Ai-based semiconductor transport device abnormality prediction system and method

Abstract

The present invention relates to an AI-based anomaly prediction system and method that collects data from various sensors installed in the drive unit of a semiconductor wafer transfer device and analyzes and predicts abnormal signs in real time using an artificial intelligence (AI) model. The system comprises a sensor module, a data collection unit, a data preprocessing unit, an AI analysis unit, an integrated prediction unit, and a visualization and alarm unit, and calculates anomaly probabilities through a CNN-LSTM-based AI model. The prediction results are visualized on a dashboard and recorded in a storage device, and are utilized for model retraining and maintenance automation. Accordingly, it is possible to prevent unexpected equipment failures in advance and improve production efficiency and maintenance stability.

Description

Technology Field

[0001] The present invention relates to an abnormality diagnosis and prediction technology for semiconductor manufacturing equipment, and more specifically, to an AI-based abnormality prediction system and method that improves the reliability of equipment by analyzing data generated from the drive unit of a semiconductor wafer transfer device based on artificial intelligence (AI) to detect signs of abnormality early and calculate the probability of failure. Background Technology

[0002] As the automation of semiconductor manufacturing lines advances, semiconductor transfer devices that precisely transport wafers have a direct impact on production efficiency and quality. Drive components such as transfer robots, servo motors, and vacuum pumps gradually deteriorate due to vibrations, temperature rises, friction, and pressure fluctuations caused by repetitive operations, and unexpected malfunctions can lead to production stoppages.

[0003] Existing maintenance methods relied primarily on periodic inspections or reactive repairs, which simultaneously resulted in unnecessary replacement costs and unexpected breakdowns. While sensor-based monitoring technologies are being applied to some extent, it is difficult to precisely identify complex abnormal patterns through simple threshold comparisons or single-signal analysis.

[0004] Accordingly, there is a need for technology that integrates various sensor data, learns and analyzes it using artificial intelligence, and can predict the probability of anomalies in real time.

[0005] The information described above disclosed in the background technology of this invention is intended only to enhance understanding of the background of the present invention and may therefore include information that does not constitute prior art. The problem to be solved

[0006] The present invention aims to solve the above problems by collecting data from various sensors attached to the main drive unit of a semiconductor wafer transfer device and predicting abnormal signs early using an AI analysis model, thereby preventing sudden failures and improving maintenance efficiency.

[0007] In addition, the analyzed prediction results are visually displayed and provided in the form of alarms, enabling operators to monitor equipment status in real time, and the prediction data is stored on a local storage device or cloud server to be utilized for model retraining and maintenance history management. means of solving the problem

[0008] The present invention is a system for collecting status data in real time from a plurality of sensors attached to a driving component of a semiconductor wafer transfer device and predicting abnormal signs through artificial intelligence (AI) analysis of the data, comprising: a sensor module unit including a vibration sensor, a temperature sensor, a noise sensor, a pressure sensor, and a current sensor attached to each driving component; a data collection unit that synchronizes and collects data from the sensor module unit; a data preprocessing unit that converts the collected data into a learnable form by performing noise removal, normalization, and outlier removal; an AI analysis unit that performs anomaly detection and probability prediction using the preprocessed data as input; an integrated prediction unit that calculates anomaly probability, anomaly grade, and replacement necessity by integrating the AI ​​analysis results; and a visualization and alarm unit that visually displays the integrated prediction results on a dashboard screen, provides visual and auditory alarms when an abnormal sign occurs, and stores the alarm occurrence history.

[0009] In one or more embodiments, the AI ​​analysis unit includes a CNN-LSTM model, extracts spatial features of sensor data through a CNN layer, and learns temporal correlations through an LSTM layer to calculate anomaly probability and anomaly score.

[0010] In one or more embodiments, the integrated prediction unit classifies the abnormality grade by part into "Normal," "Caution," and "Alert" stages based on the abnormality score and probability input from the AI ​​analysis unit, and generates an alarm through the visualization and alarm unit when the abnormality probability exceeds a threshold.

[0011] In one or more embodiments, the visualization and alarm unit displays the location of an anomaly, the probability of an anomaly, the level of an anomaly, and the status change over time of each driving unit in real time through a dashboard screen or GUI, and automatically saves the alarm occurrence history to a local storage device.

[0012] The present invention relates to a method for predicting anomalies in a semiconductor wafer transfer device, comprising: (a) collecting vibration, temperature, noise, pressure, and current data in real time from a plurality of sensors attached to a driving component; (b) preprocessing the collected data to remove noise and normalize it; (c) inputting the preprocessed data into a CNN-LSTM model to calculate an anomaly score and an anomaly probability; (d) classifying an anomaly grade according to the anomaly probability and generating an alarm; (e) visually displaying the anomaly prediction results on a dashboard screen; and (f) calculating the remaining life (RUL) based on the anomaly probability and component degradation information. Effects of the invention

[0013] According to the present invention, abnormal signs can be predicted early by integrating and analyzing multiple sensor data using an AI model (CNN-LSTM). Furthermore, according to the present invention, the accuracy of the prediction is improved by removing noise and outliers through a data preprocessing step. Additionally, according to the present invention, the integrated prediction unit can automatically classify status grades and suggest maintenance priorities. Moreover, according to the present invention, real-time status can be monitored and immediate response enabled through a visualization and alarm unit. Brief explanation of the drawing

[0014] FIG. 1 is a schematic diagram illustrating an exemplary configuration of a semiconductor wafer vision inspection device. FIG. 2 is a block diagram illustrating the overall configuration of an AI-based semiconductor transfer device abnormality prediction system according to the present invention. Figure 3 is a diagram illustrating the CNN-LSTM model architecture in an AI-based semiconductor transfer device anomaly prediction system according to the present invention. Figure 4 is a diagram illustrating a data flow diagram in an AI-based semiconductor transfer device anomaly prediction system according to the present invention. FIG. 5 is a diagram illustrating the AI ​​prediction logic flow in an AI-based semiconductor transfer device abnormality prediction system according to the present invention. Specific details for implementing the invention

[0015] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

[0016] The present inventions are provided to more fully explain the invention to those skilled in the art, and the following examples may be modified in various different forms, and the scope of the invention is not limited to the following examples. Rather, these examples are provided to make the disclosure more faithful and complete and to fully convey the spirit of the invention to those skilled in the art.

[0017] Additionally, in the drawings below, the thickness or size of each layer is exaggerated for convenience and clarity of explanation, and like reference numerals in the drawings refer to like elements. As used herein, the term "and / or" includes any one of the listed items and all combinations of one or more thereof. Furthermore, in this specification, the meaning of "connected" refers not only to cases where Member A and Member B are directly connected, but also to cases where Member C is interposed between Member A and Member B so that Member A and Member B are indirectly connected.

[0018] The terms used herein are for describing specific embodiments and are not intended to limit the invention. As used herein, the singular form may include the plural form unless the context clearly indicates otherwise. Additionally, as used herein, "comprise, include" and / or "comprising, including" specify the presence of the mentioned features, numbers, steps, actions, parts, elements, and / or groups thereof, and do not exclude the presence or addition of one or more other features, numbers, actions, parts, elements, and / or groups.

[0019] Although terms such as "first," "second," etc. are used in this specification to describe various components, parts, regions, layers, and / or parts, it is obvious that these components, parts, regions, layers, and / or parts should not be limited by these terms. These terms are used solely to distinguish one component, part, region, layer, or part from another region, layer, or part. Accordingly, the first component, part, region, layer, or part described below may refer to the second component, part, region, layer, or part without departing from the teachings of the present invention.

[0020] Spatial terms such as "beneath," "below," "lower," "above," and "upper" may be used to facilitate understanding of one element or feature depicted in the drawings and another element or feature. These spatial terms are intended to facilitate understanding of the invention according to various process or usage conditions of the invention and are not intended to limit the invention. For example, if an element or feature in the drawings is inverted, an element or feature described as "beneath" or "below" becomes "upper" or "on top." Therefore, "below" is a concept that encompasses "upper" or "below."

[0021] In addition, the sensor module, data collection unit, data preprocessing unit, AI analysis unit, integrated prediction unit, and visualization and alarm unit and / or other related devices or components according to the present invention may be implemented using any suitable hardware, firmware (e.g., application-specific semiconductor), software, or a suitable combination of software, firmware, and hardware. For example, various components of the control unit (controller) and / or other related devices or components according to the present invention may be formed on a single integrated circuit chip or on separate integrated circuit chips. In addition, various components of the control unit (controller) may be implemented on a flexible printed circuit film, or may be formed on a tape carrier package, a printed circuit board, or on the same substrate as the control unit (controller). In addition, various components of the control unit (controller) may be processes or threads executed on one or more processors in one or more computing devices, which may execute computer program instructions and interact with other components to perform the various functions mentioned below. Computer program instructions are stored in memory that can be executed on a computing device using a standard memory device, such as, for example, random access memory. Computer program instructions may also be stored on other non-transitory computer-readable media, such as, for example, CD-ROMs, flash drives, etc. Furthermore, those skilled in the art relating to the present invention should recognize that the functions of various computing devices may be combined with one another, integrated into a single computing device, or the functions of a specific computing device may be distributed across one or more other computing devices without departing from exemplary embodiments of the present invention.

[0023] FIG. 1 is a schematic diagram illustrating an exemplary configuration of a semiconductor wafer vision inspection device (100) according to one embodiment of the present invention. As shown in FIG. 1, the semiconductor wafer vision inspection device (100) may be configured to include, for example, a wafer loading unit (110), a wafer transfer unit (120), and a wafer inspection unit (130).

[0024] The wafer loading section (110) serves to load wafers brought in from the outside into the device and is equipped with a light curtain (111) to detect external intrusion or access during operation. The light curtain (111) may be composed of, for example, a Keyence GL-S40FH sensor and optically detects the approach of a worker or the passage of an object within the loading section to prevent safety accidents.

[0025] The wafer transfer unit (120) includes a wafer transfer robot (120a) and a wafer aligner (120b) to align and transfer a wafer transferred from the loading unit (110) onto the stage of the inspection unit (130). The transfer robot (120a) is capable of precise position control through a servo motor drive shaft and an LM guide structure, and the wafer aligner (120b) performs alignment based on the orientation mark of the wafer.

[0026] The inspection unit (130) includes a vision inspection module (131), an inspection stage (132), and a safety door (133). The vision inspection module (131) is equipped with a high-resolution camera and a lighting system to inspect the wafer's pattern, contamination, cracks, etc., and the inspection stage (132) has a vacuum suction structure to fix the position of the wafer. Additionally, the safety door (133) includes a door sensor that detects the external opening / closing state to prevent unauthorized opening during device operation.

[0028] FIG. 2 is a diagram illustrating the overall block configuration of an AI-based semiconductor transfer device anomaly prediction system (200) according to an embodiment of the present invention. As shown in FIG. 2, the AI-based semiconductor transfer device anomaly prediction system (200) according to the present invention is composed of a sensor module unit (210), a data collection unit (220), a data preprocessing unit (230), an AI analysis unit (240), an integrated prediction unit (250), and a visualization and alarm unit (260), and the data flow between each component is sequentially connected so that anomaly prediction results are produced in real time.

[0029] The sensor module (210) is composed of a plurality of sensors that detect vibration, temperature, noise, pressure, and current. The sensors are each attached to the main drive units of the semiconductor wafer transfer device, such as LM guides, servo motors, robot arms, and vacuum pumps, to detect status data generated during equipment operation in real time. The detected data is converted into an analog signal or a digital signal and transmitted to a subsequent data collection unit (220). With this configuration, the operational characteristics of each drive unit can be detected in a multidimensional manner, enabling early detection of abnormal signs.

[0030] The data collection unit (220) includes an IO module and an embedded device, and collects data acquired from each sensor from the sensor module unit (210) in a real-time streaming manner. The data collection unit (220) aligns the data collected from each sensor based on time and aligns it to a common time axis even if the time of occurrence of the data is different, and performs a buffering function to prevent data loss due to differences in sampling cycles. According to this configuration, temporal synchronization of data between sensors is ensured, and the aligned real-time data flow is stably transmitted to the next stage, the data preprocessing unit (230).

[0031] In the data preprocessing unit (230) above, noise removal, outlier correction, and normalization are performed on the collected data. The preprocessing process is an essential step for improving the accuracy of artificial intelligence (AI) analysis, and it maintains uniform data quality by minimizing distortion of sensor signals and measurement errors. The preprocessed data is then branched into two paths and transmitted in parallel to the AI ​​analysis unit (240) and the integrated prediction unit (250). With this configuration, temporal synchronization and quality correction are performed simultaneously during the data collection and preprocessing processes, thereby improving the accuracy of anomaly prediction.

[0032] The AI ​​analysis unit (240) includes an artificial intelligence (AI) model based on CNN-LSTM, learns the time-series characteristics of sensor data to detect abnormal signs, and calculates an anomaly probability for each component. The AI ​​analysis unit (240) decomposes the input multivariate sensor data into feature units, extracts local patterns and variation characteristics of sensor signals using a Convolutional Neural Network (CNN) layer, and analyzes long-term dependencies within the time-series data using a Long Short-Term Memory (LSTM) layer. Through this, the possibility of anomaly occurrence can be quantified by converting the change from a normal state to an abnormal state into a quantified probability value. With this configuration, the AI ​​analysis unit (240) enables precise anomaly detection by simultaneously considering individual patterns for each sensor and continuous changes on the time axis.

[0033] The integrated prediction unit (250) integrates the results transmitted from the data preprocessing unit (230) and the AI ​​analysis unit (240) to comprehensively calculate the remaining useful life (RUL) and the probability of anomalies for each component. Based on the AI ​​analysis results, the integrated prediction unit (250) classifies the condition of the component into stages of "Normal," "Caution," and "Alert," and if the anomaly grade exceeds a preset threshold, it transmits an alarm signal to the visualization and alarm unit (260) to trigger an alarm. Additionally, the integrated prediction unit (250) automatically records the calculated prediction results in a local storage device (Local DB), thereby enabling history-based condition tracking and long-term performance analysis, and the stored data can be used for retraining the AI ​​analysis model. With this configuration, the integrated prediction unit (250) can support precise maintenance judgments by simultaneously considering the possibility of anomalies occurring in the component and the state of deterioration progression.

[0034] The visualization and alarm unit (260) displays the anomaly prediction results output from the integrated prediction unit (250) in real-time in the form of a dashboard, and provides visual and auditory alarms when signs of an anomaly are detected. The visualization and alarm unit (260) includes a graphical user interface (GUI) and displays information such as the probability of anomaly occurrence, status level for each part, and predicted remaining lifespan (RUL) in the form of charts, gauges, color displays, or alarm icons. Accordingly, the operator can intuitively recognize the status of each drive unit and simultaneously check the time of anomaly occurrence and the deterioration trend of the corresponding part. In addition, the displayed data is automatically recorded in a local storage device in the form of a real-time log, which can be used as reference data for history-based maintenance decisions or for future model retraining. With this configuration, the visualization and alarm unit (260) can maximize the response efficiency of the equipment operator by providing AI analysis results in an intuitive and immediate form.

[0035] With this configuration, the system forms a closed-loop structure in which the entire process of data collection, preprocessing, artificial intelligence analysis, prediction calculation, and result display is linked in real time. Accordingly, when abnormal signs occur, analysis results are immediately visualized, allowing operators to respond without delay, and the learning efficiency of the prediction model is improved as the data flow is cyclically updated. As a result, the system can simultaneously improve the accuracy and responsiveness of fault prediction, thereby achieving stable operation and maintenance efficiency for semiconductor wafer transfer devices.

[0037] FIG. 3 is a diagram illustrating the CNN-LSTM model architecture in an AI-based semiconductor transfer device anomaly prediction system according to the present invention. As shown in FIG. 3, the CNN-LSTM model consists of an input layer, a CNN layer, an LSTM layer, and an output layer, and processes time-series data collected from multiple sensors step by step to predict anomaly signs with quantified probabilities.

[0038] First, the Input Layer receives multimodal sensor data containing different physical quantities such as vibration, temperature, noise, pressure, and current. The input data is aligned by time to form a sequence vector, and data for each sensor channel is passed to the CNN layer.

[0039] Next, the CNN layer plays the role of extracting spatial features from the input time-series data. This layer learns local patterns within the data, such as vibration frequency fluctuations, the shape of the temperature rise curve, and abnormal fluctuations in the current waveform, through convolution filters. The output of the CNN layer is converted into a multidimensional feature map and input into the LSTM layer.

[0040] Subsequently, the LSTM layer learns temporal patterns. The LSTM (Long Short-Term Memory) structure efficiently models long-term dependencies between data by utilizing an internal gate mechanism to remember data from past time points and forget unnecessary information. Accordingly, it is possible to detect cumulative anomalies that are not identified at a single time point.

[0041] Finally, the output layer calculates the anomaly probability and remaining useful life (RUL) of each component based on the time series features learned in the LSTM layer. The output result is transmitted to the integrated prediction unit (250) and used for classifying condition grades by component and establishing maintenance plans.

[0042] With this configuration, the CNN-LSTM model enables anomaly detection and lifespan prediction with higher accuracy than simple statistics-based prediction models by simultaneously reflecting the spatial patterns and temporal correlations of sensor data.

[0044] FIG. 4 is a diagram illustrating a data flow diagram in an AI-based semiconductor transfer device anomaly prediction system according to the present invention. This diagram represents the overall data processing flow of collecting, preprocessing, analyzing, and visualizing sensor data generated from the drive unit of a semiconductor wafer transfer device. As shown in FIG. 4, the data flow includes a multi-sensor data collection step (S410), a preprocessing and normalization step (S420), a CNN-LSTM analysis step (S430), a component-wise prediction step (S440), an integrated prediction step (S450), and a dashboard visualization and alarm step (S460).

[0045] First, in the multi-sensor data collection step (S410), status data such as vibration, temperature, noise, pressure, and current is acquired in real time by the sensor module (210). The data acquired from each sensor is sorted by time and integrated by the data collection unit (220). In this step, buffering processing is performed to synchronize data between sensors and prevent data loss.

[0046] Next, in the preprocessing and normalization step (S420), noise included in the collected data is removed, and normalization is performed to correct for differences in data scales per sensor. In this step, outlier removal and missing data correction are performed together, so that the data quality is homogenized into a form suitable for AI analysis.

[0047] Subsequently, in the CNN-LSTM analysis step (S430), the preprocessed time series data is input into the AI ​​analysis unit (240), and spatial features and temporal patterns are learned sequentially. The CNN layer extracts local variation characteristics, and the LSTM layer learns long-term dependency relationships to quantitatively identify complex anomalies.

[0048] Next, in the component-wise prediction step (S440), an anomaly score and an anomaly probability are calculated for each drive unit. Based on the AI ​​analysis results, the state of each component is classified as normal or abnormal, and if the anomaly level of each component exceeds a set threshold, the result is transmitted to the integrated prediction unit (250).

[0049] In the Integrated Prediction stage (S450), the prediction results for each component are aggregated to calculate the Remaining Useful Life (RUL) and Anomaly Probability. This stage calculates the Health Index for the entire system by comprehensively considering the degradation status of individual components, and the results are automatically recorded in the local database.

[0050] Finally, in the Dashboard Visualization & Alarm stage (S460), the integrated prediction results are displayed in real-time in the form of a dashboard, and visual and auditory alarms are triggered when abnormal signs are detected. Through the dashboard, operators can intuitively check the condition grade, probability of anomaly, and predicted remaining life of each part, and immediately identify the time of anomaly to perform maintenance measures.

[0051] With this configuration, the data flow illustrated in FIG. 4 automates the entire process from the collection of sensor data to visualization, allowing for the real-time reflection of anomaly prediction results without delay in data flow. Therefore, the present invention can significantly improve the anomaly detection accuracy and response speed of a semiconductor transfer device.

[0053] FIG. 5 is a diagram illustrating the AI ​​prediction logic flow of an AI-based semiconductor transfer device anomaly prediction system according to the present invention. This logic represents a series of processing procedures that calculate anomaly probability in real time based on sensor data collected from a semiconductor wafer transfer device, predict the remaining lifespan (RUL), and automatically derive a maintenance plan. As illustrated in FIG. 5, the AI ​​prediction logic flow includes a Real-time Data Input step (S510), a Component-wise Model Analysis step (S520), an Anomaly Probability Calculation step (S530), a RUL Estimation step (S540), a Maintenance Priority Decision step (S550), and an Alarm & Maintenance Plan Generation step (S560).

[0054] First, in the Real-time Data Input step (S510), multi-sensor data acquired from the sensor module unit (210) and the data collection unit (220) is input in real time to the AI ​​analysis unit (240). In this step, the noise removal and normalization results performed in the preprocessing unit (230) are reflected and supplied in a data form suitable for the AI ​​analysis model.

[0055] Next, in the component-wise model analysis step (S520), an AI analysis model corresponding to each actuator operates independently to identify abnormal patterns in the sensor data. In this step, a CNN-LSTM-based sub-model learns individual features of each component, such as vibration, temperature, and noise, to detect signs of abnormality, and transmits the corresponding signal to the abnormality probability calculation unit.

[0056] In the Anomaly Probability Calculation step (S530), the anomaly probability for each part is calculated based on the AI ​​analysis results. In this step, the possibility of an anomaly occurring is quantified as a probability value between 0 and 1 by evaluating the pattern similarity of the current data compared with normal state data. The calculated anomaly probability is combined with threshold judgment logic and used as a criterion for determining whether to generate a warning.

[0057] Next, in the remaining life prediction step (RUL Estimation) (S540), the remaining life (RUL, Remaining Useful Life) of each component is estimated based on the probability of anomalies and the rate of component degradation. The RUL can be calculated in units of time or the number of operations, and the prediction result is stored in the integrated prediction unit (250) and used to calculate the maintenance cycle of the equipment.

[0058] In the Maintenance Priority Decision step (S550), the maintenance priority is automatically determined by comparing and evaluating the abnormality probability and remaining life data for each part. For example, parts with a high abnormality probability and a short remaining life are designated as priority replacement targets, and this result is transmitted to the visualization and alarm unit (260).

[0059] Finally, in the Alarm & Maintenance Plan Generation step (S560), an alarm signal is generated according to the determined priority, and the timing for maintenance or replacement of the corresponding part is automatically planned. The generated maintenance plan is displayed on the dashboard screen, and the operator can perform maintenance measures immediately after confirming the alarm.

[0060] With this configuration, the AI ​​prediction logic flow automates the entire process from the input of sensor data to the derivation of maintenance plans, thereby simultaneously improving the speed of failure prediction and the efficiency of maintenance decision-making.

[0062] With this configuration, the present invention enables the real-time evaluation of component degradation status and anomaly probability by integrating and analyzing multi-sensor data from a semiconductor wafer transfer device based on artificial intelligence (AI). Accordingly, predictive maintenance becomes possible compared to existing maintenance systems centered on periodic inspections or reactive repairs, and production stoppages caused by sudden failures can be effectively prevented. Furthermore, by simultaneously considering spatial features and temporal patterns through a CNN-LSTM-based AI analysis structure, the accuracy of anomaly detection and prediction reliability are significantly improved compared to existing technologies that relied on single-sensor signal analysis. In addition, the entire process of data collection, analysis, prediction, and display is linked in real-time by a closed-loop system architecture including an integrated prediction unit and a visualization / alarm unit, allowing operators to take immediate response measures. Moreover, since prediction results are automatically recorded in a local database, they can be utilized for long-term performance analysis and AI model retraining. Therefore, the present invention provides the effect of maximizing production line stability and operational efficiency by simultaneously improving the failure prediction accuracy, responsiveness, and maintenance efficiency of semiconductor manufacturing equipment.

[0064] The above description is merely one embodiment for implementing an exemplary AI-based semiconductor transfer device abnormality prediction system and method according to the present invention. The present invention is not limited to the above-described embodiment, and the technical spirit of the present invention extends to the scope in which various modifications can be made by anyone with ordinary knowledge in the field to which the invention belongs, without departing from the gist of the invention as claimed in the following patent claims. Explanation of the symbols

[0065] 100; Semiconductor wafer vision inspection device 110; Wafer loading section 111; Light curtain 120; Wafer transfer unit 120a; Wafer transfer robot 120b; Wafer aligner 130; Wafer inspection unit 131; Vision inspection module 132; Inspection stage 133; Safety Door 200; AI-based Anomaly Prediction System 210; Sensor module unit 220; Data acquisition unit 230; Data Preprocessing Department 240; AI Analysis Department 250; Integrated Prediction Unit 260; Visualization and Alarm Unit

Claims

Claim 1 A system for collecting status data in real time from multiple sensors attached to drive components of a semiconductor wafer transfer device and predicting abnormal signs through artificial intelligence (AI) analysis of said data, comprising: a sensor module unit including vibration sensors, temperature sensors, noise sensors, pressure sensors, and current sensors attached to each drive component; a data collection unit that collects data from the sensor module unit by synchronizing it on a time basis and collecting it in a real-time streaming manner; a data preprocessing unit that converts the collected data into a learnable form by performing noise removal, normalization, and outlier removal; an AI analysis unit that performs anomaly detection and probability prediction using the preprocessed data as input - the AI ​​analysis unit includes a CNN-LSTM model that takes the preprocessed multivariate time series data as input, extracts spatial features of sensor data through a CNN layer, and learns temporal correlations through an LSTM layer, and calculates anomaly probability and anomaly score for each drive component -; an integrated prediction unit that calculates anomaly probability, anomaly grade, and replacement necessity by integrating the AI ​​analysis results - the integrated prediction unit determines the state of each drive component as "Normal", based on the anomaly probability and anomaly score from the AI ​​analysis unit Classifying into "Caution" and "Alert," calculating the Remaining Life (RUL) based on the above abnormality probability and component degradation information, determining maintenance priorities based on the above abnormality probability and remaining life, and generating an alarm through the visualization and alarm unit when the abnormality probability exceeds a threshold - ; and the visualization and alarm unit that visually displays the integrated prediction results from the above integrated prediction unit on a dashboard screen in real time, provides visual and auditory alarms when abnormal signs occur, and saves the alarm occurrence history - the visualization and alarm unit displays the location of abnormality occurrence, abnormality probability, abnormality grade, and status changes over time of each drive component in real time through a dashboard screen or GUI, and automatically saves the alarm occurrence history to a local storage device - ;AI-based semiconductor transfer device anomaly prediction system including Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete

Images