motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The multidimensional feature fingerprint includes time-domain feature vectors, frequency-domain feature vectors, and time-series morphological encoding; The temporal feature vector is composed of the feature components of the motion feature; The frequency domain feature vector extracts the peak frequency and corresponding amplitude of the power spectral density within a preset frequency band by performing a fast Fourier transform on the acceleration time-series data in the motion response data. The temporal morphological encoding is achieved by performing piecewise linear fitting on the position-time curve in the motion response data, extracting the slope sign of each segment and the fitting residual to form an encoding sequence.

10. A coordinate measuring and motion recognition system based on semiconductor processing equipment according to claim 1, characterized in that, The method of identifying specific fault types by matching the abnormal fingerprint database includes: For abnormal measurement actions identified under active detection, their time-domain feature vector, frequency-domain feature vector, and time-series morphological code are extracted to form a fingerprint to be matched. Calculate the comprehensive similarity between the fingerprint to be matched and the abnormal fingerprints of each known fault type in the abnormal fingerprint database. The comprehensive similarity is a weighted fusion of temporal feature vector similarity, frequency domain feature vector similarity, and temporal morphological similarity. When the highest overall similarity value is greater than the preset matching threshold, the corresponding fault type is output.

Images