Automatic control system based on full-automatic intelligent surface treatment equipment

Abstract

The invention discloses an automatic control system based on full-automatic intelligent surface treatment equipment, which relates to the field of industrial production surface treatment and comprises a data acquisition module, a data analysis module, a process decision module, a control execution module, a fault monitoring module, a man-machine interaction module, a data storage and management module, a self-adaptive learning module, a remote monitoring and maintenance module and a green energy-saving module. An advanced sensor is combined with a micro-nano sensing technology, a deep reinforcement learning and transfer learning combined algorithm is applied, genetic optimization fuzzy reasoning is adopted, through magnetic suspension and compound control, faults are monitored based on deep learning, a VR technology is introduced to realize human-computer interaction, and a block chain technology is adopted to store data. And a meta-reinforcement learning algorithm is used to automatically adjust decision rules and parameters, and the remote collaborative robot carries out maintenance operation on the equipment. The system is accurate in data acquisition, scientific in analysis and decision, accurate in control, timely in fault diagnosis, convenient in man-machine interaction, efficient in data management, energy-saving and capable of comprehensively improving the production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production surface treatment, and particularly to an automatic control system based on a fully automatic intelligent surface treatment device. Background Art

[0002] In the field of industrial production, surface treatment plays a crucial role in enhancing product performance, extending service life, and improving appearance. Traditional surface treatment equipment relies on a large amount of manual operation, which is not only inefficient but also difficult to ensure consistent processing quality due to human factors. Manual operation is prone to errors, resulting in an increase in the defective rate of products and an increase in production costs. At the same time, the manual labor intensity is high, and in some harsh environments, such as high-temperature, high-humidity, or work scenarios involving exposure to harmful chemicals, it poses a threat to the physical health of workers.

[0003] With the development of automation technology, some surface treatment equipment has achieved a certain degree of automation. However, the existing automatic surface treatment equipment has many limitations. On the one hand, the intelligent level of the control system of the equipment is relatively low, and it is unable to make flexible and accurate parameter adjustments according to complex and changeable process requirements and real-time production conditions. When facing workpieces with different materials, shapes, and surface treatment requirements, it is difficult to adaptively optimize the processing process, resulting in poor processing effects. On the other hand, the data acquisition and analysis capabilities of the existing equipment are limited, and it cannot comprehensively and accurately obtain the equipment operation data and key information during the processing process. This makes the monitoring and diagnosis of equipment failures untimely and inaccurate, and equipment maintenance often lags behind, resulting in long downtime and seriously affecting production efficiency.

[0004] In addition, under the general trend of Industry 4.0 and intelligent manufacturing, the requirements for informatization, intelligence, and networking in the production process are getting higher and higher. The existing automatic control system for surface treatment equipment lacks effective data management and remote monitoring functions, and cannot achieve efficient storage, analysis, and sharing of production data, nor can it be seamlessly docked with the overall production management system of the enterprise. Enterprises cannot grasp the equipment operation status and production progress in real time, which is not conducive to the reasonable allocation of production resources and the optimization of the production process. Therefore, it is urgent to develop an advanced automatic control system based on a fully automatic intelligent surface treatment device. Summary of the Invention

[0005] The automatic control system based on a fully automatic intelligent surface treatment device proposed by the present invention aims to solve the problems mentioned in the above prior art.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An automatic control system based on a fully automatic intelligent surface treatment device, comprising: Data acquisition module: Collect the operation data of the surface treatment equipment. The quantum dot sensor collects the temperature of the processing chamber, the fiber Bragg grating pressure sensor collects the pressure, and the Hall effect speed sensor collects the speed of the conveying device; Data analysis module: Receive the data from the data acquisition module, analyze it using deep reinforcement learning and transfer learning algorithms, construct a neural network model based on the attention mechanism, process the collected data, and introduce the Bayesian optimization algorithm to automatically adjust the hyperparameters of the neural network; Process decision-making module: Based on the results of the data analysis module and the preset process standards, make decisions using the fuzzy inference algorithm optimized by the genetic algorithm, introduce the multi-objective decision-making theory, and balance the cost and efficiency objectives; Control execution module: Receive the instructions from the process decision-making module, use the servo drive system based on magnetic levitation technology to control the temperature of the processing chamber and the speed parameters of the conveying device, and adopt the adaptive fuzzy-proportional integral differential composite control algorithm to automatically adjust the control parameters according to the operating state of the system; Fault monitoring module: Monitor the operating state of the equipment in real time, analyze the data using the fault feature extraction algorithm and fault classification algorithm based on deep learning, diagnose faults by comparison, and use time series analysis and long short-term memory network to give early warnings of equipment faults; Human-machine interaction module: Provide a human-machine interaction interface, support the input of operation parameters and query of equipment information through touch screen, keyboard and voice commands, use VR and AR technologies to experience the operating state and process flow of the equipment immersively, and use the deep neural network speech recognition algorithm based on the Transformer architecture combined with the language model for semantic understanding of voice commands; Data storage and management module: Store the collected data, analysis results, decision-making schemes and fault information in a distributed database based on blockchain technology, and use the data mining algorithm based on graph neural network to analyze the historical data to provide a basis for optimizing the process.

[0007] Furthermore, it also includes: Adaptive learning module: Use the meta-reinforcement learning algorithm to automatically adjust the decision-making rules and parameters of the process decision-making module according to the real-time feedback and historical data during the operation of the equipment, and dynamically optimize the model structure and parameters of the data analysis module.

[0008] Furthermore, it also includes: Remote monitoring and maintenance module: Connect the equipment to the remote server through 6G communication technology, conduct remote monitoring, fault diagnosis and maintenance of the equipment, obtain the operating state and processing information of the equipment through the mobile phone APP or computer terminal, and use the remote collaborative robot to perform maintenance operations on the equipment.

[0009] Further, the data acquisition module adopts multi-sensor fusion technology to fuse the data of different types of sensors. The fusion algorithm uses a multi-sensor fusion algorithm based on evidence theory. Through the formula to fuse the data, m(A) is the fused belief assignment, and m 1 (B) and m 2 (C) are the belief assignments of different sensors respectively. A sensor self-calibration mechanism is introduced to correct the measurement errors of the sensors in real time; The sensor sampling period t s is determined according to the process complexity, and the formula is , where k is an empirical constant and C is the process complexity level.

[0010] Further, the neural network model in the data analysis module adopts multi-modal transfer learning technology to process the collected data vector D=(T, P, V, ⋯). T is the temperature of the processing chamber, P is the pressure, and V is the speed of the transfer device; the surface treatment quality Q is predicted, and the prediction formula is Q = f(D), where f is the neural network mapping function.

[0011] Further, in the control execution module, the servo drive system adopts an algorithm combining sliding mode variable structure control and fuzzy-PID composite control. The switching function s(x) of the sliding mode control satisfies the design rules, and the system is controlled through the control law u = u eq +u sw . u eq is the equivalent control, and u sw is the switching control; For temperature control, an adaptive fuzzy-proportional integral differential PID composite control algorithm is adopted, combining fuzzy control and PID control, and the control parameters are automatically adjusted according to the system operation state. The control formula is , u(t) is the control output, e(t) is the error between the set temperature and the actual temperature, and Kp, Ki, and Kd are the proportional, integral, and differential coefficients respectively. The coefficients will be adjusted in real time according to the fuzzy rules.

[0012] Further, the fault feature extraction algorithm in the fault monitoring module uses a convolutional sparse autoencoder CSAE to extract features from the collected signals. The CSAE automatically learns the sparse representation of the signals and extracts representative fault features. The fault classification algorithm uses an ensemble learning method to combine multiple different classifiers; A fault dictionary F={f1, f2, ⋯, fn} is established, and the fault is diagnosed by comparing the fault features with the fault dictionary. The deep learning algorithm support vector machine is used to classify the faults. A fault prediction mechanism is introduced, and time series analysis and long short-term memory network LSTM are used to give early warnings of equipment faults.

[0013] Furthermore, the voice command recognition in the human-computer interaction module adopts the method of fine-tuning the pre-trained language model. Based on the language model pre-trained on general corpus, it fine-tunes the professional terminology and operating instructions of the surface treatment equipment, introduces gesture recognition technology, and operates the equipment through gestures.

[0014] Furthermore, the data mining algorithm in the data storage and management module adopts the deep reinforcement graph neural network DRGNN, combined with the idea of ​​reinforcement learning, and optimizes it on the basis of the graph neural network to mine the correlation and dynamic change rules between data, provide decision support for process optimization, and use data watermarking technology to protect the copyright and trace the source of the stored data.

[0015] Furthermore, it also includes: Green energy-saving module: real-time monitoring of equipment energy consumption, according to the needs of surface treatment tasks and equipment operating status, the use of optimization algorithms to adjust the equipment's operating parameters to achieve energy conservation and emission reduction, specifically by adjusting the temperature of the processing chamber and the speed of the conveyor, while maintaining processing quality, to reduce energy consumption, the optimization algorithm uses particle swarm optimization algorithm for parameter optimization.

[0016] Compared with the prior art, the present invention has the following beneficial effects: In terms of data collection, the use of advanced quantum dots, fiber grating and other sensors, combined with micro-nano sensing and self-calibration technology, has greatly improved the accuracy and reliability of data collection. It can accurately capture subtle changes in equipment operation and provide a solid data foundation for subsequent analysis.

[0017] The data analysis module uses deep reinforcement learning, multimodal transfer learning, and a neural network model based on attention mechanism and Bayesian optimization to comprehensively and deeply mine data information, significantly improve the prediction accuracy of surface treatment quality, discover potential problems in advance, and effectively reduce the defective rate.

[0018] The process decision module uses the fuzzy reasoning algorithm optimized by genetic algorithm and multi-objective decision-making theory to quickly formulate scientific and reasonable decision-making plans, taking into account multiple goals such as processing quality, cost and efficiency, so as to optimize the production process.

[0019] The control execution module adopts magnetic levitation technology, sliding mode variable structure and fuzzy-PID composite control algorithm, which greatly improves the accuracy and response speed of equipment parameter adjustment and ensures the high quality and stability of surface treatment.

[0020] The fault monitoring module is based on deep learning and ensemble learning. It can not only diagnose faults quickly and accurately, but also use time series analysis and LSTM to predict faults, greatly reducing equipment downtime and improving production efficiency.

[0021] The human-computer interaction module incorporates VR, AR, advanced voice and gesture recognition technologies, providing operators with a convenient, intuitive and intelligent interaction experience, reducing the operation difficulty and improving the operation efficiency.

[0022] The data storage and management module utilizes blockchain and deep reinforcement graph neural network technologies to ensure data security, traceability, efficiently mine data associations, and provide strong support for continuous process optimization. The green energy-saving module can optimize the device operation parameters in real time according to production requirements, reduce energy consumption, and achieve green production. Overall, the patent system comprehensively improves the intelligent level and production efficiency of the surface treatment equipment. Brief Description of the Drawings

[0023] Figure 1 It is a schematic block diagram of an automatic control system for a fully automatic intelligent surface treatment equipment proposed by the present invention; Figure 2 It is a schematic diagram for comparing the stability of processing quality; Figure 3 It is a schematic diagram for comparing the fault diagnosis time; Figure 4 It is a schematic diagram for comparing the energy consumption. Detailed Embodiments

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention.

[0026] In addition, the terms "first" and "second" are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined. In addition, the terms "mounted", "connected" and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations. The present invention will be further described in detail below with reference to the accompanying drawings.

[0027] Refer to Figures 1-4 : An automatic control system based on a fully automatic intelligent surface treatment device, comprising: Data acquisition module: This module uses a variety of advanced sensors for data acquisition. A quantum dot sensor is used to collect the temperature T of the processing chamber. Its ultra-high sensitivity can sense tiny temperature changes, and the response time is as short as milliseconds; a fiber Bragg grating pressure sensor is used to collect the pressure P, which can effectively resist electromagnetic interference and ensure the accuracy of pressure data; a Hall effect speed sensor is used to collect the speed V of the conveying device. The sensor sampling period ts is determined according to the formula where k is taken as 8 (empirical constant), C is the process complexity level. For example, when the simple process C = 1, ts = 8 seconds; when the complex process C = 5, ts = 1.6 seconds. At the same time, micro-nano sensing technology can capture extremely tiny physical quantity changes during the operation of the device, and the self-calibration mechanism can correct the sensor measurement error in real time to ensure the long-term stability of the data.

[0028] Data analysis module: After receiving the data from the data acquisition module, it uses a machine learning algorithm that combines deep reinforcement learning and transfer learning. A neural network model based on the attention mechanism is constructed to process the collected data vector D=(T,P,V,⋯) and predict the surface treatment quality Q, that is, Q=f(D). The Bayesian optimization algorithm automatically adjusts the neural network hyperparameters to improve the model performance. Taking the treatment of metal workpieces as an example, the model can accurately predict quality indicators such as surface roughness by learning a large amount of historical data. Multi-modal transfer learning integrates multi-modal data such as images and sounds, enabling the model to have a more comprehensive understanding of the surface treatment process.

[0029] Process decision-making module: Based on the results of the data analysis module and the preset process standards, it makes decisions using a fuzzy inference algorithm optimized by a genetic algorithm. The adjustment scheme R is determined through the formula where w i is the weight of each factor. For example, the weight of the temperature factor w1 = 0.3, the weight of the pressure factor w 2 = 0.2, etc.; r i is the output corresponding to the fuzzy rule. The genetic algorithm can quickly find the optimal combination of fuzzy rules and weights in a complex search space, and the multi-objective decision-making theory takes into account quality, cost, and efficiency. For example, under the premise of ensuring the surface treatment quality, the processing time and energy consumption are reasonably reduced.

[0030] Control execution module: Receives instructions from the process decision-making module and adjusts the device parameters using a servo drive system based on magnetic levitation technology. The temperature control uses an adaptive fuzzy-proportional integral derivative composite control algorithm, and the formula is , where K p , K i , K d will be adjusted in real time according to the fuzzy rules. The magnetic levitation technology reduces mechanical friction and wear, and improves the control accuracy and response speed. The sliding mode variable structure control is combined with the fuzzy-PID composite control to enhance the system robustness and ensure stable operation under external disturbances and parameter changes.

[0031] Fault monitoring module: Monitors the operating status of the device in real time, and uses a fault feature extraction algorithm based on deep learning and a fault classification algorithm of support vector machines. A fault dictionary F = {f1, f2, ⋯, fn} is established, and the fault is diagnosed by comparing the fault features with the dictionary. The convolutional sparse autoencoder (CSAE) extracts the fault features, and the ensemble learning method is used for fault classification. At the same time, time series analysis and long short-term memory network (LSTM) are used to predict possible future faults of the device in advance, such as predicting motor faults, sensor faults, etc.

[0032] Human-machine interaction module: Provides a friendly human-machine interaction interface, supporting touch screen, keyboard, or voice command operations. Using virtual reality (VR) and augmented reality (AR) technologies, operators can immerse themselves in the device operating status and process flow through VR devices, and view device parameters and fault prompts in the actual scene through AR devices. The voice command recognition uses a deep neural network voice recognition algorithm based on the Transformer architecture combined with a language model for semantic understanding, and the gesture recognition technology increases the interaction diversity.

[0033] Data Storage and Management Module: All kinds of collected data, including real-time parameters during equipment operation (such as temperature, pressure, rotational speed, etc.), environmental data collected by sensors, data on the change of material properties during surface treatment, etc., as well as the analysis results of these data, generated decision-making schemes, and monitored fault information, are all stored in a distributed database based on blockchain technology. The application of blockchain technology brings unique advantages to data storage. It adopts a distributed ledger structure, where data is divided into multiple data blocks and dispersed and stored on multiple nodes in the network. Each node holds a complete or partial copy of the ledger. This decentralized storage method not only improves the redundancy and reliability of data. Even if some nodes fail, the data will not be lost. Moreover, through consensus mechanisms (such as proof of work, proof of stake, etc.), the consistency and immutability of data are ensured. For example, when a new data block needs to be added to the blockchain, the nodes in the network will verify the legality and integrity of the data. Only after being recognized by the majority of nodes can the data block be successfully added. To further ensure data security, the system adopts advanced data encryption technology. Before data storage, the data will be encrypted using a symmetric encryption algorithm (such as the AES algorithm) or an asymmetric encryption algorithm (such as the RSA algorithm). The symmetric encryption algorithm encrypts and decrypts data through a single key, with fast encryption and decryption speeds, suitable for processing large amounts of data. The asymmetric encryption algorithm uses a public key and a private key, where the public key is used to encrypt data and the private key is used to decrypt data, with higher security and is often used for encrypting key information. At the same time, a hash function is also used to perform a digest process on the data to generate a hash value of a fixed length for verifying the integrity of the data. Once the data is tampered with, the hash value will change. In terms of data analysis, a data mining algorithm based on graph neural networks is used. This algorithm models historical data in the form of a graph, where each element in the data is regarded as a node in the graph, and the association relationship between elements is represented as an edge. For example, the mutual influence relationship between different operating parameters of equipment, the causal relationship between process parameters and treatment effects, etc., can all be clearly presented in the graph. The graph neural network learns the feature representations of nodes and edges through multiple layers of information transfer and aggregation operations, thereby discovering hidden association relationships and patterns in the data. For example, by analyzing the relationship between the combination of process parameters (such as temperature, time, treatment liquid concentration, etc.) and the corresponding treatment effects (such as surface finish, coating adhesion, etc.) in historical data, the differences in treatment effects under different combinations of process parameters are found, providing a scientific basis for process optimization. In addition, to protect the copyright of data and achieve data traceability, data watermarking technology is adopted. Data watermark is an imperceptible piece of information that is embedded into the original data without affecting the normal use of the data. The data watermark can contain content such as the copyright information of the data, source identification, usage permissions, etc. During the storage, transmission, and use of the data, the watermark is always bound to the data.When copyright protection or data traceability is required, watermark information can be accurately extracted from the data through a special watermark extraction algorithm, thereby determining the owner and usage history of the data and preventing the data from being illegally stolen and tampered with.

[0034] The present invention also includes the following modules: Adaptive Learning Module: Using the meta-reinforcement learning algorithm, it promotes the optimization and adjustment of the system in a highly intelligent and flexible manner. During the operation of the device, various sensors continuously collect real-time data, such as the temperature and pressure fluctuations in the processing chamber, the actual operating speed of the conveying device, and the key quality indicators of the products after surface treatment. These real-time feedback data, together with the historically accumulated data, constitute the information basis of the adaptive learning module. The meta-reinforcement learning algorithm plays a key role here. It regards the entire automatic control system as a process of interaction between an intelligent agent and the environment. The intelligent agent here is the adaptive learning module, and the environment covers various operating conditions of the device and different requirements of the surface treatment tasks. The intelligent agent continuously conducts experiments in the environment, takes different actions (such as adjusting the decision rules and parameters of the process decision-making module, or modifying the model structure and parameters of the data analysis module), and learns the optimal strategy based on the rewards obtained after the actions (for example, the improvement of the product quality processed, the increase in production efficiency, and the reduction in energy consumption). Specifically for the adjustment of the process decision-making module, when the device processes workpieces of different materials or specifications, the adaptive learning module will dynamically rewrite the decision rules in the process decision-making module using the meta-reinforcement learning algorithm based on real-time feedback and historical experience. For example, originally for the surface coating treatment of a certain metal material, the decision rule was set to be carried out at specific temperatures and pressures. But when encountering new material characteristics, the algorithm will analyze the data of past similar situations, try to adjust the value ranges and combination methods of parameters such as temperature and pressure, and re-establish the decision rule to achieve a better coating effect. At the same time, for relevant parameters, fine-tuning will also be carried out according to the actual operating effects, perhaps with adjustments accurate to several decimal places, to ensure the accuracy of the process decision-making. In terms of the optimization of the data analysis module, the meta-reinforcement learning algorithm will evaluate the performance of the current model structure in processing new data and completing new tasks. If it is found that the existing model is insufficient in extracting data features during some complex surface treatment processes, resulting in inaccurate analysis results, the algorithm will automatically try to change the number of layers, the number of neurons, or adjust structural parameters such as connection weights of the model. For example, increasing the number of convolutional layers of a convolutional neural network to better capture the subtle features in image data; or adjusting the number of hidden layer nodes of a recurrent neural network to optimize the processing ability for time series data. In parameter optimization, through methods such as gradient descent, the weight parameters of the model are iteratively updated, enabling the model to process data more efficiently and improving the accuracy and efficiency of the analysis. The meta-reinforcement learning algorithm has the ability to quickly learn new environments and new tasks. When facing new surface treatment requirements, such as adopting a new type of surface treatment process, the algorithm can quickly extract key information from historical data and real-time feedback, adjust each module of the system within a short time, enable the entire automatic control system to quickly adapt to the new task requirements, demonstrate excellent adaptability and intelligence level, and ensure that the surface treatment work is always carried out efficiently and accurately.

[0035] In the present invention, the following modules are further included: Remote monitoring and maintenance module: By means of advanced 6G communication technology, a connection bridge is built between the device and the remote server. The 6G communication technology has many excellent characteristics. Its ultra-high bandwidth capability can support data transmission rates of several gigabits per second or even higher. During the operation of the device, multiple high-definition cameras installed at key parts of the device can capture real-time images of the internal operation of the device and the surface treatment at a high frame rate (such as 60 frames per second or higher). The amount of data of these high-definition videos is huge, and through the ultra-high bandwidth of the 6G network, it can be transmitted to the remote server in real time without lag or delay, enabling the operator to clearly and intuitively observe the operating state of the device. At the same time, the ultra-low latency characteristic of 6G communication is also crucial. The data collected from the device end, including the operating parameters of the device monitored in real time by various sensors (such as temperature sensors, pressure sensors, vibration sensors, etc.), can be transmitted to the remote server almost instantaneously (the latency can be as low as milliseconds). This enables the remote server to analyze and process the data in a timely manner. Once abnormal device operation is detected, such as too high temperature, pressure fluctuation exceeding the normal range, etc., the early warning mechanism can be triggered immediately. The characteristic of massive connection provides strong support for the intelligent management of the device. In addition to the device itself, each sub-module in the system, such as different functional units in the automatic control system, remote cooperation robots, etc., can establish connections with the remote server through the 6G network. In this way, the remote server can manage and monitor multiple devices and modules simultaneously, realizing the efficient operation and maintenance of a large-scale device cluster. The operator can conveniently access the remote server through a specially developed mobile phone APP or a computer terminal. On the mobile phone APP, the interface design is simple and intuitive, and the operator can view the operating parameter charts, processing progress information, and fault early warning prompts of the device in real time. Through the computer terminal, richer operation functions and detailed data displays can be obtained, such as in-depth analysis of the device operation data, generation of operation reports, etc. When the device needs maintenance, the operator can use the remote cooperation robot to perform remote maintenance operations. These robots are equipped with high-precision robotic arms and a variety of professional tools, and are equipped with advanced vision and sensing systems. The operator remotely controls the robot through the computer terminal or mobile phone APP based on the high-definition video images and sensor data transmitted by the device in real time. For example, when a certain component inside the device fails and needs to be replaced, the operator can remotely control the robotic arm of the robot, grab the appropriate component, and replace it according to the preset operation process, greatly improving the maintenance efficiency, reducing the losses caused by device failure shutdown, and at the same time reducing the cost and risk of on-site manual maintenance.

[0036] In the present invention, the sensors in the data acquisition module adopt multi-sensor fusion technology to fuse the data of different types of sensors, improving the accuracy and reliability of the data. The fusion algorithm adopts a multi-sensor fusion algorithm based on the evidence theory, and the data is fused through the formula where m(A) is the fused belief assignment, and m 1 (B) and m 2 (C) are the belief assignments of different sensors respectively. Meanwhile, a sensor self-calibration mechanism is introduced, which can correct the measurement errors of the sensors in real time to ensure the long-term stability of the data.

[0037] In the present invention, the data analysis module adopts a neural network model that utilizes multi-modal transfer learning technology. When applying the models pre-trained in other similar surface treatment tasks, first, a basic neural network model is trained from a large amount of data of completed similar surface treatment tasks. These models may structurally include multiple convolutional layers, fully connected layers, etc. Through learning from past data, they have mastered some general surface treatment-related features and patterns. For example, a model pre-trained in the task of metal surface polishing may have learned to identify the parameter change laws in different polishing stages and their correlation with surface quality. When migrating these pre-trained models to the surface treatment task of this device, they are not directly used, but rather some parameters of the models are adjusted and optimized. Specifically, considering factors such as the unique operating parameters of this device and the characteristics of the processing materials, a small amount of initial data of this device is used to fine-tune the pre-trained model, enabling the model to quickly adapt to the processing characteristics of this device. At the same time, this neural network model integrates multi-modal data such as images and sounds. In terms of image data acquisition, the device is equipped with a high-definition industrial camera to capture real-time images of the material surface during the surface treatment process. These images can intuitively present information such as the texture changes and coating coverage on the material surface. Through image processing techniques, such as edge detection and feature extraction, the key information in the images is converted into feature vectors that can be understood by a computer. For sound data, high-precision microphones are arranged around the device to collect sound signals during the operation of the device. Different device operating states and surface treatment processes will generate sounds with different characteristics, such as the sound differences between normal motor operation and when a fault occurs, or the friction sound changes in different stages of material surface treatment. The sound signals are processed through Fourier transform, etc., converted from the time domain to the frequency domain, and key features such as frequency and amplitude are extracted. The processed multi-modal feature data such as images and sounds are fused, and common fusion methods include early fusion and late fusion. Early fusion combines data of different modalities at the feature extraction stage and then inputs them into the neural network for unified learning; late fusion allows data of different modalities to perform feature extraction and learning in their respective sub-networks, and finally, the output results of each sub-network are combined and comprehensively judged at the decision-making layer. Through multi-modal transfer learning, the complementary information between different modal data is fully utilized, enabling the model to not only understand the surface treatment process from a single dimension but also comprehensively analyze from multiple perspectives, thereby having a more comprehensive understanding of the surface treatment process. At the same time, the training efficiency and performance of the model are improved, and it can more accurately analyze and predict the device operation and surface treatment results.

[0038] In the present invention, the servo drive system in the control execution module adopts an algorithm combining sliding mode variable structure control and fuzzy-PID composite control. The sliding mode variable structure control can enhance the robustness of the system and ensure the stability of the system in the presence of external disturbances and parameter variations. At the same time, combining the advantages of fuzzy-PID control can further improve the stability and accuracy of parameter adjustment. The switching function s(x) of the sliding mode control satisfies certain design rules, and the system is controlled through the control law u = u eq +u sw , where u eq is the equivalent control and u sw is the switching control.

[0039] In the present invention, the fault feature extraction algorithm in the fault monitoring module utilizes the convolutional sparse autoencoder (CSAE), which is an advanced technology integrating the characteristics of the convolutional neural network (CNN) and the sparse autoencoder. In the signal acquisition stage, various sensors are distributed on the device, such as vibration sensors, temperature sensors, pressure sensors, etc., which will collect signal data such as vibration, temperature, and pressure during the operation of the device in real time. These raw signal data usually contain a large amount of noise and redundant information and need to be preprocessed, such as filtering, normalization, etc., to improve the data quality. The preprocessed signals are then input into the CSAE. The convolutional layer part of the CSAE is similar to the CNN, and multiple convolutional kernels of different sizes and parameters slide on the signal to extract the local features of the signal. Different from the traditional CNN, the convolutional operation in the CSAE aims to learn the sparse representation of the signal. Sparse representation means that only a few key features can be activated, which can remove a large amount of redundant information and highlight the key features in the signal. In the encoding stage, after multiple convolutional and pooling operations on the input signal, it is mapped to a low-dimensional feature space to form a sparse code. In the decoding stage, these sparse codes attempt to reconstruct the original signal through deconvolution operations. In the continuous training process, the CSAE adjusts the network parameters by minimizing the reconstruction error, enabling it to automatically learn the most representative sparse representation of the signal and then extract the key fault features. For example, when the device has a bearing wear fault, the CSAE can accurately extract the unique frequency features related to wear from the vibration signal. In the fault classification link, an ensemble learning method is adopted. Specifically, multiple different types of classifiers are selected, such as support vector machine (SVM), random forest (RandomForest), multi-layer perceptron (MLP), etc. Each classifier is trained based on the fault features extracted by the CSAE, and they classify the faults from different perspectives. Then, through specific combination strategies, such as the voting method (the minority obeys the majority), the averaging method (taking the average of the probability outputs of each classifier), etc., the results of these classifiers are fused. This way can comprehensively utilize the advantages of different classifiers, reduce the biases and errors that may occur in a single classifier, and thus significantly improve the accuracy and reliability of fault classification, ensuring that the fault type can be quickly and accurately judged when the device fails, providing strong support for maintenance and repair.

[0040] In the present invention, in terms of speech command recognition, the human-computer interaction module adopts an advanced method of fine-tuning a pre-trained language model. First of all, the pre-trained language model is trained on a large-scale general corpus. These general corpora have a wide range of sources, covering various types of text data such as news, literature, academic papers, and social media texts. Through unsupervised learning, the model can learn the basic grammar rules of the language, lexical semantic relationships, and common language expression patterns, thus possessing basic language understanding and generation capabilities. On this basis, for the field of surface treatment equipment, fine-tuning operations are required. Professionals will collect a large number of professional terms related to surface treatment equipment, such as "plasma treatment", "chemical vapor deposition", "surface roughness", etc., as well as various operation instructions, such as "start the cleaning program", "adjust the coating thickness parameter", "pause the current processing task", etc. These professional data are used as the corpus for fine-tuning, and through supervised learning, the pre-trained model is made to learn the language patterns and semantic characteristics in the field of surface treatment equipment. During the fine-tuning process, some parameters of the model will be adjusted so that the model can more accurately recognize and understand speech commands related to surface treatment equipment. For example, by optimizing the attention mechanism of the model, making it pay more attention to the role of professional terms in the instructions, thereby improving the recognition accuracy and semantic understanding ability. At the same time, the human-computer interaction module also introduces gesture recognition technology. This technology uses multiple high-definition cameras and depth sensors installed on the device to capture the gesture actions of the operator in real time. The camera can obtain the two-dimensional image information of the gesture, and the depth sensor can provide the position and distance information of the gesture in the three-dimensional space. These data will be transmitted to the gesture recognition algorithm module of the system. The algorithm first preprocesses the obtained image and depth data, including operations such as noise reduction and normalization, to improve the quality of the data. Then, a feature extraction-based method is used to extract the key features of the gesture from the preprocessed data, such as the degree of finger bending, the orientation of the palm, the movement trajectory of the gesture, etc. These features are matched with pre-defined specific gesture templates. When the matching degree reaches a certain threshold, the system can recognize the gesture intention of the operator and convert it into corresponding device operation instructions. For example, when the operator makes a specific sliding gesture, the system can recognize it as an instruction to adjust the device operation speed; making a fist gesture can be recognized as an instruction to pause the device operation, greatly increasing the diversity and convenience of the interaction.

[0041] In the present invention, the data mining algorithm in the data storage and management module utilizes the Deep Reinforcement Graph Neural Network (DRGNN), which integrates various advanced concepts and mechanisms. From the basic structure of the graph neural network, it models various types of data generated during the operation of the surface treatment equipment, such as equipment parameters, processing time, product quality indicators, etc., in the form of a graph. Each data point is regarded as a node in the graph, and the association relationship between nodes is represented by edges. For example, the mutual influence relationship between the operating parameters of different components of the equipment, or the sequential association between different processing steps. In this way, the complex topological structure between data can be presented comprehensively and intuitively. On this basis, combined with the idea of reinforcement learning, an interaction mechanism between the agent and the environment is introduced. The agent continuously explores in the environment constructed by the graph neural network, trying different data mining strategies. It selects actions (such as extracting features from certain nodes, adjusting the weights of edges, etc.) based on the current state (i.e., the information contained in the nodes and edges in the graph), and adjusts its own strategy according to the rewards obtained after executing the actions (such as the degree of discovering effective association relationships, the contribution to the process optimization decision-making, etc.). Through continuous iterative training, the agent can gradually learn the optimal data mining method, so as to more effectively mine the association relationships and dynamic change rules between data. For example, in the surface treatment process, the agent can discover the hidden non-linear relationship between the slight change of the equipment temperature parameter and the surface hardness of the final product through DRGNN, thereby providing key clues for process optimization and helping engineers formulate a more accurate process parameter adjustment plan. In terms of data security and copyright protection, this module adopts the data watermarking technology. Specifically, key information such as copyright information and data source identification is converted into a watermark signal through a specific encryption algorithm. This watermark signal is not simply superimposed on the surface of the data, but is embedded in the feature space of the data in an imperceptible way. During the storage, transmission and use of the data, the watermark is always closely combined with the data. When copyright protection and traceability are required, the watermark information can be accurately extracted from the data through a special watermark extraction algorithm, so as to determine the legal owner and usage rights of the data, effectively preventing the data from being illegally tampered with and stolen, and providing a reliable guarantee for the safe use and management of the data.

[0042] In the present invention, the following modules are further included: Green Energy-saving Module: Integrating high-precision energy consumption monitoring sensors, it can monitor the energy consumption of each component of the equipment, such as the power system, heating device, cooling system, etc. in real time and accurately with a response speed of milliseconds, and transmit these energy consumption data to the data processing center of the system in a high-frequency manner. In terms of the demand analysis of surface treatment tasks, this module will deeply analyze the type of tasks (for example, chemical coating treatment, physical grinding treatment, heat treatment, etc.), the characteristics of the processing materials (including parameters such as the thermal conductivity, hardness, and chemical activity of the materials), and the expected processing effects (such as surface finish, coating thickness accuracy, etc.). At the same time, combining the current operating status information of the equipment, such as real-time parameters such as the temperature, pressure, and rotation speed of each component, advanced and complex optimization algorithms are used to dynamically adjust the operating parameters of the equipment. Taking temperature control as an example, a multi-point temperature sensor network is deployed inside the processing chamber, which can accurately obtain the temperature distribution at different positions inside the chamber. The green energy-saving module will fine-tune the power of the heating element through an intelligent PID (Proportional-Integral-Derivative) control algorithm according to the task requirements, ensuring that the processing chamber meets the temperature requirements of the processing process while avoiding excessive energy consumption. For the speed adjustment of the conveyor device, the system will comprehensively consider factors such as the chemical reaction time and physical change process of the material during processing, and use a fuzzy control algorithm to reasonably reduce the conveyor speed on the premise of ensuring the uniformity and quality of surface treatment, thereby reducing the energy consumption of the motor. In the selection of optimization algorithms, in addition to the particle swarm optimization algorithm, the simulated annealing algorithm can also be combined. By simulating the temperature reduction principle in the physical annealing process, global search is carried out in the solution space to avoid the algorithm falling into a local optimal solution. With the core goal of minimizing energy consumption, collaborative optimization of multiple operating parameters of the equipment is carried out to achieve a more efficient energy-saving and emission-reduction effect, helping the surface treatment equipment to achieve dual optimization in terms of environmental protection and economy.

[0043] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An automatic control system based on fully automatic intelligent surface treatment equipment, characterized in that: Includes the following modules: Data acquisition module: collects the operating data of the surface treatment equipment, the quantum dot sensor collects the temperature of the processing chamber, the fiber Bragg grating pressure sensor collects the pressure, and the Hall effect speed sensor collects the speed of the conveyor; Data analysis module: receives data from the data collection module, uses deep reinforcement learning and transfer learning algorithms to analyze, builds a neural network model based on the attention mechanism, processes the collected data, and introduces the Bayesian optimization algorithm to automatically adjust the hyperparameters of the neural network; Process decision module: Based on the results of the data analysis module and the preset process standards, the fuzzy reasoning algorithm based on genetic algorithm optimization is used to make decisions, and multi-objective decision theory is introduced to take into account both cost and efficiency goals; Control execution module: receives instructions from the process decision module, uses a servo drive system based on magnetic suspension technology to control the temperature of the processing chamber and the speed parameters of the conveyor, and uses an adaptive fuzzy-proportional integral differential composite control algorithm to automatically adjust the control parameters according to the system operation status; Fault monitoring module: monitors the equipment operation status in real time, uses the fault feature extraction algorithm and fault classification algorithm based on deep learning to analyze the data, diagnoses faults through comparison, and uses time series analysis and long short-term memory network to provide early warning of equipment faults; Human-computer interaction module: provides a human-computer interaction interface, supports touch screen, keyboard and voice command input of operating parameters and query of equipment information, uses VR and AR technology to immersively experience the operating status and process flow of the equipment, and uses a deep neural network speech recognition algorithm based on the Transformer architecture combined with a language model for semantic understanding of voice command recognition; Data storage and management module: The collected data, analysis results, decision-making plans and fault information are stored in a distributed database based on blockchain technology, and historical data are analyzed using a data mining algorithm based on a graph neural network to provide a basis for process optimization.

2. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: Also includes: Adaptive learning module: Utilizes the meta-reinforcement learning algorithm to automatically adjust the decision rules and parameters of the process decision module based on real-time feedback and historical data during equipment operation, and dynamically optimizes the model structure and parameters of the data analysis module.

3. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: Also includes: Remote monitoring and maintenance module: Connect the device to the remote server through 6G communication technology to remotely monitor, diagnose faults and maintain the equipment. Obtain the operating status and processing information of the equipment through mobile phone APP or computer terminal, and use remote collaborative robots to perform maintenance operations on the equipment.

4. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The data acquisition module uses multi-sensor fusion technology to fuse the data of different types of sensors. The fusion algorithm uses a multi-sensor fusion algorithm based on evidence theory. The data is fused, m(A) is the trust distribution after fusion, m1(B) and m2(C) are the trust distribution of different sensors respectively, and the sensor self-calibration mechanism is introduced to correct the measurement error of the sensor in real time; Sensor sampling period t s Determined according to the complexity of the processing process, the formula is , where k is an empirical constant and C is the process complexity level.

5. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The neural network model in the data analysis module uses multimodal transfer learning technology to process the collected data vector D=(T,P,V,⋯), where T is the processing chamber temperature, P is the pressure, and V is the conveyor speed; the surface treatment quality Q is predicted using the prediction formula Q=f(D), where f is the neural network mapping function.

6. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The servo drive system in the control execution module adopts an algorithm combining sliding mode variable structure control and fuzzy-PID composite control. The switching function s(x) of the sliding mode control meets the design rules. Through the control law u=u eq +u sw To control the system, eq For equivalent control, u sw To switch control; For temperature control, an adaptive fuzzy-proportional integral differential PID composite control algorithm is used, combining fuzzy control and PID control to automatically adjust the control parameters according to the system operating status. The control formula is: , u(t) is the control output, e(t) is the error between the set temperature and the actual temperature, K p , K i , K d They are proportional, integral and differential coefficients respectively, and the coefficients will be adjusted in real time according to fuzzy rules.

7. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The fault feature extraction algorithm in the fault monitoring module uses a convolutional sparse autoencoder (CSAE) to extract features from the collected signals. CSAE automatically learns the sparse representation of the signal and extracts representative fault features. The fault classification algorithm uses an integrated learning method to combine multiple different classifiers. A fault dictionary F={f1,f2,⋯,fn} is established. Faults are diagnosed by comparing fault features with the fault dictionary. The deep learning algorithm support vector machine is used to classify faults. A fault prediction mechanism is introduced, and time series analysis and long short-term memory network LSTM are used to provide early warning of equipment failures.

8. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The voice command recognition in the human-computer interaction module adopts the method of fine-tuning the pre-trained language model. Based on the language model pre-trained on general corpus, it fine-tunes the professional terminology and operating instructions of surface treatment equipment, introduces gesture recognition technology, and operates the equipment through gestures.

9. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: The data mining algorithm in the data storage and management module adopts the deep reinforcement graph neural network DRGNN, combined with the idea of ​​reinforcement learning, and optimizes it on the basis of the graph neural network to mine the correlation and dynamic change rules between data, provide decision support for process optimization, and use data watermarking technology to protect the copyright and trace the source of the stored data.

10. The automatic control system based on the fully automatic intelligent surface treatment equipment according to claim 1 is characterized in that: Also includes: Green energy-saving module: real-time monitoring of equipment energy consumption, according to the needs of surface treatment tasks and equipment operating status, the use of optimization algorithms to adjust the equipment's operating parameters to achieve energy conservation and emission reduction, specifically by adjusting the temperature of the processing chamber and the speed of the conveyor, while maintaining processing quality, to reduce energy consumption, the optimization algorithm uses particle swarm optimization algorithm for parameter optimization.

Images