A method, system, device and medium for operation and maintenance prediction of components of a semiconductor device

Abstract

The application provides a component operation and maintenance prediction method, system, device and medium of a semiconductor equipment, the method comprising: collecting multi-factor operation data during equipment operation, and performing normalization processing on the multi-factor operation data to form normalized task execution time, normalized energy consumption data and normalized current waveform sequence; based on the normalized multi-factor operation data, comparing the corresponding data of the first operation arm and the second operation arm according to the task type and the residual time window to form task time residual features, energy consumption residual features and current waveform residual features, and jointly constructing a mirror residual feature vector; inputting the mirror residual feature vector into a prediction model, the prediction model analyzing the residual features of the continuous time window to generate an abnormal score set, and determining the working state of the fixed 180° opposite double operation arms through a double threshold value determination mechanism to obtain abnormal result data; and outputting a warning prompt based on the abnormal result data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor, and particularly relates to a component operation and maintenance prediction method, system, device and medium of a semiconductor equipment. BACKGROUND

[0002] In a semiconductor wafer manufacturing equipment, a mechanical arm is responsible for key actions such as wafer grabbing, carrying, loading and unloading, and is a core component for normal operation of a production line. Common mechanical arm structures on the market mainly include single operation arm and double operation arm. Among them, the fixed 180° opposite double operation arm is widely used in the industry because of its relatively simple structure, low cost, and higher efficiency than the single operation arm.

[0003] However, the fixed 180° opposite double operation arm will also have problems such as component wear, energy efficiency decline, and unstable action in long-term operation. The existing technology usually only relies on a single data to judge, for example, whether the execution time of a task is too long or whether the energy consumption is excessive. This method is easily affected by occasional fluctuations or noise, and the result is either false positives or false negatives, making it difficult to accurately reflect the true health status. More importantly, the existing method rarely uses dynamic signals in the running process, such as the waveform of the current change over time, which can reveal hidden problems such as motor overload and jamming, but is often overlooked.

[0004] In addition, the traditional method is mostly based on static threshold judgment and does not analyze the operation arm running process as a whole in time series, so it is difficult to discover potential degradation trends at an early stage. The traditional method often waits until the device has a significant failure before the system issues an alarm, lacking forward-looking support for operation and maintenance work.

[0005] Therefore, how to utilize various running information such as task execution time, energy consumption data and current waveform sequence under the structural characteristics of the fixed 180° opposite double operation arm, and identify abnormalities by comparing the differences between the double operation arms under the same task, has become a technical problem that needs to be solved in the field. SUMMARY

[0006] The present application provides a component operation and maintenance prediction method, system, device and medium of a semiconductor equipment to solve the problems in the prior art.

[0007] In a first aspect, the present application provides a component operation and maintenance prediction method of a semiconductor equipment, comprising:

[0008] Multi-factor operation data collection and preprocessing, collecting multi-factor operation data of the fixed 180° opposite double operation arms during the device operation process, including task execution duration, energy consumption data and current waveform sequence, and performing normalization processing on the multi-factor operation data to form normalized task execution duration, normalized energy consumption data and normalized current waveform sequence;

[0009] Mirror residual feature vector construction, based on the normalized task execution duration, normalized energy consumption data and normalized current waveform sequence, comparing the corresponding data of the first operation arm and the second operation arm according to the task type and the residual time window to form task duration residual features, energy consumption residual features and current waveform residual features, and jointly constructing a mirror residual feature vector based on the three kinds of residual features;

[0010] Prediction model construction and anomaly detection, inputting the mirror residual feature vector into a prediction model, the prediction model analyzing the mirror residual feature vector of the continuous time window to generate an abnormal score set, and determining the working state of the fixed 180° opposite double operation arms through a double threshold judgment mechanism to obtain abnormal result data;

[0011] Operation and maintenance prompt generation, based on the abnormal result data, outputting a warning prompt, the warning prompt including operation and maintenance prompts and abnormal diagnosis prompts.

[0012] In one possible design, the mirror residual feature vector construction includes:

[0013] Constructing a residual time window, constructing a task duration sample set, an energy consumption sample set and a current waveform sample set based on the task type in each residual time window, and collecting each sample set according to the operation arm identifier to form a first operation arm sample set and a second operation arm sample set, and starting residual feature analysis when the residual analysis condition is met;

[0014] In the task duration sample set, the task duration sample median and the task duration sample median absolute deviation of the first operation arm and the second operation arm are calculated to construct the task duration residual feature, and the standardized task duration residual is formed through standardization processing;

[0015] In the energy consumption sample set, the energy consumption sample median and the energy consumption sample median absolute deviation of the first operation arm and the second operation arm are calculated to construct the energy consumption residual feature, and the standardized energy consumption residual is formed through standardization processing;

[0016] In the current waveform sample set, the current waveform sample median and the current waveform sample median absolute deviation of the first operation arm and the second operation arm are calculated to construct the current waveform residual feature, and the standardized current waveform residual feature is formed through standardization processing;

[0017] The standardized task duration residual, the standardized energy consumption residual and the standardized current waveform residual are combined to form a mirror residual feature vector under the same residual time window and task type.

[0018] In a possible design, the prediction model construction and the anomaly detection include:

[0019] A prediction model is constructed based on the time sequence characteristics of the mirror residual feature vector;

[0020] The mirror residual feature vector composed of the standardized task execution duration residual, the standardized energy consumption residual and the standardized current waveform residual is taken as an input of the prediction model, the working state of the fixed 180° opposite double manipulator is predicted and analyzed, and a multi-dimensional anomaly score set is output;

[0021] Based on the anomaly score set, a double-threshold judgment mechanism combining a bias threshold and a confidence threshold is used to make an anomaly judgment on the working state of the fixed 180° opposite double manipulator.

[0022] In a possible design, the residual time window is a continuous operation process of the manipulator, which is divided according to a fixed time length;

[0023] In each residual time window, according to the task type, the tasks falling into the residual time window are screened out, the normalized task execution duration, the normalized energy consumption data and the normalized current waveform sequence of each task are taken as a task duration sample, an energy consumption sample and a current waveform sample respectively for sampling, and a task duration sample set, an energy consumption sample set and a current waveform sample set are formed;

[0024] According to the manipulator identifier, the task duration sample set, the energy consumption sample set and the current waveform sample set are collected to form a first manipulator sample set and a second manipulator sample set;

[0025] The corresponding data of the first manipulator sample set and the second manipulator sample set are compared to obtain a task duration residual feature, an energy consumption residual feature and a current waveform residual feature.

[0026] In a possible design, the component operation and maintenance prediction method is based on the structural characteristics of the fixed 180° opposite double manipulator, under the same residual time window and task type, the normalized task execution duration, the normalized energy consumption data and the normalized current waveform sequence are sampled to form corresponding task duration samples, energy consumption samples and current waveform samples, which are mirror operations.

[0027] The task duration sample, the energy consumption sample and the current waveform sample are collected according to the operation arm identifier to form a first operation arm sample set and a second operation arm sample set, and comparison is performed, and the difference is the mirror residual.

[0028] In a possible design, the double-threshold determination mechanism includes:

[0029] When the comprehensive abnormality score is greater than or equal to the preset emergency threshold, it is determined that the operation arm is abnormal, and the abnormality level is identified as a third-level abnormality.

[0030] When the comprehensive abnormality score is less than the preset emergency threshold and greater than or equal to the preset deviation threshold, it is determined that the current residual time window is abnormally deviated, and the continuous counter is incremented by 1.

[0031] When the comprehensive abnormality score is abnormally deviated in the continuous N residual time windows, it is determined that the operation arm is abnormal, and the abnormality level is identified as a first-level abnormality.

[0032] When the comprehensive abnormality score is abnormally deviated in the continuous M residual time windows, it is determined that the operation arm is abnormal, and the abnormality level is identified as a second-level abnormality.

[0033] When the comprehensive abnormality score is less than the preset deviation threshold, the continuous counter is cleared.

[0034] N and M are positive integers, and M is greater than N, and the emergency threshold is greater than the deviation threshold.

[0035] In a possible design, the prediction model is a deep neural network model, and parameter pruning and 8-bit integer quantization methods are used for optimization.

[0036] In a second aspect, the application provides a component operation and maintenance prediction system of a semiconductor device, including:

[0037] The multi-factor operation data acquisition and preprocessing module acquires multi-factor operation data of the fixed 180° opposed double operation arms during device operation, including task execution duration, energy consumption data and current waveform sequence, and performs normalization processing on the multi-factor operation data to form normalized task execution duration, normalized energy consumption data and normalized current waveform sequence.

[0038] The mirror residual feature vector construction module compares the corresponding data of the first operation arm and the second operation arm according to the task type and the residual time window based on the normalized task execution duration, the normalized energy consumption data and the normalized current waveform sequence, forms the task duration residual feature, the energy consumption residual feature and the current waveform residual feature, and jointly constructs the mirror residual feature vector based on the three residual features.

[0039] The prediction model construction and anomaly detection module inputs the mirror residual feature vector into a prediction model, the prediction model analyzes the mirror residual feature vectors of consecutive time windows, generates a set of anomaly scores, and determines the working state of the fixed 180° opposing double operation arms through a double threshold judgment mechanism to obtain anomaly result data;

[0040] The operation and maintenance prompt generation module outputs a warning prompt based on the anomaly result data, and the warning prompt includes an operation and maintenance prompt and an anomaly diagnosis prompt.

[0041] In a third aspect, the present application provides an electronic device, comprising:

[0042] a processor; and

[0043] a memory for storing executable instructions of the processor;

[0044] The processor is configured to execute any one of the possible methods described in the first aspect by executing the executable instructions.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by a processor to implement any one of the possible methods described in the first aspect.

[0046] The present application provides a visual monitoring method, system, device and medium for a semiconductor device. The method collects multi-factor operation data including task execution time, energy consumption data and current waveform sequence during the operation of the semiconductor device, and constructs a mirror residual feature vector under the condition of fixed 180° opposing double operation arms. The method can reflect the difference in task execution based on the symmetry comparison of the double operation arms. The method further combines a prediction model and a double threshold judgment mechanism to realize hierarchical anomaly detection of the operation arm working state and generate operation and maintenance prompts and diagnosis prompts, so as to identify potential component degradation at an early stage, reduce the risk of failure downtime, and improve the forward-looking and reliability of device operation and maintenance.

[0047] Further, by constructing a residual time window and distinguishing task types to collect samples, the robustness of residual feature calculation is ensured. The median and median absolute deviation of task duration, energy consumption and current waveform are used to extract residual features, which enhances the anti-interference ability. A deep neural network is introduced into the prediction model, and parameter pruning and 8-bit integer quantization methods are used for lightweight optimization to ensure the real-time inference ability of the prediction model on edge devices. At the same time, the double threshold judgment mechanism sets deviation threshold and emergency threshold and uses continuous window counting method to effectively reduce the false positive rate and false negative rate.

[0048] In summary, the visual monitoring method, system, device and medium of the semiconductor equipment have the following beneficial effects:

[0049] The application performs normalization preprocessing on the task execution duration, energy consumption data and current waveform sequence, so that different types of tasks have a unified comparison basis in terms of speed, energy efficiency and dynamics. Based on the normalized data, an image residual feature vector is constructed. Under the structure of fixed 180° opposite double operation arms, by comparing and analyzing the differences between the two operation arms under the same task, signs of component performance changes can be found earlier. Combined with the residual time window mechanism and the robust calculation method of the median absolute deviation, the interference of incidental fluctuations and noise on the results is effectively weakened, making the residual feature more stable and reliable. By introducing a lightweight deep neural network prediction model and running it in real time on the edge, the dynamic analysis of the residual changes in the continuous time window can timely identify abnormal states and improve the accuracy of detection.

[0050] Therefore, the application not only improves the accuracy and real-time performance of anomaly detection, but also provides interpretable diagnostic information for operation and maintenance personnel, significantly improving the overall operation and maintenance efficiency and reliability of semiconductor equipment. BRIEF DESCRIPTION OF DRAWINGS

[0051] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application.

[0052] Figure 1 is a flowchart of a component operation and maintenance prediction method of a semiconductor equipment according to an example embodiment of the application;

[0053] Figure 2 is a flowchart of the construction of an image residual feature vector according to an example embodiment of the application;

[0054] Figure 3 is a flowchart of the construction of a prediction model and anomaly detection according to an example embodiment of the application;

[0055] Figure 4 is a structural diagram of a component operation and maintenance prediction system of a semiconductor equipment according to an example embodiment of the application;

[0056] Figure 5 is a structural diagram of an electronic device according to an example embodiment of the application.

[0057] Through the above drawings, the specific embodiments of the application have been shown, and will be described in more detail in the following. These drawings and written descriptions are not intended to limit the scope of the concept of the application by any means, but to illustrate the concept of the application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0058] The exemplary embodiments will be described in detail hereinbelow with reference to the drawings. In the following description, the same numbers in different drawings represent the same or similar elements unless otherwise represented. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they only describe example implementations consistent with some aspects of the present application, as detailed in the appended claims.

[0059] The embodiment provides a component operation and maintenance prediction method, system, device and medium of a semiconductor equipment, which is suitable for a semiconductor equipment with fixed 180° opposite double operation arms. In this scenario, the two operation arms usually cooperatively complete wafer grabbing, carrying, loading and unloading tasks in a mirror image manner. With the increase of running time, the operation arm driving components, motors and transmission mechanisms will gradually deteriorate, which is manifested as that the task execution time, energy consumption data and current waveform sequence gradually show asymmetric differences between the double operation arms. Therefore, the present application realizes early detection and operation and maintenance prompt of component deterioration by constructing mirror image residual features and combining a lightweight prediction model.

[0060] Figure 1 FIG. 1 is a flowchart of a component operation and maintenance prediction method of a semiconductor equipment according to an example embodiment of the present application. As shown in FIG. 1, the component operation and maintenance prediction method of the semiconductor equipment provided by the embodiment includes the following steps. Figure 1 As shown in FIG. 1, the component operation and maintenance prediction method of the semiconductor equipment provided by the embodiment includes the following steps.

[0061] In the embodiment, the “task” refers to an independent action unit completed by the wafer transfer operation arm in a process cycle under the semiconductor equipment control system, each task has a clear start time stamp and end time stamp, which is used to describe the complete execution process of the action. Typical tasks include grabbing, carrying, loading and unloading actions.

[0062] In order to facilitate subsequent data processing, the embodiment introduces the following identification:

[0063] Task identification (TaskID): represents the original task instance recorded in the semiconductor equipment control system log, which is used to uniquely identify the specific execution process of an operation arm in a process cycle;

[0064] Operation arm identification (ArmID): suitable for fixed 180° opposite double operation arms, used to distinguish the first operation arm (Arm_A) and the second operation arm (Arm_B);

[0065] Task type (TaskType): used to represent the action category of the task, including grabbing, carrying, loading and unloading, the task type is written into the system log after being bound with the task identification, which is used to clearly define the action attribute of each task.

[0066] Step S101, multi-factor operation data collection and preprocessing, collecting multi-factor operation data of fixed 180° opposite double operating arms during device operation, including task execution duration, energy consumption data and current waveform sequence, and performing normalization processing on the multi-factor operation data to form normalized task execution duration, normalized energy consumption data and normalized current waveform sequence.

[0067] In this step, the semiconductor device collects multi-factor operation data of fixed 180° opposite double operating arms during normal operation, including first operating arm (Arm_A) and second operating arm (Arm_B). The multi-factor operation data includes task execution duration, energy consumption data and current waveform sequence, and the specific collection process is as follows:

[0068] Task execution duration:

[0069] The start timestamp (t start ) and end timestamp (t end ) of each task are extracted from the system log of the semiconductor device control system, and the task execution duration of each task is calculated:

[0070]

[0071] Where:

[0072] ExecutionTime j is the task execution duration of task j;

[0073] is the start timestamp of task j;

[0074] is the end timestamp of task j;

[0075] Bind each task execution duration with operating arm identifier (ArmID), task identifier (TaskID) and task type (TaskType).

[0076] Energy consumption data:

[0077] The voltage and current of the first operating arm and the second operating arm are collected by the power collection module or external power sensor of the semiconductor device, and collected at a fixed sampling period;

[0078] According to the voltage and current of the first operating arm and the second operating arm, the instantaneous power of the first operating arm and the second operating arm is calculated respectively, and the instantaneous power curve of the task execution interval [t start , t end ] is formed by continuous splicing;

[0079] The instantaneous power curve is integrated in the task execution interval [t start ,t end ] to obtain the energy consumption data (TotalEnergy) of the task, and the calculation formula is as follows:

[0080]

[0081] Wherein:

[0082] TotalEnergy j is the energy consumption data of the task j;

[0083] is the start time stamp of the task j;

[0084] is the end time stamp of the task j;

[0085] P(t) is the instantaneous power of the task j.

[0086] Each energy consumption data (TotalEnergy) is bound with the operation arm identifier (ArmID), the task identifier (TaskID) and the task type (TaskType).

[0087] Current waveform sequence:

[0088] The current signal I(t) of the first operation arm and the second operation arm is collected in real time in the task execution process through the current detection module or the Hall sensor of the semiconductor device, and the current waveform sequence (CurrentWaveform) of the task execution interval [t start ,t end ] is obtained, wherein:

[0089] The current waveform sequence of the task TaskID=j is as follows:

[0090]

[0091] Wherein:

[0092] CurrentWaveform j (t) is the current waveform sequence of the task j, and t is the sampling time in the task execution process;

[0093] is the current signal sampling point of the first operation arm;

[0094] j, the task identifier is j;

[0095] is the start time stamp of the task j;

[0096] End Time Stamp of Task j;

[0097] Bind each current waveform sequence with Task ID and Task Type.

[0098] Preprocess the collected multi-factor operation data, including:

[0099] Normalization processing:

[0100] For each task execution duration, use the minimum-maximum normalization method to form the normalized task execution duration, which is used to eliminate the differences of different task periods:

[0101]

[0102] wherein, ExecutionTime j is the task execution duration of task j, ExecutionTime j is the normalized task execution duration of task j, ExecutionTime max is the maximum value of the task execution duration in the collected task execution duration sequence, ExecutionTime min is the minimum value of the task execution duration in the collected task execution duration sequence.

[0103] For each energy consumption data, use the minimum-maximum normalization method to form the normalized energy consumption data to ensure the comparability between different tasks:

[0104]

[0105] wherein, TotalEnergy j is the energy consumption data of task j, TotalEnergy j is the normalized energy consumption data of task j, TotalEnergy max is the maximum value in the collected energy consumption data sequence, TotalEnergy min is the minimum value in the collected energy consumption data sequence.

[0106] For each current waveform sequence, use the Z-score standardization method to form the normalized current waveform sequence to eliminate the differences of different motor rated powers:

[0107]

[0108] wherein, CurrentWaveform j (t) is the current waveform sequence of task j, CurrentWaveform j(t) is the normalized current waveform sequence of task j, μ is the current waveform mean, and σ is the current waveform standard deviation.

[0109] Filtering and smoothing processing:

[0110] The moving average method is used to remove accidental spikes from the normalized task execution duration data and normalized energy consumption data.

[0111] The Butterworth filter is used on the normalized current waveform sequence, with a cutoff frequency set to twice the rated drive frequency to retain low-frequency trends related to degradation while suppressing high-frequency noise.

[0112] After preprocessing, the two operating arm corresponding data sets required for mirror residual feature construction are obtained:

[0113]

[0114] Among them:

[0115] TaskID, task identification;

[0116] ArmID, operating arm identification;

[0117] TaskType, task type, used to label each task according to the typical action of the semiconductor equipment operating arm, including grabbing, carrying, loading and unloading;

[0118] ExecutionTime j ' is the normalized task execution duration of task j;

[0119] TotalEnergy j ' is the normalized energy consumption data of task j;

[0120] is the normalized current waveform sequence of task j.

[0121] After preprocessing the collected multi-factor operation data, the dimension difference under different tasks and different motor power conditions can be eliminated, noise interference can be reduced, and the one-to-one correspondence of the multi-factor operation data of the two operating arms at the task level can be ensured. This not only provides clean and comparable input data for subsequent mirror residual feature construction, but also ensures the repeatability and reliability of the anomaly detection process.

[0122] It should be noted that each task includes normalized task execution duration, normalized energy consumption data and normalized current waveform sequence to ensure that different tasks have comparability in speed, energy efficiency and dynamics.

[0123] Step S102, mirror residual feature vector construction, based on the normalized task execution time, normalized energy consumption data and normalized current waveform sequence, comparing the corresponding data of the first operating arm and the second operating arm according to the task type and the residual time window, forming the task time residual feature, the energy consumption residual feature and the current waveform residual feature, and jointly constructing the mirror residual feature vector based on the three kinds of residual features.

[0124] In this step, first, the running data of the double operating arms are compared under the same residual time window and task type, and the results are uniformly represented as residual features, and then combined into a mirror residual feature vector, which is used to reflect the asymmetry of the operating arms in the task execution process.

[0125] Through the construction of the mirror residual feature, the noise interference caused by the overall operating condition change of the operating arm can be effectively filtered out, and the difference in execution time, energy consumption and current waveform of the fixed 180° opposite double operating arms is highlighted, so that the component deterioration signs can be sensitively captured, and high-quality input is provided for the subsequent prediction model.

[0126] Step S103, prediction model construction and anomaly detection, inputting the mirror residual feature vector into the prediction model, the prediction model analyzing the mirror residual feature vector of the continuous time window, generating an abnormal score set, and determining the working state of the fixed 180° opposite double operating arms through a double threshold judgment mechanism to obtain abnormal result data.

[0127] In this step, the prediction model learns and infers the mirror residual feature vector using time series features, outputs abnormal score results corresponding to task execution time, energy consumption and current waveform, and realizes hierarchical judgment of abnormal levels by combining a double threshold judgment mechanism.

[0128] Through the combination of the prediction model and the double threshold judgment mechanism, the accuracy and robustness of the anomaly detection are ensured, and false positives and false negatives caused by single threshold methods are avoided, and the abnormal state can be classified, providing reliable basis for operation and maintenance prompt and diagnosis prompt.

[0129] Step S104, operation and maintenance prompt generation, outputting an early warning prompt based on the abnormal result data, the early warning prompt including operation and maintenance prompt and abnormal diagnosis prompt.

[0130] In this step, the early warning prompt based on the abnormal result data of step S103 includes operation and maintenance prompt and abnormal diagnosis prompt.

[0131] The operation and maintenance prompt is:

[0132] When the abnormal level is level one, normal early warning is performed.

[0133] When the abnormality level is a secondary abnormality, a prompt of "recommended maintenance" is given;

[0134] When the abnormality level is a tertiary abnormality, a prompt of "check immediately" is given.

[0135] The abnormality diagnosis prompt is:

[0136] The task duration abnormality score (Score_T), the energy consumption abnormality score (Score_E) and the current waveform abnormality score (Score_I) are obtained.

[0137] When the task duration abnormality score (Score_T) is greater than or equal to a preset task duration abnormality threshold, a prompt of "check transmission resistance and operation arm jamming" is given.

[0138] When the energy consumption abnormality score (Score_E) is greater than or equal to a preset energy consumption abnormality threshold, a prompt of "check motor efficiency and bearing lubrication state" is given.

[0139] When the current waveform abnormality score (Score_I) is greater than or equal to a preset current waveform abnormality threshold, a prompt of "check motor driver, encoder or load balancing" is given.

[0140] The pre-alarm release rule is set as follows:

[0141] When the comprehensive abnormality scores of R consecutive residual time windows are all less than a preset deviation threshold, the operation and maintenance prompt is released. R is a positive integer, and in the embodiment, R is 5.

[0142] Figure 2 FIG. 1 is a flow diagram of a mirror residual feature vector construction method according to an example embodiment of the present application. As shown in FIG. 1, the mirror residual feature vector construction method provided in the embodiment includes the following steps. Figure 2

[0143] In the embodiment, in order to ensure the stability and comparability of the residual features, a time window mechanism is introduced before the residual features are constructed, and the task duration residual feature, the energy consumption residual feature and the current waveform residual feature are calculated in the constructed residual time window.

[0144] In step S1021, a residual time window is constructed, a task duration sample set, an energy consumption sample set and a current waveform sample set are constructed based on the task type in each residual time window, each sample set is collected according to the operation arm identifier to form a first operation arm sample set and a second operation arm sample set, and residual feature analysis is started when the residual analysis condition is met.

[0145] In this step, continuous time windows are used for data aggregation. The residual time window is the continuous operation process of the operation arm, which is divided according to a fixed time length: ​

[0146]

[0147] wherein, is the start time of the kth residual time window, The rolling update is performed with a step W (e.g. W = 10 min).

[0148] In each of the residual time windows, according to the task type, the tasks falling into the residual time window are screened out, and the normalized task execution time, the normalized energy consumption data and the normalized current waveform sequence of each task are sampled as the task time sample, the energy sample and the current waveform sample respectively to form a task time sample set, an energy sample set and a current waveform sample set. Each normalized current waveform sequence corresponds to a task instance and participates in the set statistics as an independent sample unit.

[0149] Subsequently, according to the operation arm identifier, the task time sample set, the energy sample set and the current waveform sample set are collected to form a first operation arm sample set and a second operation arm sample set.

[0150] Finally, the corresponding data of the first operation arm sample set and the second operation arm sample set are compared to obtain the task time residual feature, the energy consumption residual feature and the current waveform residual feature.

[0151] To ensure the statistical stability of the residual feature, the embodiment sets a residual analysis condition, and only when the residual analysis condition is met, the residual feature analysis is started. Specifically:

[0152] When the task time sample, the energy sample and the current waveform sample in the first operation arm sample set are all greater than or equal to the preset minimum sample threshold (n min ) and the task time sample, the energy sample and the current waveform sample in the second operation arm sample set are all greater than or equal to the preset minimum sample threshold (n min ), the task time residual feature, the energy consumption residual feature and the current waveform residual feature in the residual time window are calculated;

[0153] Otherwise, the residual time window is invalid and does not participate in the prediction analysis.

[0154] In the embodiment, the preset minimum sample threshold n min is 5-10 samples.

[0155] The embodiment maintains the timing consistency and low delay of online anomaly detection on the one hand, and ensures the statistical reliability of the residual feature on the other hand, avoiding occasional false alarms caused by fluctuation of production capacity or shortage of tasks.

[0156] It should be noted that, based on the structural characteristics of the fixed 180° opposite double operating arms, the normalized task execution duration, normalized energy consumption data and normalized current waveform sequence are sampled under the same residual time window and task type, to form corresponding task duration samples, energy samples and current waveform samples, i.e. mirror operation, in the embodiment.

[0157] The task duration samples, energy samples and current waveform samples are collected according to the operating arm identifier to form the first operating arm sample set and the second operating arm sample set, and compared, and the difference is the mirror residual.

[0158] In step S1022, the median and median absolute deviation of the task duration samples of the first operating arm and the second operating arm are calculated in the task duration sample set, to construct the task duration residual feature, and the standardized task duration residual is formed by standardization processing.

[0159] In this step, in each residual time window (Win k ), the median and median absolute deviation of the task duration samples of the first operating arm and the second operating arm are calculated according to the task type, based on the task duration sample set in the first operating arm sample set and the second operating arm sample set, to construct the task duration residual feature. Specifically:

[0160] All task duration sample sets that satisfy the task type τ (TaskType = τ) and belong to the residual time window (Win k ) are {ExcutionTime j | TaskType = τ, j ∈ Win k}, then:

[0161] The median of the task duration samples of the first operating arm is , and the median absolute deviation of the task duration samples is .

[0162] The median of the task duration samples of the second operating arm is , and the median absolute deviation of the task duration samples is .

[0163] Wherein, τ is the task type, and T is the task duration sample set belonging to the task type τ.

[0164] It should be noted that the median (Median) is the middle value after sorting the samples, and the median absolute deviation (MAD) is the median of the absolute deviation relative to the median, which is a robust statistical method in the prior art.

[0165] Then, the task duration residual feature is calculated based on the median of the task duration sample:

[0166]

[0167] Wherein:

[0168] τ, is the task type;

[0169] is the median of the task duration sample of the first operating arm in the kth residual time window;

[0170] is the median of the task duration sample of the second operating arm in the kth residual time window.

[0171] The task duration residual feature is subjected to Z-score standardization to form a standardized task duration residual:

[0172]

[0173] Wherein:

[0174] is the task duration residual feature with the task type τ and belonging to the residual time window Win k

[0175] is the absolute deviation of the median of the task duration sample of the first operating arm;

[0176] is the absolute deviation of the median of the task duration sample of the second operating arm.

[0177] It should be noted that the median of the task duration sample and the absolute deviation of the median of the task duration sample are adopted to weaken the influence of occasional abnormalities on the statistical sample, so that the task duration residual feature more stably reflects the systematic speed difference of the two operating arms in the same task. Through Z-score standardization, the task duration residual feature is unified to a dimensionless space, which is convenient for being combined with the standardized energy consumption residual and the standardized current waveform residual to form a mirror residual feature vector.

[0178] In step S1023, the median of the energy consumption sample and the absolute deviation of the median of the energy consumption sample of the first operating arm and the second operating arm are calculated in the energy consumption sample set, so as to construct an energy consumption residual feature, and a standardized energy consumption residual is formed through standardization.

[0179] In this step, in each residual time window (Win k ​According to the task type, the energy consumption sample median and the energy consumption sample median absolute deviation of the two operating arms are calculated based on the energy consumption sample set in the first operating arm sample set and the second operating arm sample set, and the energy consumption residual feature is constructed therefrom. Specifically:

[0180] All energy consumption sample sets satisfying the task type τ (TaskType = τ) and belonging to the residual time window (Win k ) are {TotalEnergy k |TaskType = τ, j ∈ Win k}, then:

[0181] The energy consumption sample median of the first operating arm is , and the energy consumption sample median absolute deviation is .

[0182] The energy consumption sample median of the second operating arm is , and the energy consumption sample median absolute deviation is .

[0183] Wherein, τ is the task type, and T is the energy consumption sample set belonging to the task type τ.

[0184] It should be noted that the median (Median) is the middle value after sorting the samples, and the median absolute deviation (MAD) is the median of the absolute deviation relative to the median, which is a robust statistical method in the prior art.

[0185] Then, the energy consumption residual feature is calculated based on the energy consumption sample median:

[0186]

[0187] Wherein:

[0188] τ, is the task type;

[0189] , is the energy consumption sample median of the first operating arm in the kth residual time window;

[0190] , is the energy consumption sample median of the second operating arm in the kth residual time window.

[0191] The energy consumption residual is subjected to Z-score standardization to form the standardized energy consumption residual:

[0192]

[0193] Wherein:

[0194] the energy consumption residual feature of the first operating arm in the residual time window Win k ;

[0195] the absolute deviation of the median of the energy consumption sample of the first operating arm;

[0196] the absolute deviation of the median of the energy consumption sample of the second operating arm.

[0197] It should be noted that the median of the energy consumption sample and the absolute deviation of the median of the energy consumption sample are used to weaken the influence of occasional abnormal energy consumption data (such as current fluctuations or sampling spikes in a single task) on the statistical quantity, so that the energy consumption residual feature can more stably reflect the energy consumption difference of the two operating arms under similar tasks. Through Z-score standardization processing, the energy consumption residual feature is unified to a dimensionless space, which is convenient for forming a mirror residual feature vector together with the standardized task time residual and the standardized current waveform residual.

[0198] In step S1024, the median of the current waveform sample and the absolute deviation of the median of the current waveform sample of the first operating arm and the second operating arm are calculated in the current waveform sample set, so as to construct the current waveform residual feature, and the standardized current waveform residual feature is formed through standardization processing.

[0199] In this step, the median of the current waveform sample and the absolute deviation of the median of the current waveform sample of the two operating arms are calculated according to the task type based on the current waveform sample in each residual time window (Win k ), and the current waveform residual feature is constructed. Specifically:

[0200] All current waveform samples satisfying the task type τ (TaskType = τ) and belonging to the residual time window (Wink) are , then:

[0201] The median of the current waveform sample of the first operating arm is , and the absolute deviation of the median of the current waveform sample is ;

[0202] The median of the current waveform sample of the second operating arm is , and the absolute deviation of the median of the current waveform sample is .

[0203] Wherein, τ is the task type, and T is the current waveform sample set belonging to the task type τ.

[0204] It should be noted that the median is the middle value of the sorted sample, and the median absolute deviation (MAD) is the median of the absolute deviation relative to the median, which belongs to the robust statistical method in the prior art.

[0205] Then, the current waveform residual feature is calculated based on the median of the current waveform sample:

[0206]

[0207] Wherein:

[0208] τ, is the task type;

[0209] is the median of the first operating arm current waveform sample in the kth residual time window;

[0210] is the median of the second operating arm current waveform sample in the kth residual time window.

[0211] The current waveform residual feature is subjected to Z-score standardization to form a standardized current waveform residual:

[0212]

[0213] Wherein:

[0214] is the current waveform residual feature with the task type τ and belonging to the residual time window Win k

[0215] is the median absolute deviation of the current waveform sample of the first operating arm;

[0216] is the median absolute deviation of the current waveform sample of the second operating arm.

[0217] In this embodiment, the standardized current waveform residual feature is used to comprehensively characterize the differences in speed, energy efficiency and dynamic behavior of the two operating arms.

[0218] It should be noted that the median of the current waveform sample and the median absolute deviation of the current waveform sample are used to weaken the influence of abnormal values caused by occasional interference (such as transient electrical noise and sampling jitter) in a single current waveform, so that the current waveform residual feature can more robustly reflect the dynamic differences of the two operating arms under similar tasks. Through Z-score standardization, the current waveform residual feature is unified to a dimensionless space, which is convenient for being combined with the standardized task time residual and the standardized energy consumption residual to form a mirror residual feature vector.

[0219] ​Step S1025, under the same residual time window and task type, the standardized task duration residual, the standardized energy consumption residual and the standardized current waveform residual are jointly constructed into a mirror residual feature vector.

[0220] In this step, based on the standardized task duration residual, the standardized energy consumption residual and the standardized current waveform residual, in each residual time window, a mirror residual feature vector is constructed according to the task type:

[0221]

[0222] Wherein:

[0223] WinID, residual time window identifier;

[0224] TaskType, task type;

[0225] ZT, standardized task execution duration residual;

[0226] ZE, standardized energy consumption residual;

[0227] ZI, standardized current waveform residual.

[0228] In the kth residual time window, when the task type is τ, the mirror residual feature vector is specifically represented as:

[0229]

[0230] It should be noted that by jointly constructing the three types of standardized residuals into a feature vector under the unified time window and task type, the differences in speed (task duration), energy efficiency (energy consumption) and dynamic behavior (current waveform) of the two operating arms can be reflected at the same time, thereby enhancing the comprehensiveness and stability of the degradation discrimination. The mirror residual feature vector as the input of the prediction model provides multi-dimensional support for the subsequent double-threshold judgment mechanism.

[0231] Figure 3 is a flowchart of the prediction model construction and anomaly detection according to an example embodiment of the present application. As Figure 3 shown, the prediction model construction and anomaly detection method provided by the present embodiment comprises:

[0232] Step S1031, based on the time sequence characteristics of the mirror residual feature vector, a prediction model is constructed.

[0233] In the embodiment, the prediction model is a deep neural network model, and parameter pruning and 8-bit integer quantization are used for lightweight optimization. The prediction model uses a gated recurrent unit (GRU) network, which is a simplified structure of a recurrent neural network and has the characteristics of small parameter quantity and high computational efficiency, and is suitable for processing sequences of image residual feature vectors in continuous time windows to extract the law of residual evolution over time.

[0234] The prediction model can be implemented based on existing open source frameworks such as TensorFlow Lite, PyTorch Mobile, etc. The present scheme does not involve the training process of the model, and only lightweight optimization is performed when implemented, including:

[0235] Parameter pruning: by analyzing the sensitivity of each weight in the prediction model to the output result, the neural connection or convolution kernel coefficient with low contribution is set to zero, thereby reducing the calculation amount and storage requirement. In implementation, a sparse weight matrix can be automatically generated using a model compression tool such as TensorFlow Model Optimization Toolkit, and a sparse computing library is deployed on the edge device for efficient inference. In the embodiment, parameter pruning is used to reduce redundant calculation of the prediction model when running on the edge server or local gateway, to ensure real-time processing of the image residual feature vector at the window level time granularity.

[0236] 8-bit integer quantization method: in the case of keeping the prediction model structure unchanged, the floating-point weight and activation value are uniformly compressed to 8-bit integer representation to significantly reduce the memory occupation and calculation overhead of the prediction model. In implementation, the quantization conversion interface provided by PyTorch Mobile or TensorFlow Lite can be used to complete weight quantization in the offline stage, and the quantized inference kernel is called on the edge to perform execution. In the embodiment, the 8-bit integer quantization method is used to further compress the model weight and calculation precision, to reduce the storage and calculation requirement, so that the prediction model can stably run under the limited computing power condition of the edge server or local gateway, to meet the real-time requirement of online anomaly detection.

[0237] In step S1032, the image residual feature vector composed of the normalized task execution duration residual, the normalized energy consumption residual, and the normalized current waveform residual is taken as the input of the prediction model, the working state of the fixed 180° opposed dual operating arms is predicted and analyzed, and a multi-dimensional abnormal score set is output.

[0238] In this step, the mirror residual feature vector output in step S1025 is input into the prediction model to obtain a set of abnormal scores. The set of abnormal scores includes component-level abnormal scores and a comprehensive abnormal score. Wherein:

[0239] The component-level abnormal score is the task duration abnormal score (Score_T), the energy consumption abnormal score (Score_E), and the current waveform abnormal score (Score_I) respectively output by the prediction model based on the standardized task duration residual (ZT), the standardized energy consumption residual (ZE), and the standardized current waveform residual (ZI) in the mirror residual feature vector. The component-level abnormal score is used to represent the specific deviation in each dimension and provides a detailed basis for subsequent operation and maintenance prompts.

[0240] The comprehensive abnormal score is obtained by weighted averaging the task duration abnormal score (Score_T), the energy consumption abnormal score (Score_E), and the current waveform abnormal score (Score_I):

[0241]

[0242] Wherein, β1, β2, and β3 are components of each weight, and satisfy β1+β2+β3=1.

[0243] In this embodiment, the comprehensive abnormal score provides data basis for the double-threshold mechanism.

[0244] Step S1033, based on the set of abnormal scores, a double-threshold judgment mechanism combining deviation threshold and confidence threshold is used to perform abnormal judgment on the working state of the fixed 180° opposed double operating arms.

[0245] In this step, the set of abnormal scores is subjected to double-threshold judgment to reduce false positives. The double-threshold judgment mechanism includes:

[0246] When the comprehensive abnormal score is greater than or equal to the preset emergency threshold, it is determined that the operating arm has an abnormality, and the abnormality level is identified as a third-level abnormality;

[0247] When the comprehensive abnormal score is less than the preset emergency threshold and greater than or equal to the preset deviation threshold, it is determined that the current residual time window has an abnormal deviation, and the continuous counter is incremented by 1;

[0248] When the comprehensive abnormal score has an abnormal deviation in the continuous N residual time windows, it is determined that the operating arm has an abnormality, and the abnormality level is identified as a first-level abnormality;

[0249] When the comprehensive abnormal score has an abnormal deviation in the continuous M residual time windows, it is determined that the operating arm has an abnormality, and the abnormality level is identified as a second-level abnormality;

[0250] When the comprehensive abnormality score is less than the preset deviation threshold, the continuous counter is cleared.

[0251] Wherein, N and M are positive integers, and M is greater than N, the emergency threshold is greater than the deviation threshold. In this embodiment, N is 3, and M is 10.

[0252] Through the above-mentioned double threshold determination mechanism, it can be ensured that the abnormality detection result not only considers the difference amplitude, but also considers the evidence stability, thereby effectively reducing the false positive rate.

[0253] When an anomaly is detected, an abnormality result data is constructed:

[0254]

[0255] Wherein:

[0256] TriggerWinID, the residual time window where the abnormal deviation occurs;

[0257] ContextWinIDSet, the continuous residual time window where the anomaly occurs;

[0258] TaskType, the task type;

[0259] AnonalyLevel, the abnormality level, including first-level abnormality, second-level abnormality and third-level abnormality;

[0260] ScoreSet, the abnormality score set, task duration abnormality score (Score_T), energy consumption abnormality score (Score_E), current waveform abnormality score (Score_I) and comprehensive abnormality score (Score).

[0261] The abnormality result data is recorded in the log to generate a log record.

[0262] It should be noted that the TriggerWinID, ContextWinIDSet and TaskType fields are used for traceability and log recording in engineering implementation, which is convenient for subsequent analysis of the context window range where the anomaly occurs, and is not used as the basis data for subsequent warning prompt.

[0263] Figure 4 is a structural schematic diagram of a component operation and maintenance prediction system of a semiconductor device according to an example embodiment. As Figure 4 shown, the component operation and maintenance prediction system 400 of the semiconductor device provided in this embodiment includes a multi-factor operation data acquisition and preprocessing module 410, a mirror residual feature vector construction module 420, a prediction model construction and abnormality detection module 430 and an operation and maintenance prompt generation module 440.

[0264] The multi-factor operation data acquisition and preprocessing module 410 acquires multi-factor operation data of the fixed 180° opposite double operation arms during device operation, including task execution duration, energy consumption data and current waveform sequence, and performs normalization processing on the multi-factor operation data to form normalized task execution duration, normalized energy consumption data and normalized current waveform sequence.

[0265] The mirror residual feature vector construction module 420 compares the corresponding data of the first operation arm and the second operation arm according to the task type and the residual time window based on the normalized task execution duration, the normalized energy consumption data and the normalized current waveform sequence, forms the task duration residual feature, the energy consumption residual feature and the current waveform residual feature, and jointly constructs the mirror residual feature vector based on the three residual features.

[0266] The prediction model construction and anomaly detection module 430 inputs the mirror residual feature vector into a prediction model, the prediction model analyzes the mirror residual feature vector of the continuous time window to generate an abnormal score set, and determines the working state of the fixed 180° opposite double operation arms through a double-threshold judgment mechanism to obtain abnormal result data.

[0267] The operation and maintenance prompt generation module 440 outputs an early warning prompt based on the abnormal result data, and the early warning prompt includes operation and maintenance prompts and abnormal diagnosis prompts.

[0268] Figure 5 is a structural schematic diagram of an electronic device according to an example embodiment. As shown in Figure 5 The electronic device 500 provided in the embodiment includes a processor 501 and a memory 502; wherein:

[0269] The memory 502 is used for storing computer programs, and the memory can also be a flash (flash memory).

[0270] The processor 501 is used for executing execution instructions stored in the memory to realize each step in the above method. For details, please refer to the related description in the previous method embodiment.

[0271] Optionally, the memory 502 can be independent or integrated with the processor 501.

[0272] When the memory 502 is a device independent of the processor 501, the electronic device 500 can further include:

[0273] The bus 503 is used for connecting the memory 502 and the processor 501.

[0274] The embodiment further provides a readable storage medium, and the readable storage medium stores a computer program. When at least one processor of an electronic device executes the computer program, the electronic device executes the method provided by various embodiments.

[0275] The embodiment further provides a program product, and the program product includes a computer program stored in a readable storage medium. At least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to enable the electronic device to implement the method provided by various embodiments.

[0276] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0277] It will be understood that the application is not limited to the precise structures hereinbefore described and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof.

Claims

1. A method for predicting the maintenance of components in semiconductor equipment, characterized in that, include: Multi-factor operation data acquisition and preprocessing: During equipment operation, multi-factor operation data of fixed 180° opposing double operating arms are collected, including task execution time, energy consumption data and current waveform sequence. The multi-factor operation data is then normalized to form normalized task execution time, normalized energy consumption data and normalized current waveform sequence. The mirror residual feature vector is constructed by comparing the corresponding data of the first and second operating arms according to the normalized task execution time, normalized energy consumption data and normalized current waveform sequence, according to the task type and residual time window, forming task duration residual features, energy consumption residual features and current waveform residual features, and then constructing the mirror residual feature vector based on the three residual features. The prediction model is constructed and anomaly detection is performed by inputting the mirror residual feature vector into the prediction model. The prediction model analyzes the mirror residual feature vector of the continuous time window, generates anomaly score set, and uses a dual threshold judgment mechanism to judge the working state of the fixed 180° opposing double manipulators and obtain anomaly result data. Operation and maintenance prompts are generated, and early warning prompts are output based on the abnormal result data. The early warning prompts include operation and maintenance prompts and abnormal diagnosis prompts.

2. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The construction of the mirror residual feature vector includes: Construct a residual time window, and within each residual time window, construct a task duration sample set, an energy consumption sample set, and a current waveform sample set based on the task type. Then, aggregate each sample set according to the operator arm identifier to form the first operator arm sample set and the second operator arm sample set. Finally, start residual feature analysis when the residual analysis conditions are met. In the task duration sample set, the median of the task duration samples of the first operating arm and the second operating arm and the absolute deviation of the median of the task duration samples are calculated to construct the task duration residual features, and standardized task duration residuals are formed through standardization processing. In the energy consumption sample set, the median of the energy consumption samples of the first operating arm and the second operating arm and the absolute deviation of the median of the energy consumption samples are calculated to construct the energy consumption residual characteristics, and standardized energy consumption residuals are formed through standardization processing. In the current waveform sample set, the median and absolute deviation of the current waveform sample of the first operating arm and the second operating arm are calculated to construct the current waveform residual features, and standardized current waveform residual features are formed through standardization processing. Under the same residual time window and task type, the standardized task duration residual, standardized energy consumption residual, and standardized current waveform residual are jointly constructed into a mirror residual feature vector.

3. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The prediction model construction and anomaly detection include: A prediction model is constructed based on the time series characteristics of mirror residual feature vectors; The mirror residual feature vector, composed of standardized task execution time residual, standardized energy consumption residual, and standardized current waveform residual, is used as the input of the prediction model to predict and analyze the working state of the fixed 180° opposing dual operating arms, and outputs a multi-dimensional set of anomaly scores. Based on the aforementioned abnormal score set, a dual-threshold judgment mechanism combining deviation threshold and confidence threshold is adopted to determine the abnormal working state of the fixed 180° opposing dual manipulators.

4. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The residual time window is a division of the continuous operation process of the operating arm into fixed time lengths; Within each residual time window, tasks falling into the residual time window are selected according to task type. The normalized task execution time, normalized energy consumption data and normalized current waveform sequence of each task are sampled as task duration sample, energy consumption sample and current waveform sample, respectively, to form a task duration sample set, energy consumption sample set and current waveform sample set. Based on the operator arm identification, the task duration sample set, energy consumption sample set, and current waveform sample set are collected to form the first operator arm sample set and the second operator arm sample set. By comparing the corresponding data of the first manipulator sample set with the second manipulator sample set, the residual characteristics of task duration, energy consumption, and current waveform are obtained.

5. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The component maintenance prediction method is based on the structural characteristics of a fixed 180° opposing double operating arm. Under the same residual time window and task type, the normalized task execution time, normalized energy consumption data and normalized current waveform sequence are sampled to form corresponding task time samples, energy consumption samples and current waveform samples, which is the mirror operation. The task duration samples, energy consumption samples, and current waveform samples are collected according to the operator arm identifier to form the first operator arm sample set and the second operator arm sample set, and then compared. The difference between them is the mirror residual.

6. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The dual threshold determination mechanism includes: When the comprehensive anomaly score is greater than or equal to the preset emergency threshold, the operating arm is determined to be abnormal and the anomaly level is marked as level three. When the overall anomaly score is less than the preset emergency threshold and greater than or equal to the preset deviation threshold, it is determined that there is an abnormal deviation in the current residual time window, and the continuous counter is incremented by 1. When the comprehensive anomaly score shows abnormal deviations in N consecutive residual time windows, the operating arm is determined to be abnormal and the anomaly level is marked as Level 1 anomaly. When the comprehensive anomaly score shows abnormal deviations in all M consecutive residual time windows, the operating arm is determined to be abnormal and the anomaly level is identified as Level 2 anomaly. When the overall anomaly score is less than the preset deviation threshold, the continuous counter is reset to zero. Where N and M are both positive integers, and M is greater than N, and the emergency threshold is greater than the deviation threshold.

7. The component maintenance prediction method for semiconductor equipment according to claim 1, characterized in that, The prediction model is a deep neural network model, and it is optimized using parameter pruning and 8-bit integer quantization.

8. A component maintenance prediction system for semiconductor equipment, characterized in that, include: The multi-factor operation data acquisition and preprocessing module acquires multi-factor operation data of the fixed 180° opposing double operating arms during equipment operation, including task execution time, energy consumption data and current waveform sequence, and performs normalization processing on the multi-factor operation data to form normalized task execution time, normalized energy consumption data and normalized current waveform sequence. The mirror residual feature vector construction module compares the corresponding data of the first and second operating arms according to the normalized task execution time, normalized energy consumption data and normalized current waveform sequence, according to the task type and residual time window, to form task duration residual features, energy consumption residual features and current waveform residual features, and jointly constructs a mirror residual feature vector based on the three residual features. The prediction model construction and anomaly detection module inputs the mirror residual feature vector into the prediction model. The prediction model analyzes the mirror residual feature vector of the continuous time window, generates anomaly score set, and uses a dual threshold judgment mechanism to judge the working state of the fixed 180° opposing double manipulators and obtain anomaly result data. The operation and maintenance prompt generation module outputs early warning prompts based on the abnormal result data. The early warning prompts include operation and maintenance prompts and abnormal diagnosis prompts.

9. An electronic device, characterized in that, include: processor; as well as, Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 7.

Images