Three-dimensional intelligent production and inspection method and system

Abstract

According to the three-dimensional intelligent production and inspection method and system, automation, intelligence and precision of the production process are achieved through cooperative work of data acquisition, intelligent control, robot operation and a man-machine interaction function unit, the production efficiency is improved, the production cost is reduced, and the product quality is guaranteed. Specifically, production data are collected in real time and preprocessed; performing anomaly detection on the preprocessed data; when the data is normal and the performance evaluation meets the requirement, calculating an optimal decision vector, including estimating the state of the current moment by using the current production data and the state of the previous moment; the digital twin model predicts a state vector at the next moment by using the state estimation at the current moment and historical data of the state estimation; taking the state vector of the next moment as a target, and calculating an optimal decision vector through decision analysis; estimating future system state and performance corresponding to the optimal decision vector through predictive analysis; and if the prediction analysis reaches the standard, executing the optimal decision vector.

Description

Technical Field

[0001] The present invention relates to the field of automated production and quality control, and specifically to an intelligent manufacturing solution that combines digital twin technology and three-dimensional visualization technology, aiming to achieve digitization, intelligence and visualization of factory production processes. Background Art

[0002] With the advent of the Industrial 4.0 era, intelligent manufacturing has become the key to the transformation and upgrading of the manufacturing industry. Traditional production methods and inspection methods can no longer meet the high efficiency, precision and intelligence requirements of modern production. Therefore, the development of a device and system that can achieve three-dimensional intelligent production and inspection is of great significance for improving production efficiency, reducing production costs and ensuring product quality. Summary of the invention

[0003] The purpose of the present invention is to provide a three-dimensional intelligent production and inspection method and system, which realizes the automation, intelligence and precision of the production process through the collaborative work of data acquisition, intelligent control, robot operation, and human-computer interaction functional units, improves production efficiency, reduces production costs, and ensures product quality.

[0004] The specific technical solutions are as follows:

[0005] A three-dimensional intelligent production and inspection method, comprising:

[0006] Collect production data in real time and preprocess it; perform anomaly detection on the preprocessed data; perform performance evaluation when the data is normal; when the performance evaluation meets the requirements, calculate the optimal decision vector, including:

[0007] State estimation: Use the preprocessed production data at the current moment and the state estimation at the previous moment to estimate the state at the current moment

[0008] Will The historical data of state estimation is input into the digital twin model to predict the state vector at the next moment.

[0009] Decision analysis: As the goal, the optimal decision vector is calculated through decision analysis;

[0010] Predictive analysis: Estimate the future system state and performance based on the state vector x'(t) predicted by the optimal decision vector and the state history data;

[0011] Execute decision: If the prediction analysis meets the criteria, the optimal decision vector is executed.

[0012] Furthermore, the production data includes the current assembly progress of the product, the position of each component of the product, the temperature of the installation area, the installation angle, and the installation pressure.

[0013] Furthermore, anomaly detection uses a density-based local anomaly factor algorithm, with the following specific processes: a) collecting historical data to form a benchmark data set for normal operation; b) calculating the LOF score for real-time preprocessed production data; c) if the LOF score exceeds a threshold T, it is marked as an anomaly;

[0014] Furthermore, the performance evaluation uses a weighted summation method to comprehensively evaluate multiple performance indicators, including: a) defining a set of performance indicators; b) scoring each indicator; c) calculating a weighted total score.

[0015] Furthermore, the digital twin model adopts the LSTM model.

[0016] Furthermore, decision analysis includes: a. defining the decision space according to the actual situation, which includes control variables; b. simulating the impact of each decision; c. evaluating the expected effect of each decision; d. considering uncertainty; e. multi-objective optimization to balance the needs of quality and efficiency and obtain the optimal decision vector.

[0017] Furthermore, an extended Kalman filter is used for state estimation, where the state vector is defined based on production data.

[0018] Furthermore, predictive analytics estimates future system states and performance through time series prediction models.

[0019] A three-dimensional intelligent production and inspection system, including a data acquisition unit, an intelligent control unit, a robot operation unit and a human-machine interaction unit;

[0020] The data acquisition unit is responsible for the acquisition and preprocessing of production data, including three-dimensional visual recognition and sensor acquisition. The former is responsible for the acquisition of three-dimensional image data of the product, and captures the three-dimensional image data in real time through the image processor. The latter acquires the production status and environmental data of the product through designated sensors, including but not limited to temperature sensors, humidity sensors, pressure sensors, and voltage sensors.

[0021] The intelligent control unit is the core of the whole system, which is responsible for receiving all real-time production data from the data acquisition unit and the data from the human-machine interaction unit, and finally issuing control instructions to the robot unit to perform production operations, while outputting the three-dimensional production process of the product to the human-machine interaction unit in real time to realize production monitoring and decision support; the intelligent control unit integrates anomaly detection module, performance evaluation module, state estimation module, digital twin module, decision analysis module, prediction analysis module and execution decision module;

[0022] The robot operation unit is the execution part of the system, which is responsible for completing the operation of the product according to the instructions of the intelligent control unit. The unit adopts an industrial robot and a matching control system. The control system receives instructions from the intelligent control unit and realizes precise position and posture control by controlling the joints and drivers of the robot.

[0023] The human-computer interaction unit is an interface for the system to interact with the product manager, operator and inspector. It receives product information input by the product manager, operating instructions input by the operator and product quality standards input by the inspector, and displays the production status and progress of the product output by the intelligent control module in real time, and displays the product operating instruction log and product production performance. The unit includes a product information input module, a production progress viewing module and a quality inspection module. The information input module provides three parts of information input functions: product information input by the product manager, operating instructions input by the operator and product quality standards input by the inspector, which are used for anomaly detection, performance evaluation and decision analysis of the intelligent control unit; the production progress viewing module receives the production information output by the data acquisition module and the operating instructions, anomaly detection and performance evaluation data output by the intelligent control module, and displays them in real time through two-dimensional graphs, tables, logs and three-dimensional products, supporting users to view the production progress of three-dimensional products in real time through VR glasses.

[0024] Beneficial Effects

[0025] 1. Improve production efficiency: By capturing and processing the three-dimensional data of products in real time, the intelligent control unit can accurately guide the robot operation unit to perform operations such as grabbing, moving and assembling products. The construction of the digital twin model makes every link in the production process more precise and efficient, reducing waste and downtime in the production process, thereby improving overall production efficiency.

[0026] 2. Optimize product quality: The digital twin model can reflect the actual status of the product in real time, including changes in size, shape, and performance. By comparing the differences between the model and the actual product, companies can promptly identify potential problems in the production process and take corresponding measures to improve them. In addition, the application of quality inspection units also further ensures the stability and consistency of product quality and reduces the proportion of defective products.

[0027] 3. Reduce production costs: Through the construction of digital twin models, enterprises can conduct virtual simulation and prediction before production to avoid unnecessary errors and losses in the actual production process. This helps to reduce material waste, energy consumption, and downtime and maintenance costs caused by production problems. In addition, by optimizing production processes and improving production efficiency, enterprises can also reduce labor costs and time costs.

[0028] 4. Improve decision-making support capabilities: The digital twin model provides rich real-time data and analysis results, providing enterprise decision makers with powerful decision-making support capabilities. Through the analysis of model data, decision makers can gain an in-depth understanding of the operation status and bottleneck problems of the production process and formulate more reasonable production plans and resource allocation strategies. In addition, the model can also predict future production trends and market demand, helping enterprises make more accurate strategic decisions.

[0029] 5. Enhance the visualization and transparency of the production process: The digital twin model makes all aspects of the production process visualized, and enterprises can understand information such as production progress, equipment status and product quality in real time. This helps enterprises to promptly discover and solve problems in the production process and improve the controllability and stability of the production process. At the same time, a transparent production process also helps to enhance the image and credibility of the enterprise and enhance customer trust in the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 , block diagram of the three-dimensional intelligent production and inspection system.

[0031] Figure 2 , 3D intelligent production and inspection method flow chart DETAILED DESCRIPTION

[0032] The present invention is achieved through the following technical solutions.

[0033] A three-dimensional intelligent production and inspection device includes a data acquisition unit, an intelligent control unit, a robot operation unit and a human-machine interaction unit. Physical connection and data transmission are achieved between the units through specific interfaces, protocols and hardware devices.

[0034] The data acquisition unit is responsible for the acquisition and preprocessing of product data, product status and environment data during production, and is an important source of input data for the intelligent control unit. The unit includes two parts: three-dimensional visual recognition and sensor acquisition. The former is responsible for the acquisition of three-dimensional image data of the product. It captures three-dimensional image data (including point cloud data, texture information, etc.) in real time through a high-performance image processor, and identifies the type, size, angle, position and other information of product parts after real-time processing and analysis. The latter collects data such as product production status and environment through designated sensors, including but not limited to temperature sensors, humidity sensors, pressure sensors, voltage sensors, etc. The unit is connected to the intelligent control unit through a high-speed interface (including but not limited to Gigabit Ethernet, USB3,0) to ensure fast and stable data transmission. In addition, the unit also uses standard communication protocols such as TCP / IP or UDP to ensure reliable data transmission.

[0035] The intelligent control unit is the core of the whole device, which is responsible for receiving all real-time production data, historical production data (production records of the past 6 months) from the data acquisition unit, and the production plan (target output, product model, delivery date) and quality standard data (accuracy, efficiency, yield rate, energy consumption, etc.) from the human-machine interaction unit. After analysis and decision-making, it issues corresponding control instructions to the robot unit to perform production operations, and outputs the three-dimensional production process of the product to the human-machine interaction unit in real time to achieve efficient production monitoring and decision support. The intelligent control unit integrates a state estimation module, anomaly detection module, performance evaluation module, state estimation module, digital twin module, decision analysis module, prediction analysis module and execution decision module. The digital twin model uses the current state estimation and the historical data of state estimation as input to digitally model the physical production. The digital twin model is a 1:1 true restoration three-dimensional simulation model that is maintained in real time by the system and driven by actual production data. The decision analysis module takes the state vector predicted by the digital twin model at the next moment as the target, and calculates the optimal decision vector through decision analysis; the performance evaluation module aims to quantify the performance of the current production process, and uses the weighted summation method to comprehensively evaluate multiple performance indicators, which are output to the quality inspection module of the human-computer interaction unit, making it convenient for inspectors to view product production performance in real time, which is the intelligence and digitization of the inspection process.

[0036] Specifically, the intelligent control unit receives the production data from the data acquisition unit and performs anomaly detection on the data. If an anomaly occurs, the anomaly is output to the human-machine interaction unit for manual processing by the operator, and the operator inputs an operation instruction to the robot operation unit through the human-machine interaction unit. If no anomaly occurs, a performance evaluation is performed. If the evaluation result meets the quality standard input by the human-machine interaction unit, the optimal decision vector is calculated, and finally the optimal decision instruction is input to the robot operation unit for decision execution. If the evaluation result does not meet the quality standard input by the human-machine interaction unit, it is output to the human-machine interaction unit for manual processing by the operator, and the operator inputs an operation instruction to the robot operation unit through the human-machine interaction unit.

[0037] The robot operation unit is the execution part of the device, which is responsible for accurately completing the operations of grabbing, moving, assembling and so on of the product according to the instructions of the intelligent control unit, so as to improve the production efficiency and quality. The unit adopts a high-precision industrial robot and a matching control system. The control system receives instructions from the intelligent control unit and realizes precise position and posture control by controlling the joints and drivers of the robot. In order to realize the high-reliability robot operation, the unit adopts industrial-grade communication interface and protocol (including but not limited to EtherCAT, CAN bus or PROFINET), which ensures the real-time and stability of data transmission and meets the requirements of robot operation for accuracy and speed. The unit is connected to the intelligent control unit through the interface control line and communicates with the intelligent control unit in two directions. The intelligent control unit sends control instructions to the robot operation unit, such as grabbing position, moving speed, etc., to instruct the robot to perform the corresponding operation. The robot operation unit sends the executed operation instructions to the intelligent control unit for the storage of production data and the decision analysis of the intelligent control unit.

[0038] The human-computer interaction unit is an interface for the system to interact with the product manager, operator and inspector, and receives product information (model X, production plan Q = {q_production quantity, q_production cycle, q_delivery date, ...}, production process L = {l n}, etc.), the operation instructions input by the operator, and the product quality standard P = {p_precision, p_speed, p_yield, ..., p_energy consumption} input by the inspector, and display the production status and progress of the product output by the intelligent control module in real time, display the product operation instruction log and product production performance. This unit includes a product information input module, a production progress viewing module, and a quality inspection module. The information input module provides three parts of information input functions, product information (model, production plan, production process, etc.) input by the product manager, the operation instructions input by the operator, and the product quality standard input by the inspector, which are used for anomaly detection and performance evaluation of the intelligent control unit, respectively. The production progress viewing module receives the product production progress, product location, environmental information output by the data acquisition module, and the operation instructions, anomaly detection, and product quality evaluation data output by the intelligent control module, and displays them in real time through two-dimensional graphs, tables, logs, and three-dimensional products, supporting users to view the production progress of three-dimensional products in real time through VR glasses.

[0039] Assuming that the 3D production scene and 3D product model have been constructed, the operation process of the 3D intelligent production and inspection system is as follows:

[0040] Data acquisition and preprocessing: define the data acquisition unit to collect data S = {s_progress, s_temperature, s_angle, s_force, s_position} in real time, where s_progress and s_position are obtained by the three-dimensional recognition model, and s_temperature, s_angle and s_force are obtained by the sensor module. The data S is time-stamped to form D(t) = {d_progress(t), d_temperature(t), d_angle(t), d_force(t), d_position(t)} data, and the D(t) data is preprocessed to obtain clean, standardized, and information-rich data D'(t) = f_preprocess(D(t)). The preprocessing steps include data cleaning: using statistical methods (such as Z-score or IQR) to remove outliers, and using interpolation methods (such as linear interpolation or nearest neighbor interpolation) to handle missing data; data standardization: unifying data of different scales to the same range, using Min-Max scaling or Z-score method. Use the same method as above to preprocess the historical data H to obtain H'.

[0041] Anomaly detection: A(t) = f_anomaly(D'(t), H', T), using statistical methods and machine learning techniques to identify abnormal conditions in the production process, such as abnormal CPU installation angle, excessive pressure and other problems. Input: the real state D'(t) and historical operation data H' collected in real time in step 1, the abnormal threshold T entered by the operator through the human-computer interaction unit, and output whether there is an abnormality. We use the density-based local outlier factor (LOF) algorithm. Specific process: a) Collect historical data H' to form a benchmark data set for normal operation; b) Calculate the LOF score for the new observation data D'(t): calculate the k-nearest neighbor distance, calculate the local reachable density, and calculate the LOF score; c) If the LOF score exceeds the threshold T, it is marked as abnormal: A(t) = 1 if LOF(P(t))>T else 0. For example, if the LOF score of the CPU installation angle is found to be 1.8, which exceeds the preset threshold of 1.5, the system will issue an abnormal warning. If normal, proceed to the next step.

[0042] Performance evaluation: P(t) = f_evaluate(M(t), G), the performance evaluation step aims to quantify the performance of the current production process. Use the weighted summation method to comprehensively evaluate multiple performance indicators. Specific process: a) Define the performance indicator set P = {p_accuracy, p_speed, p_yield, p_energy consumption}; b) Score each indicator: s_accuracy = (1-|actual position-target position| / allowable error)*100, s_speed = (actual assembly time / target time)*100, s_yield = (number of products passing quality inspection / total number of products)*100, s_energy consumption = (target energy consumption / actual energy consumption)*100; c) Calculate the weighted total score: P_total = w_accuracy*s_accuracy+w_speed*s_speed+w_yield*s_yield+w_energy consumption*s_energy consumption, where w is the weight, reflecting the relative importance of each indicator. For example, if the accuracy score is 95, the speed score is 90, the yield score is 98, and the energy consumption score is 92, and the weights are 0.4, 0.3, 0.2, and 0.1 respectively, the total score is: 0.4*95+0.3*90+0.2*98+0.1*92=93.9. If the performance evaluation meets the requirements, proceed to the next step.

[0043] State estimation: Accurately infer the current state of the system based on real-time collected data. Input: preprocessed current moment collected data D'(t) and the state estimate of the previous moment Output: state estimate at the current moment This step is achieved using the Extended Kalman Filter Specific process: define the state vector x = [x_pos, y_pos, θ, F, T]', where (x_pos, y_pos) is the position of the product parts, θ is the installation angle, F is the installation force, and T is the temperature, which are obtained from d_position(t), d_angle(t), d_force(t), and d_temperature(t) in D'(t); prediction step: predict based on the state estimation and system model at the previous moment Where f is the nonlinear state transfer function, u is the control input, and w is the process noise; calculate the Kalman gain K(t) = P - (t)H(t) ’ (H(t)P - (t)H(t)'+R(t)) - 1, where P - (t) is the prior error covariance matrix, H(t) is the observation matrix, and R(t) is the observation noise covariance matrix; then, use the current observation data to correct this prediction: x^(t) = x - (t)+K(t)(z(t)-h(x -(t))), that is, the final estimate = predicted value + Kalman gain * (actual observation value - predicted observation value), where x^(t) is the final state estimate (posterior estimate), K(t) is the Kalman gain, z(t) is the actual observation value, h(x - (t)) is the observed value predicted based on the prior estimate. (Using the preprocessed current moment collected data D'(t) and the state estimate of the previous moment ), estimate the current state

[0044] Will The historical data of state estimation is input into the digital twin model to obtain the state estimation at the next moment. The digital twin model adopts the LSTM model.

[0045] Decision analysis: The purpose is to find the optimal production adjustment strategy through a systematic approach, taking into account various factors and uncertainties. It breaks down complex decision problems into manageable steps, uses mathematical models and calculation methods to simulate, evaluate and optimize decisions, and ultimately comes up with an optimal solution x'(t) = f_analysis(M(t),Ω) that can both improve product quality and ensure production efficiency. Specific process: a. Define decision space: Ω = {ω_adjust pressure, ω_adjust speed, ω_adjust temperature, ω_pause production}; b. Simulate the impact of each decision: for ω_adjust pressure: M_pressure(t+Δt) = f_simulate(M(t), ω_adjust pressure, Δt); c. Evaluate the expected effect of each decision: U_pressure = f_utility(M_pressure(t+Δt), G) to evaluate the impact of adjusting pressure on assembly quality and efficiency; d. Consider uncertainty: R(ω_adjust pressure) = ∫L(M_pressure(t+Δt), θ)p(θ|D)dθ, θ includes uncertain factors such as material elasticity change and environmental humidity; e. Multi-objective optimization: ω* = argmin_ω{w_quality*R(ω)-w_efficiency*U(ω)} to balance the needs of quality and efficiency and finally obtain the optimal decision vector.

[0046] Prediction analysis: The optimal decision vector ω* generated in the decision analysis phase is converted into a predicted state vector x'(t) by the extended Kalman filter method, and x'(t) is added to the historical data H' for prediction analysis F(t+Δt)=f_forecast(M(t),H'). The prediction analysis step uses a time series prediction model to estimate the future system state and performance, for example, to predict the assembly quality and efficiency within the next Δt time if the current operation continues. We use a long short-term memory network (LSTM) for prediction, F(t+Δt)=f_forecast(M(t),H'). Specific process: a) Prepare training data: Extract feature sequences and target sequences from historical data H; b) Train LSTM model: model=LSTM(input_shape=(sequence_length,n_features))model.fit(X_train,y_train,epochs=100,batch_size=32); c) Use the trained model for prediction: F(t+Δt)=model.predict(M(t)). For example, based on the current assembly status and historical data, the LSTM model may predict that the yield rate will drop by 2% and the energy consumption will increase by 5% after 2 hours.

[0047] Execution decision: If the system state and performance predicted in the prediction analysis phase meet the expected indicators (production quality and production efficiency), the decision instructions generated in the decision analysis phase are sent to the robot operation unit. For example, if ω* is the adjustment pressure, the pressure adjustment instruction C(t) = f_control(ω*) is sent to the robot operation unit. If the system state and performance predicted in the prediction analysis phase do not meet the expected indicators, continue the decision analysis.

[0048] Human-computer interaction and inspection process. Throughout the process, the system interacts with the operator through an intuitive visual interface, displays real-time status, prediction results, and decision-making suggestions, and allows the operator to make necessary interventions. At the same time, quality inspectors can view the three-dimensional status of the product in real time through VR glasses. The results of performance evaluation and anomaly detection reflect the product quality in the entire production process in real time. Quality inspectors can make qualification judgments based on the data output by the system and complete product inspection.

[0049] Iteration: The system iterates continuously to achieve continuous optimization, collecting data for each new production step; conducting a new round of decision analysis and prediction; and executing optimized operations. Through this closed-loop process, the system can continuously improve the quality and efficiency of product production, quickly respond to abnormal situations, and provide strong support for operators and quality inspectors.

[0050] The following is combined with reference Figure 1 It should be understood that the embodiments shown and described in the drawings are only exemplary and are intended to explain the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0051] Taking the assembly and inspection process of a laptop component as an example, the three-dimensional intelligent production and inspection device and system of this embodiment are described in detail as follows:

[0052] The three-dimensional intelligent production and inspection device includes a data acquisition unit, an intelligent control unit, a robot operation unit and a human-computer interaction unit.

[0053] Notebook component assembly 3D intelligent production and inspection system workflow:

[0054] (1) Initialization and data collection

[0055] When the system starts, the 3D production scene and the 3D model of the notebook product are loaded. The robots on the assembly line start working, preparing to install core components such as CPU, memory, and hard disk.

[0056] The data acquisition unit collects the following data in real time:

[0057] s_progress: obtain the current assembly progress through the 3D recognition model (such as CPU installation is 50% complete);

[0058] s_position: The 3D recognition model obtains the precise position coordinates of each component;

[0059] s_temperature: The sensor monitors the temperature of the CPU installation area;

[0060] s_angle: The sensor measures the CPU installation angle;

[0061] s_force: Sensor records CPU mounting pressure.

[0062] The collected raw data undergo the following data preprocessing: data cleaning: use the Z-score method to remove outliers, such as abnormally high installation force data; missing data processing: use linear interpolation to fill in the occasional missing sensor data; data standardization: unify data of different scales (such as temperature and pressure) into the range of 0-1.

[0063] (2) Anomaly Detection

[0064] Use the LOF algorithm to detect anomalies. For example, if the LOF score of the CPU installation angle exceeds the preset threshold of 1.5, the system will mark it as an anomaly and output it to the human-computer interaction unit. The operator will decide whether to handle the anomaly; if there is no anomaly, the execution will continue.

[0065] (3) Performance evaluation

[0066] The system evaluates the current production performance: accuracy: CPU installation position deviation; speed: actual assembly time vs. target time; yield rate: the proportion of CPUs that pass quality inspection; energy consumption: actual energy consumption vs. target energy consumption.

[0067] The system calculates a weighted total score, such as 93.9 points (out of 100), which reflects the overall production performance. If it meets the requirements, it will proceed to the next step.

[0068] (4) State Estimation

[0069] The extended Kalman filter is used to estimate the current system state. For example, based on the previous state and new observation data, the system estimates the current position, installation angle, pressure, and temperature of the CPU.

[0070] (5) The extended Kalman filter estimates the current system state and the historical data of state estimation are input into the digital twin model to predict the state vector at the next moment.

[0071] (6) Decision Analysis

[0072] The system considers the following decision options: ω_Adjust pressure: fine-tune CPU installation pressure; ω_Adjust speed: change robot installation speed; ω_Adjust temperature: adjust the temperature of the working area; ω_Pause production: pause the production line when a serious anomaly is found. The system simulates the impact of each decision, evaluates the expected effect, and considers uncertain factors (such as CPU batch differences). Finally, the system weighs quality and efficiency and selects the optimal decision.

[0073] (7) Predictive analysis

[0074] Based on the state vector x'(t) predicted by the optimal decision vector and the state history data, the LSTM model is used to estimate the future system state and performance. For example, based on the current CPU installation status and historical data, the system may predict the yield rate and energy consumption in 2 hours. If the yield rate and energy consumption are not improved compared to the current values, the decision is not executed and the decision analysis is repeated; otherwise, the decision is continued.

[0075] (8) Execution decision

[0076] If the predictive analysis shows that continuing the current operation can achieve the expected indicators, the system will execute the optimal decision previously selected. For example, send a command to the robot to slightly increase the CPU installation pressure by 2%.

[0077] Human-computer interaction and quality inspection

[0078] Throughout the process, operators monitor production status through a visual interface, view decision suggestions, and intervene when necessary.

[0079] Quality inspectors use VR glasses to view the three-dimensional status of CPU installation in real time. The system continuously outputs performance evaluation and abnormality detection results to assist quality inspectors in judging product eligibility.

[0080] The system iterates in a continuous loop to achieve continuous optimization: re-collect data for each newly installed CPU component; conduct a new round of decision analysis and prediction; and execute the optimized operation. Through this closed-loop process, the system can continuously improve the assembly quality and efficiency of notebook components, quickly respond to abnormal situations, and provide strong support for operators and quality inspectors.

[0081] The three-dimensional intelligent production and inspection device and system of this embodiment realizes comprehensive intelligent monitoring and management of the assembly and inspection process of notebook components by constructing a digital twin model. During the production process, the system can automatically decide on production operation instructions, dispatch production resources, and conduct real-time monitoring and adjustments. During the inspection process, the system ensures that the quality and performance of the notebook meet the standards through real-time anomaly detection and performance evaluation. The entire system improves production efficiency, reduces costs, and improves product quality and customer satisfaction.

Claims

1. A three-dimensional intelligent production and inspection method, characterized in that: Collect and pre-process production data in real time; Perform anomaly detection on preprocessed data; perform performance evaluation when the data is normal; When the performance evaluation meets the requirements, the optimal decision vector is calculated, including: ① State estimation: Use the preprocessed production data at the current moment and the state estimation at the previous moment to estimate the state at the current moment ② The historical data of state estimation is input into the digital twin model to predict the state vector at the next moment. ③Decision analysis: As the goal, the optimal decision vector is calculated through decision analysis; ④ Prediction analysis: Estimate the future system state and performance based on the state vector x'(t) predicted by the optimal decision vector and the state history data; ⑤ Execute decision: If the prediction analysis meets the requirements, the optimal decision vector will be executed.

2. A three-dimensional intelligent production and inspection method according to claim 1, characterized in that: The production data includes the current assembly progress of the product, the location of each component of the product, the temperature of the installation area, the installation angle, and the installation pressure.

3. A three-dimensional intelligent production and inspection method according to claim 1 or 2, characterized in that: Anomaly detection uses a density-based local anomaly factor algorithm. The specific process is: a) collect historical data to form a benchmark data set for normal operation; b) calculate the LOF score for real-time preprocessed production data; c) if the LOF score exceeds the threshold T, it is marked as an anomaly.

4. A three-dimensional intelligent production and inspection method according to claim 3, characterized in that: The performance evaluation uses a weighted summation method to comprehensively evaluate multiple performance indicators, including: a) defining a set of performance indicators; b) scoring each indicator; c) calculating a weighted total score.

5. A three-dimensional intelligent production and inspection method according to claim 4, characterized in that: The digital twin model adopts the LSTM model.

6. A three-dimensional intelligent production and inspection method according to claim 5, characterized in that: Decision analysis includes: a. defining the decision space according to the actual situation, which includes control variables; b. simulating the impact of each decision; c. evaluating the expected effect of each decision; d. considering uncertainty; e. multi-objective optimization to balance the needs of quality and efficiency and obtain the optimal decision vector.

7. A three-dimensional intelligent production and inspection method according to claim 6, characterized in that: The extended Kalman filter is used for state estimation, where the state vector is defined based on production data.

8. A three-dimensional intelligent production and inspection method according to claim 7, characterized in that: Predictive analytics estimates future system states and performance through time series forecasting models.

9. A three-dimensional intelligent production and inspection system, based on any method described in rights 1-8, characterized in that: It includes data acquisition unit, intelligent control unit, robot operation unit and human-machine interaction unit; The data acquisition unit is responsible for the acquisition and preprocessing of production data, including three-dimensional visual recognition and sensor acquisition. The former is responsible for the acquisition of three-dimensional image data of the product, and captures the three-dimensional image data in real time through the image processor. The latter acquires the production status and environmental data of the product through designated sensors, including but not limited to temperature sensors, humidity sensors, pressure sensors, and voltage sensors. The intelligent control unit is the core of the whole system, which is responsible for receiving all real-time production data from the data acquisition unit and the data from the human-machine interaction unit, and finally issuing control instructions to the robot unit to perform production operations, while outputting the three-dimensional production process of the product to the human-machine interaction unit in real time to realize production monitoring and decision support; the intelligent control unit integrates anomaly detection module, performance evaluation module, state estimation module, digital twin module, decision analysis module, prediction analysis module and execution decision module; The robot operation unit is the execution part of the system, which is responsible for completing the operation of the product according to the instructions of the intelligent control unit. The unit adopts an industrial robot and a matching control system. The control system receives instructions from the intelligent control unit and realizes precise position and posture control by controlling the joints and drivers of the robot. The human-computer interaction unit is an interface for the system to interact with the product manager, operator and inspector. It receives product information input by the product manager, operating instructions input by the operator and product quality standards input by the inspector, and displays the production status and progress of the product output by the intelligent control module in real time, and displays the product operating instruction log and product production performance. The unit includes a product information input module, a production progress viewing module and a quality inspection module. The information input module provides three parts of information input functions: product information input by the product manager, operating instructions input by the operator and product quality standards input by the inspector, which are used for anomaly detection, performance evaluation and decision analysis of the intelligent control unit; the production progress viewing module receives the production information output by the data acquisition module and the operating instructions, anomaly detection and performance evaluation data output by the intelligent control module, and displays them in real time through two-dimensional graphs, tables, logs and three-dimensional products, supporting users to view the production progress of three-dimensional products in real time through VR glasses.

Images