Control system for battery manufacturing device

Abstract

A control system for a battery manufacturing device, belonging to the technical field of battery manufacturing device control. The control system for a battery manufacturing device comprises a first processor, a second processor and a neural network processor, the first processor running a first operating system, and the second processor running a second operating system; the first processor is used for collecting first information of a battery manufacturing device; the second processor is used for collecting second information of the battery manufacturing device, acquiring the first information, and extracting feature information of the first information and the second information, the first information and the second information being associated with information carried by the battery manufacturing device; the neural network processor is used for acquiring the feature information, and, on the basis of the feature information, acquiring first control parameters; the first processor is further used for acquiring the first control parameters, determining second control parameters on the basis of the first control parameters, and, on the basis of the second control parameters, controlling the battery manufacturing device. The control system for a battery manufacturing device improves the data acquisition efficiency and the real-time control capability for battery manufacturing devices.

Description

Control systems for battery manufacturing equipment

[0001] This application claims priority to Chinese patent application No. 202311288011.X filed on September 28, 2023, with the invention name “Control System for Battery Manufacturing Equipment”. The entire contents of the above Chinese patent application are incorporated into this application by reference. Technical Field

[0002] The present invention relates to the technical field of battery manufacturing equipment control, and specifically provides a control system for battery manufacturing equipment. Background Art

[0003] With the continuous development of industrial automation, big data, and artificial intelligence technologies, especially the current success of AI models such as ChatGPT, more and more industrial equipment and production lines will need to evolve from automation to intelligence. Therefore, data will become the key to industrial upgrading. However, due to the wide variety of industrial equipment connection protocols and interface types, and the large amount of data, data collection is becoming increasingly difficult, transmission delays, and data security issues are becoming increasingly prominent.

[0004] To this end, edge computing technology is increasingly being applied to industrial automation, pushing data processing and analysis to the device side, thereby improving response speed and data security. Edge computing technology requires edge gateways to serve as a bridge between devices and the cloud, enabling the collection, processing, and transmission of device data. However, current edge gateway products on the market have several issues, such as limitations in connection protocols and interface types, low data collection frequency, a lack of edge computing capabilities, and poor responsiveness. Therefore, the lithium battery industry requires a highly responsive battery manufacturing equipment control system with machine learning capabilities to meet the current and future high-frequency big data and AI-based intelligent manufacturing requirements for edge computing.

[0005] Accordingly, the art requires a new battery manufacturing equipment control solution to solve the above problems.

[0006] Summary of the Invention

[0007] In order to overcome the above-mentioned defects, the present invention is proposed to provide a solution or at least partially solve the above-mentioned technical problems.

[0008] In a first aspect, the present invention provides a control system for battery manufacturing equipment, the control system comprising a first processor, a second processor, and a neural network processor, wherein the first processor runs a first operating system and the second processor runs a second operating system;

[0009] The first processor is used to collect first information;

[0010] The second processor is configured to collect second information, obtain the first information, and extract characteristic information of the first information and the second information, wherein the first information and the second information are associated with information carried by the battery manufacturing equipment;

[0011] The neural network processor is used to obtain the characteristic information, and obtain a first control parameter based on the characteristic information;

[0012] The first processor is further configured to obtain the first control parameter, determine a second control parameter based on the first control parameter, and control the battery manufacturing equipment based on the second control parameter.

[0013] In one technical solution, the first processor includes a first data acquisition module;

[0014] The first data acquisition module acquires first information at a sampling period that is less than a preset threshold.

[0015] In one technical solution, the first processor further includes a determination module and a control module;

[0016] The determination module is used to obtain the first control parameter obtained by the neural network processor, and determine the second control parameter based on the first control parameter;

[0017] The control module is used to control the battery manufacturing equipment based on the second control parameter.

[0018] In one technical solution, the first processor further includes a data preprocessing module;

[0019] The data preprocessing module is used to perform filtering processing on the first information.

[0020] In one technical solution, the second processor includes a second data acquisition module;

[0021] The second data acquisition module acquires second information at a sampling period greater than or equal to a preset threshold.

[0022] In one technical solution, the second processor further includes a feature extraction module;

[0023] The feature extraction module is used to extract features from the first information and the second information to obtain a feature vector set.

[0024] In one technical solution, the second processor further includes a conversion module;

[0025] The conversion module is used to convert the first information from the time domain to the frequency domain.

[0026] In one technical solution, the second processor further includes a data reporting module.

[0027] The data reporting module is used to upload training data to the cloud to train the neural network model in the cloud; wherein, the neural network model outputs a first control parameter, and the training data is associated with the first information, the second information and the preset first control parameter.

[0028] In one technical solution, the control system further includes a first storage module and a second storage module;

[0029] The first storage module is used to store the first information and the first control parameter obtained by the first processor;

[0030] The second storage module is used to store the second information, the feature information extracted by the second processor, and the first control parameter obtained by the neural network processor.

[0031] In one technical solution, the first operating system is an RTOS system; the second operating system is a Linux system.

[0032] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0033] The present invention provides a control system for battery manufacturing equipment. The control system includes a first processor, a second processor, and a neural network processor, wherein the first processor runs a first operating system and the second processor runs a second operating system. The first processor is configured to collect first information; the second processor is configured to collect second information, obtain the first information, and extract feature information of the first and second information, wherein the first and second information are associated with information carried by the battery manufacturing equipment. The neural network processor is configured to obtain the feature information and obtain a first control parameter based on the feature information. The first processor is further configured to obtain the first control parameter, determine a second control parameter based on the first control parameter, and control the battery manufacturing equipment based on the second control parameter. The present invention, based on an architecture comprising dual processors running dual operating systems and a neural network processor, collects information carried by the battery manufacturing equipment, meeting real-time requirements while improving system throughput and resource utilization, thereby resolving latency issues in data collection. The neural network processor performs computational tasks of a neural network model on the information carried by the battery manufacturing equipment, enabling the present invention to perform edge computing tasks on the battery manufacturing equipment. The first processor determines a second control parameter based on the first control parameter output by the neural network model, and then controls the battery manufacturing equipment based on the second control parameter, thereby achieving real-time control of the battery manufacturing equipment based on the results of the edge computing tasks and improving the control efficiency of the battery manufacturing equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The disclosure of the present invention will be more easily understood with reference to the accompanying drawings. Those skilled in the art will readily appreciate that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Furthermore, similar numbers in the drawings represent similar components, wherein:

[0035] FIG1 is a schematic diagram of the main structure of a control system of a battery manufacturing equipment according to an embodiment of the present invention;

[0036] FIG2 is a schematic structural diagram of a control system of a battery manufacturing device according to an embodiment of the present invention;

[0037] FIG3 is a schematic diagram of a partial structure of a control system of a battery manufacturing equipment according to one embodiment of the present invention;

[0038] FIG4 is a working principle diagram of a control system of a battery manufacturing equipment according to an embodiment of the present invention;

[0039] FIG5 is a flow chart of a method for implementing a control system of a battery manufacturing device according to an embodiment of the present invention;

[0040] FIG6 is a flow chart of a method for implementing a control system of battery manufacturing equipment according to another embodiment of the present invention.

[0041] List of reference numerals: 11: first processor; 12: second processor; 13: neural network processor; 41: first tension sensor; 42: second tension sensor; 43: third tension sensor. DETAILED DESCRIPTION

[0042] Some embodiments of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0043] In the description of the present invention, "module" and "processor" may include hardware, software, or a combination of both. A module may include hardware circuitry, various suitable sensors, communication ports, and memory. It may also include software components, such as program code, or a combination of software and hardware. A processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. A processor has data and / or signal processing capabilities. A processor may be implemented in software, hardware, or a combination of both. Non-transitory computer-readable storage media include any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, and the like. The term "A and / or B" refers to all possible combinations of A and B, such as only A, only B, or both A and B. The terms "at least one of A or B" or "at least one of A and B" have similar meanings to "A and / or B" and may include only A, only B, or both A and B. The singular forms "one" and "the" may also include the plural forms.

[0044] As mentioned in the background technology, current edge gateway products have some problems, such as limitations on connection protocols and interface types, low data collection frequency, lack of edge computing capabilities, and low responsiveness.

[0045] To this end, the present invention provides a control system for battery manufacturing equipment, comprising a first processor, a second processor, and a neural network processor, wherein the first processor runs a first operating system and the second processor runs a second operating system; the first processor is used to collect first information; the second processor is used to collect second information, obtain the first information, and extract feature information of the first information and the second information, wherein the first information and the second information are associated with information carried by the battery manufacturing equipment;

[0046] The neural network processor is used to obtain the characteristic information and obtain the first control parameter based on the characteristic information; the first processor is also used to obtain the first control parameter, determine the second control parameter based on the first control parameter, and control the battery manufacturing equipment based on the second control parameter.

[0047] The present invention is based on the architecture of dual processors running dual operating systems plus a neural network processor, which can collect information carried by battery manufacturing equipment, meet real-time requirements while improving the system's throughput and resource utilization, and solve the delay problem in data collection; the neural network processor performs the calculation task of the neural network model on the information carried by the battery manufacturing equipment, so that the present invention can perform edge computing tasks on the battery manufacturing equipment; the first processor determines the second control parameter based on the first control parameter output by the neural network model, and then controls the battery manufacturing equipment based on the second control parameter, realizing real-time control of the battery manufacturing equipment according to the results of the edge computing task, and improving the control efficiency of the battery manufacturing equipment.

[0048] Please refer to FIG1 , which is a schematic diagram of the main structure of a control system of a battery manufacturing equipment according to an embodiment of the present invention.

[0049] As shown in Figure 1, the control system of the battery manufacturing equipment in the embodiment of the present invention mainly includes a first processor 11, a second processor 12, and a neural network processor 13, wherein the first processor 11 runs a first operating system and the second processor 12 runs a second operating system; the first processor 11 is used to collect first information; the second processor 12 is used to collect second information, obtain the first information, and extract feature information of the first information and the second information, wherein the first information and the second information are associated with the information carried by the battery manufacturing equipment; the neural network processor 13 is used to obtain the feature information and obtain a first control parameter based on the feature information; the first processor is also used to obtain the first control parameter, determine a second control parameter based on the first control parameter, and control the battery manufacturing equipment based on the second control parameter. The architecture of dual processors running dual systems plus a neural network processor can realize multi-task parallel processing, and by running a real-time operating system (RTOS) on the first processor, the ability to process real-time tasks, critical tasks and real-time control of battery manufacturing equipment is improved, thereby reducing the delay of battery manufacturing equipment control.

[0050] In the technical solution of the present invention, the first information and the second information are associated with information carried by the battery manufacturing equipment. In practical applications, those skilled in the art may determine, based on the event to be controlled (such as tension control during winding and unwinding, coating surface density control during coating, etc.), to collect information carried by the battery manufacturing equipment related to the event as the first information and the second information.

[0051] Specifically, for example, in the scenario of controlling the tension during the winding and unwinding process, the first information may be the tension feedback value fed back by the winding and unwinding device, and the second information may be the width, thickness, and material of the coiled material output by the winding and unwinding device, the tape speed, spindle diameter, torque, winding diameter, and unwinding diameter output by the winding and unwinding device, the current speed and tension setting value of the motor controlled by the winding and unwinding device output by the winding and unwinding device, and the ambient temperature and humidity collected by the thermometer / dew point meter in the factory. Alternatively, the first information may be the width, thickness, and material of the coiled material of the winding and unwinding device, the tape speed, spindle diameter, torque, winding diameter, and unwinding diameter output by the winding and unwinding device, the current speed and tension setting value of the motor controlled by the winding and unwinding device output by the winding and unwinding device, and the ambient temperature and humidity collected by the thermometer / dew point meter in the factory, and the second information may be the tension feedback value fed back by the winding and unwinding device.

[0052] For another example, in the scenario of controlling the coating surface density during the coating process, the first information may be the coating surface density feedback value output by the coating machine, and the second information may be the viscosity, concentration, and water content of the slurry output by the manufacturing execution system (MES), the belt speed, the set coating thickness, width, the coating surface density setting value, the die head temperature, the mold cavity pressure, the oven temperature output by the coating machine, and the ambient temperature and humidity collected by the factory thermometer / hygrometer / dew point meter. Alternatively, the first information may be the viscosity, concentration, and water content of the slurry output by the manufacturing execution system (MES), the belt speed, the set coating thickness, width, the coating surface density setting value, the die head temperature, the mold cavity pressure, the oven temperature output by the coating machine, and the ambient temperature and humidity collected by the factory thermometer / hygrometer / dew point meter, and the second information may be the coating surface density feedback value output by the coating machine.

[0053] In one technical solution, the first information may be real-time data output by the battery manufacturing equipment collected at a sampling period less than a preset threshold, or device data output by the battery manufacturing equipment collected at a sampling period greater than or equal to the preset threshold; and the second information may be real-time data output by the battery manufacturing equipment collected at a sampling period less than the preset threshold, or device data output by the battery manufacturing equipment collected at a sampling period greater than or equal to the preset threshold, without limitation herein. Those skilled in the art may determine the sampling period for collecting data based on the specific content of the first information / second information.

[0054] In the technical solution of the present invention, there is no specific limitation on the first information and the second information to be collected. Those skilled in the art can determine the first information and the second information to be collected based on the event to be controlled. Therefore, the first information and the second information to be collected for the tension control event in the above-mentioned winding and unwinding process, and the first information and the second information to be collected for the coating surface density control event in the coating process should not be used as limitations on the technical solution of the present invention.

[0055] In one technical solution, the first processor 11 sends the collected first information to the second processor 12 through the first input / output interface, the second processor 12 receives the first information sent by the first processor 11 through the second input / output interface, and then sends the feature information to the neural network processor 13 through the third input / output interface. The neural network processor 13 receives the feature information through the fourth input / output interface and sends the acquired first control parameter to the second processor 12 through the fourth input / output interface. After the second processor 12 obtains the first control parameter from the third input / output interface, it sends the first control parameter to the first processor 11 through the second input / output interface. Then, the first processor 11 determines the second control parameter based on the first control parameter, and controls the battery manufacturing equipment based on the second control parameter.

[0056] As shown in Figure 2, the control system of the present invention is based on a dual-processor, dual-operating system, and neural network processor architecture. The hardware design can adopt the NXP i.MX8M Plus chip, which has four 1.6GHz Cortex-A53 ARM cores, an integrated independently running Cortex-M7 core, and a neural processing unit (NPU) running at a speed of 2.3TOPS. To improve system responsiveness, dual-channel DDR4 RAM memory is used, providing a maximum bandwidth of 50.6GB / s.

[0057] In one technical solution, in order to avoid the impact of excessive processor temperature on system stability, dynamic voltage and frequency adjustment technology is adopted. By reading the temperature sensor built into the processor, the chip temperature is detected in real time, and the processor's operating frequency and voltage are adaptively changed according to the processor's current load and power consumption requirements to ensure system stability while achieving the goal of balancing performance and energy efficiency.

[0058] In one technical solution, specifically, the first processor 11 may be a Cortex-M7 core, the second processor 12 may be a Cortex-A53 ARM core, and the neural network processor 13 may be a neural processing unit (NPU). A first operating system runs on the first processor 11, and a second operating system runs on the second processor 12. To achieve efficient task scheduling, in this embodiment, the first and second operating systems use a hybrid task scheduling algorithm: a periodic task model (PTM) static priority scheduling algorithm is used in the first operating system, and a dynamic priority scheduling algorithm is used in the second operating system.

[0059] Specifically, within the first operating system, tasks to be executed are first analyzed and organized, clarifying information such as the resources, time, and dependencies required for each task. Then, based on real-time requirements, all pending tasks are ranked by importance or urgency, and a preemptive scheduling strategy is adopted, meaning that high-priority tasks can preempt running lower-priority tasks.

[0060] A detailed time schedule is also developed, including the time required for the start, end and intermediate stages of each task, and ensuring that there are no conflicts and cross-influences between tasks. In order to improve the real-time performance of the system, additional factors such as context switching, delay and interrupt handling are considered in time planning.

[0061] It can also determine a reasonable allocation plan based on the resource requirements (such as CPU and memory) of different task types and priorities, and monitor and manage resource usage. Finally, it starts each pending task one by one according to the schedule, and monitors its running status through real-time monitoring. If an abnormality or error is found, the corresponding operation is stopped immediately and an alarm is issued.

[0062] After all tasks are completed, each task is evaluated and feedback is provided. Scheduling strategies are modified based on actual conditions, and task execution statistics and analysis are performed to optimize overall system performance. Because the static priority scheduling algorithm determines the priority of each process at creation time, it can meet the real-time system's requirements for task response time and reasonability.

[0063] In the second operating system, all tasks are initially sorted from high to low priority, the highest priority task is selected for execution, and the priority of each task is continuously updated according to actual conditions to ensure that the system can respond to high-priority tasks in a timely manner. When a new task enters the system, it is inserted into the appropriate position according to its priority, and all task lists are reordered.

[0064] For low-priority tasks that have not been executed for a long time, a preemptive scheduling strategy is adopted so that they can be executed in a timely manner when necessary. Load balancing technology is used to balance the number of threads running on each processor to avoid system crashes or delays caused by overloading of some processors.

[0065] The dynamic priority scheduling algorithm enables the process with the highest priority to obtain CPU resources as quickly as possible, thereby speeding up the response speed. In a dynamic environment, the execution time and resource requirements of different processes may change. By dynamically adjusting the priority of each process, we can better adapt to such changes.

[0066] In one technical solution, the first processor 11 includes a first data acquisition module; the first data acquisition module acquires first information at a sampling period less than a preset threshold.

[0067] Specifically, the first data acquisition module collects first information from the battery manufacturing equipment at a sampling period less than a preset threshold, which may be 10 milliseconds. This means the first data acquisition module supports data acquisition with a sampling period less than 10 milliseconds, such as CAN bus data and microsecond-level analog data. By enabling the first data acquisition module to collect data from the battery manufacturing equipment at a sampling period less than the preset threshold, real-time data transmission is ensured.

[0068] In one technical solution, the first processor 11 also includes a determination module and a control module; the determination module is used to obtain the first control parameter obtained by the neural network processor 13, and determine the second control parameter based on the first control parameter; the control module is used to control the battery manufacturing equipment based on the second control parameter.

[0069] Specifically, the first control parameter can be a PID parameter obtained by the neural network processor 13. The determination module obtains the PID parameter obtained by the neural network processor 13, and determines the second control parameter for controlling the battery manufacturing equipment based on the PID parameter and the PID control algorithm. The second control parameter can be a specific adjustment amount when controlling the battery manufacturing equipment; and then the control module can control the battery manufacturing equipment according to the second control parameter.

[0070] PID is the abbreviation of Proportional, Integral, and Differential. The PID control algorithm is a control algorithm that combines proportional, integral, and differential.

[0071] The control module of the first processor 11 provides real-time control of the equipment through interfaces such as high-speed I / O, CAN bus, network, and serial ports, supporting a system scan cycle of less than 1 millisecond. By running a real-time operating system (RTOS) on the first processor 11 to control the battery manufacturing equipment, the high-responsiveness and real-time control capabilities of the battery manufacturing equipment are enhanced.

[0072] In one technical solution, the first processor 11 further includes a data preprocessing module; the data preprocessing module is used to perform filtering processing on the first information.

[0073] Specifically, the data preprocessing module is used to filter the first information. In practical applications, different filtering methods can be used depending on the data type of the first information, such as low-pass filtering, notch filtering, median filtering, and sliding average filtering. Filtering and denoising the first information by the data preprocessing module prevents the impact of noise on the information, which could lead to inaccurate subsequent data processing.

[0074] In one technical solution, the second processor 12 includes a second data acquisition module; the second data acquisition module is used to acquire second information, and the second data acquisition module acquires the second information with a sampling period greater than or equal to a preset threshold.

[0075] Specifically, the second data acquisition module collects the second information at a sampling period greater than or equal to a preset threshold, which can be 10 milliseconds. The second data acquisition module supports PLC-specific protocols from Mitsubishi, Siemens, Omron, and Keyence, as well as standard protocols such as Modbus and OPU UA, with a maximum tag count of 20,000. The second processor 12 collects data from the battery manufacturing equipment, in addition to the first information, at a sampling period greater than or equal to the preset threshold. This enables parallel data collection, improving data transmission efficiency and system resource utilization.

[0076] In one technical solution, the second processor 12 includes a feature extraction module; the feature extraction module is used to extract features from the first information and the second information to obtain a feature vector set.

[0077] Specifically, the feature extraction module is used to extract features from the first information and the second information and obtain a feature vector set. In practical applications, the feature vector set can be automatically extracted through deep learning based on relevant information carried by the battery manufacturing equipment.

[0078] For example, for the first information in the coating surface density control scenario mentioned above, when performing feature extraction, the time domain features and frequency domain features of the coating surface density feedback value are extracted. The extracted time domain features include the average value and the deviation of the coating surface density feedback value. When extracting the frequency domain features, it is necessary to perform Fourier transform on the time domain data of the coating surface density feedback value to analyze its performance in the frequency domain, including the main frequency components and the energy distribution at the characteristic frequency. In this technical solution, the first information can also be the first information preprocessed by the data preprocessing module, that is, feature extraction is performed on the filtered first information.

[0079] In one technical solution, the second processor 12 includes a conversion module; the conversion module is used to convert the first information from the time domain to the frequency domain.

[0080] Specifically, the conversion module is configured to convert the first information. In some specific embodiments, the first information is converted from the time domain to the frequency domain using a Fourier transform by performing spectral analysis on the first information. For example, in the context of tension control for battery manufacturing equipment, the purpose of performing frequency domain analysis within a PID algorithm is to better understand the characteristics of tension in a closed-loop control system. Frequency domain analysis allows for observing and analyzing tension changes from the perspective of signal frequency, revealing the dynamic response, stability, and frequency characteristics of the closed-loop control system. Specifically, frequency domain analysis of tension can reveal how tension responds to input signals (such as excitation or interference) at different frequencies. This helps understand the closed-loop control system's sensitivity to different frequency components and its ability to suppress rapidly changing signals or periodic interference. By examining the closed-loop control system's gain and phase curves at different frequencies, it is possible to determine whether the closed-loop control system has oscillation tendencies or stability issues. If issues such as excessively high or low gain values ​​or phase delays exist, further adjustment of the PID parameters may be necessary to improve stability. Frequency domain analysis can also allow for the design of a tension PID controller to consider more important signal components within specific ranges (e.g., low or high frequencies) based on application requirements.

[0081] Frequency domain analysis allows us to understand the impact of different frequency components on tension and adjust the control strategy accordingly to optimize control performance. In short, the application of frequency domain analysis in the reinforcement learning tension PID algorithm can more comprehensively understand and master the dynamic characteristics of the closed-loop control system, thereby improving control effectiveness and stability.

[0082] In one technical solution, the second processor 12 further includes a data storage module, which is used to store part of the data locally for subsequent processing and analysis.

[0083] In one technical solution, the second processor 12 also includes a data reporting module; the data reporting module is used to upload training data to the cloud to train the neural network model in the cloud; wherein, the neural network model outputs a first control parameter, and the training data is associated with the first information, the second information and the preset first control parameter.

[0084] Specifically, the training data is uploaded to the server through the data reporting module. The training data can be the first information, the second information, and the preset first control parameter. The server evaluates, trains and optimizes the neural network model based on the training data. The data reporting module is also used to upload the feature vector set extracted from the first information and the second information, the first control parameter and the second control parameter determined based on the first control parameter for data backup. The neural network model in this embodiment can be a combination of a multi-layer perceptron (MLP) and reinforcement learning (RL). The multi-layer perceptron (MLP) is a fully connected neural network that can handle a series of complex nonlinear relationships. Reinforcement learning (RL) is a method of obtaining the optimal strategy through exploration and learning under the guidance of environmental feedback. In one embodiment, a decision model can be trained by combining MLP and RL, and the surface density of the coating machine is predicted by MLP, and then RL is used to decide how to adjust the fuzzy rules to achieve the best control effect.

[0085] The protocols supported by the data reporting module include WebAPI, MQTT, FTP, etc.

[0086] In one technical solution, the second processor 12 further includes a security control module, which is used to perform security control on the battery manufacturing equipment, such as user authentication, data encryption, access control, data physical isolation, etc., to ensure data security.

[0087] In one technical solution, the control system also includes a first storage module and a second storage module; the first storage module is used to store the first information and the first control parameter obtained by the first processor 11; the second storage module is used to store the second information, the processing results of the second processor 12 and the first control parameter obtained by the neural network processor 13.

[0088] Specifically, the first processor 11 and the second processor 12 use a shared memory method to exchange data. The first processor 11 and the second processor 12 share a first storage module. The first storage module is used to store the first information collected by the first processor 11 and the first control parameter obtained by the first processor 11. The second processor 12 can obtain the first information collected by the first processor 11 from the first storage module.

[0089] The second processor 12 and the neural network processor 13 use a shared memory method to exchange data. The second processor 12 and the neural network processor 13 share a second storage module. The second storage module is used to store the second information collected by the second processor 12, the processing results of the second processor 12 and the first control parameters obtained by the neural network processor 13. The processing results of the second processor 12 include a feature vector set extracted by the feature extraction module. The second processor 12 can obtain the first control parameters obtained by the neural network processor 13 from the second storage module and share them with the first processor 11 through the first storage module.

[0090] To ensure synchronization and mutual exclusion of memory access and avoid data corruption, the lightweight remote processor message transmission protocol (RPMsg) is used between the first processor 11 and the second processor 12. The lightweight remote processor message transmission protocol (RPMsg) is a protocol that provides a lightweight message passing mechanism in a multi-processor system, providing a reliable, efficient and low-latency cross-operating system and cross-processor communication mechanism.

[0091] As shown in Figure 3, to further improve system responsiveness, the Virtual IO Device Driver Model (Virtio) is used as an intermediate transport layer between first processor 11 and second processor 12. Virtio is a high-performance device driver framework in a virtualized environment that isolates the lightweight Remote Processor Message Transfer Protocol (RPMsg) from the underlying hardware. The Virtual IO Device Driver Model (Virtio) and Lightweight Remote Processor Message Transfer Protocol (RPMsg) layers enable message transmission between different processors while ensuring efficient performance and low latency.

[0092] In one technical solution, the first operating system is an RTOS system; the second operating system is a Linux system. Specifically, the present invention adopts a dual-system architecture of an RTOS system and a Linux system, running the RTOS system on the first processor 11 and the Linux system on the second processor 12, to meet the task processing requirements of high concurrency and high responsiveness. When the RTOS system is running on the first processor 11, the first data acquisition module of the first processor 11 collects first information at a sampling period less than a preset threshold. When the Linux system is running on the second processor 12, the second data acquisition module of the second processor 12 collects second information at a sampling period greater than or equal to the preset threshold.

[0093] Please refer to FIG4 , which is a working principle diagram of a control system of a battery manufacturing equipment according to an embodiment of the present invention.

[0094] As shown in Figure 4, in this embodiment, the control system of battery manufacturing equipment is applied as an edge gateway in the winding and unwinding equipment in the field of lithium-ion battery cell manufacturing. The equipment layer includes three tension control systems: a closed loop between the first tension sensor 41 and the unwinding motor, a closed loop between the second tension sensor 42 and the main traction motor, and a closed loop between the third tension sensor 43 and the winding motor. In this embodiment, the three tension control systems adopt the same control method, which is described below using the third tension sensor 43 and the winding motor as an example. The specific workflow is to first collect first information through the first processor 11 (the RTOS system of the Cortex-M core) with a sampling period less than a preset threshold, collect the tension feedback value of the third tension sensor 43 through a high-response bus with a sampling period less than 10 milliseconds, and to avoid the influence of noise on the signal while maintaining the real-time nature of the signal, the collected tension feedback value is filtered in real time within the RTOS system.

[0095] The second processor 12 (Linux system of Cortex-A core) is then used to collect device data for tension control with a sampling period greater than or equal to a preset threshold, and the collected device data and tension feedback value are processed, and feature extraction is performed on the processed data. The obtained feature vector set is reported, and the second processor 12 then obtains the neural network model from the server, where the neural network model is trained by the server based on the training data reported by the second processor 12 and then sent to the edge gateway. The second processor 12 sends the feature vector set and the neural network model to the neural network processor (NPU).

[0096] The neural network processor (NPU) then predicts the first control parameter of the tension data based on the feature vector set obtained from the second processor 12 (Cortex-A core Linux system) and the neural network model.

[0097] Then, the first processor 11 (RTOS system of Cortex-M core) determines the second control parameter for controlling the tension of the winding motor through the PID control algorithm based on the first control parameter of the predicted tension data, and finally performs PID closed-loop control on the winding motor based on the second control parameter, and adjusts the speed of the winding motor in real time in a closed-loop manner, thereby completing high-response control of the tension of the winding and unwinding equipment.

[0098] 5 , which is a flow chart of a method for implementing a control system of a battery manufacturing device according to an embodiment of the present invention.

[0099] As shown in FIG5 , the control system of the battery manufacturing equipment of the present invention includes the following steps when applied to the intelligent tension control of the winding and unwinding equipment:

[0100] Step S501: The tension feedback value fed back by the unwinding and winding device is collected and filtered by the first processor 11 (RTOS system). Specifically, the tension feedback value fed back by the tension sensor is collected by the first processor 11 (RTOS system). To avoid the influence of noise on the signal and to maintain the real-time performance of the signal, this embodiment uses low-pass filtering to filter the tension feedback value. The low-pass filtering algorithm is as follows: Y(n) = αX(n) + (1-α)Y(n-1)

[0101] Where α is the filter coefficient, X(n) is the tension feedback value of this sampling, Y(n-1) is the output value of the previous filtering, and Y(n) is the output value of this filtering.

[0102] At the same time, in order to avoid the periodic influence of the signal caused by the deformation or eccentricity of the mechanical structure of the rewinding and unwinding spindle, this embodiment further adopts a notch filter algorithm to perform hybrid filtering on the tension feedback value. The notch filter algorithm is as follows:

[0103] Where H(z) is the filter's transfer function, ω0 is the digital angular frequency corresponding to the current spindle speed, and r is the attenuation factor. It is understood that various filtering methods can be selected based on the type of data collected in the actual application, and this is not limited here.

[0104] Step S502: the first processor 11 (RTOS system) sends the tension feedback value after filtering to the second processor 12 (Linux system).

[0105] Step S503: The second processor 12 (Linux system) collects equipment data. The equipment data specifically includes the width, thickness, and material of the coil output by the winding and unwinding equipment; the winding and unwinding equipment's tape speed; the spindle diameter, torque, winding and unwinding diameters; the current speed and tension setting of the motor controlled by the winding and unwinding equipment; and the ambient temperature and humidity collected by the factory thermo-hygrometer / dew point meter.

[0106] Step S504: The second processor 12 (Linux system) pre-processes the tension feedback value, specifically performing spectrum analysis on the tension feedback value and converting the tension feedback value from the time domain to the frequency domain using Fourier transform. The Fourier transform is shown below:

[0107] Where F(k) represents the complex representation of the tension feedback value in the discrete frequency domain, f(n) represents the value of the tension feedback value at discrete time n, and N is the number of sampling points.

[0108] In this embodiment, the second processor preprocesses the data further comprising:

[0109] Missing value handling: For example, the data of the tension sensor may be missing, and these missing values ​​need to be filled by interpolation, using the mean, or the median.

[0110] Standardization or normalization: To ensure that the neural network model is not affected by the characteristic scale, the different parameters collected (such as winding speed, coil diameter, foil material thickness, etc.) need to be standardized or normalized so that they are on the same scale to facilitate subsequent comparison and analysis.

[0111] Handling outliers: Tension feedback values ​​may contain outliers, which may have a negative impact on data processing and model training. Statistical methods need to be used to detect and handle them.

[0112] Step S505: The second processor 12 (Linux system) performs feature extraction on the preprocessed tension feedback value and the collected device data to obtain a feature vector set. The feature vector set can include tension time-domain features (overshoot, steady-state error, and adjustment time), tension frequency-domain features, and features of data such as winding and unwinding speed, coil diameter, foil material thickness, and ambient temperature and humidity. Specifically, features closely related to winding tension are extracted, and different feature extraction methods can be used for different types of data.

[0113] For example, the thickness of the foil material has a direct impact on the winding tension, so it is directly input into the neural network model as a feature. At the same time, it is also necessary to consider extracting some derived features, such as the product of the foil thickness and the winding speed (reflecting the stiffness and stress distribution of the material) and the product of the material thickness and the coil diameter (reflecting the degree of material bending).

[0114] Ambient temperature and humidity can affect the properties and mechanical structure of the foil, thereby affecting the winding tension. Therefore, temperature and humidity are directly input into the neural network model as features. In addition, it is also necessary to consider extracting some derived features, such as the product of temperature and humidity (reflecting the combined effect of ambient humidity and temperature) and the product of temperature or humidity and winding speed (reflecting the impact of environmental factors on winding speed).

[0115] For time series features, because data such as winding speed and tension sensor readings change over time, in addition to using the original data as features, some time series features are also extracted. For example, a sliding window method can be used to calculate statistical features such as the average, maximum, minimum, and standard deviation over a period of time.

[0116] Step S506: The second processor 12 (Linux system) uploads the obtained feature vector set to the server. Specifically, the second processor 12 (Linux system) reports the feature vector set to the server for the server to evaluate, train and optimize the neural network model. The reported feature vector set can be tension time domain features (overshoot, steady-state error, adjustment time), tension frequency domain features (main frequency components, and energy distribution at characteristic frequencies), as well as tension set value, current speed of the motor controlled by the winding and unwinding equipment, current neural network model number, width, thickness, material of the coil, current ambient temperature, humidity, finished product inspection results, inspection quality distribution and other data extraction related features.

[0117] Step S507: The second processor 12 obtains from the server a neural network model that is trained and published using the training data based on reinforcement learning, and predicts the first control parameters Kp, Ki, and Kd through the neural network model based on the feature vector set.

[0118] Step S508: The second processor 12 (Linux system) sends the first control parameters Kp, Ki and Kd predicted by the neural network model to the first processor 11 (RTOS system).

[0119] Step S509: The first processor 11 (RTOS system) determines the second control parameter using a PID control algorithm based on the tension setpoint and the first control parameters Kp, Ki, and Kd determined by the neural network model. The PID algorithm continuously measures the error between the actual and desired outputs of the controlled object and adjusts the controller output based on the magnitude of the error, effectively correcting deviations in the controlled object and thus achieving a stable state. In this embodiment, the tension feedback value represents the actual output of the controlled object, the tension setpoint represents the desired output of the controlled object, and the second control parameter is determined based on the error between the tension feedback value and the tension setpoint and the first control parameter.

[0120] Specifically, the first control parameters Kp, Ki and Kd are determined based on the neural network model, and the second control parameter is calculated by the PID control algorithm. The PID control algorithm is as follows:

[0121] Among them, Kp, Ki and Kd represent proportional, integral and differential gains respectively, e(t) represents the tension deviation value at the current moment, represents the rate of change of the deviation value, and u(t) is the second control parameter for tension. Kp determines the control system's response speed to the current tension deviation value. Ki is used to eliminate steady-state errors in the tension of the rewinding and unwinding equipment. Kd is used to predict system behavior. It will issue control signals in advance at the beginning of system changes to prevent the development of tension deviation values.

[0122] Step S510: The first processor 11 (RTOS system) compensates the spindle speed setpoint based on the second control parameter, and adjusts the spindle speed in real time in a closed loop to achieve high-response tension control of the rewinding and unwinding device. In this embodiment, the spindle may be a drive roller of the rewinding and unwinding device.

[0123] Please refer to FIG6 , which is a flow chart of a method for implementing a control system of a battery manufacturing device according to another embodiment of the present invention.

[0124] As shown in FIG6 , the control system of the battery manufacturing equipment of the present invention includes the following steps when applied to the coating machine surface density control:

[0125] Step S601: Obtain the surface density feedback value of the coating machine through the first processor 11 (RTOS system) and filter it. Specifically, a high-precision surface density meter is used to collect the coating surface density feedback value of the electrode in real time, and the coating surface density feedback value is obtained and filtered by the first processor 11 (RTOS system). The filtering process can adopt median filtering, sliding average filtering, etc.

[0126] Step S602: the first processor 11 (RTOS system) sends the filtered coating surface density feedback value to the second processor 12 (Linux system).

[0127] Step S603: The second processor 12 (Linux system) obtains the equipment data of the coating machine. The equipment data can be the viscosity, concentration, and water content of the slurry output by the manufacturing execution system (MES), the tape speed, the set thickness and width of the coating, the coating surface density setting value, the die head temperature, the cavity pressure, the oven temperature output by the coating machine equipment, and the ambient temperature and humidity collected by the factory thermometer / dew point meter.

[0128] Step S604: The second processor 12 (Linux system) preprocesses the acquired device data and the filtered coating surface density feedback value. The preprocessing may be data cleaning, data conversion, etc. Data cleaning includes removing invalid, erroneous or incomplete data and checking the integrity of the data; data conversion includes Fourier transforming the coating surface density feedback value and converting the coating surface density feedback value from the time domain to the frequency domain.

[0129] In this embodiment, the second processor 12 pre-processes the data further including:

[0130] Missing value processing, for example, data such as temperature, humidity, and slurry viscosity may be missing, and these missing values ​​need to be filled by interpolation, using the mean, or median.

[0131] Standardization or normalization: To ensure that the neural network model is not affected by the characteristic scale, the collected data from different devices needs to be standardized or normalized, such as coating speed, slurry viscosity, oven temperature, etc., because these parameters may have different magnitudes and dimensions. After standardization or normalization, these parameters can be compared and analyzed on the same scale and can also be directly used for neural network model training.

[0132] Handling outliers: Outliers may appear in measurement values ​​such as coating surface density feedback, slurry viscosity, and baking temperature. These outliers may have adverse effects on data processing and neural network model training. Statistical methods are needed to detect them and decide whether to delete or replace them.

[0133] Step S605: The second processor 12 (Linux system) analyzes and extracts features from the preprocessed data to obtain a feature vector set. Data analysis includes calculating the mean, median, variance, standard deviation, etc. of the data; analyzing the correlation between the coating surface density and other parameters (such as coating speed, slurry viscosity, oven temperature, etc.); extracting features specifically includes extracting the time domain features of the coating surface density, which include the overshoot, steady-state error, adjustment time, coating machine operating speed, and surface density fluctuation ratio of the surface density adjustment process, or the average surface density in the past 10 minutes; and frequency domain features, which include the periodic variation of the surface density of the coating machine, the intensity of each frequency component, and extracting relevant features of the equipment data to obtain a feature vector set.

[0134] Step S606: The second processor 12 (Linux system) uploads the feature vector set to the server for the server to evaluate, train and optimize the neural network model.

[0135] Step S607: The second processor 12 (Linux system) obtains the neural network model based on reinforcement learning, trained using training data and released from the server, and performs prediction through the neural network model based on the feature vector set, outputs the three fuzzy rule matrices P[m,n], I[m,n] and D[m,n] of the fuzzy control loop of the surface density control system, and sends the fuzzy rule matrix to the first processor 11 (RTOS system).

[0136] Step S608: The first processor 11 (RTOS system) determines the surface density deviation e(t) and the deviation change rate based on the coating surface density feedback value and the surface density setting value. And the surface density deviation e(t) and the deviation change rate Fuzzification is performed to obtain fuzzy sets.

[0137] Step S609: The first processor 11 (RTOS system) performs fuzzy reasoning on the fuzzy rule matrices P[m,n], I[m,n] and D[m,n] and the fuzzy sets to obtain fuzzy output quantities Kp, Ki and Kd.

[0138] Step S610: The first processor 11 (RTOS system) converts the fuzzy outputs Kp, Ki, and Kd into second control parameters through a PID control algorithm. The PID control algorithm is as follows:

[0139] Among them, Kp, Ki and Kd are the proportional, integral and differential gains of fuzzy inference output, e(t) represents the surface density deviation value at the current moment, represents the rate of change of the deviation value, and u(t) is the second control parameter of the coating die head. Kp determines the control system's response speed to the current areal density deviation value. Ki is used to eliminate the steady-state error of the coater's areal density. Kd is used to predict system behavior. It will issue a control signal in advance at the beginning of system changes to prevent the development of areal density deviation values.

[0140] Step S611: The first processor 11 (RTOS system) controls the coating machine based on the second control parameter of the coating die head, thereby completing high-response intelligent control of the surface density of the coating machine.

[0141] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0142] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A control system for battery manufacturing equipment, characterized in that: The control system includes a first processor, a second processor and a neural network processor, wherein the first processor runs a first operating system, and the second processor runs a second operating system; The first processor is used to collect first information; The second processor is used to collect second information, obtain the first information, and extract characteristic information of the first information and the second information, wherein the first information and the second information are associated with information carried by the battery manufacturing equipment; The neural network processor is used to obtain the characteristic information, and obtain a first control parameter based on the characteristic information; The first processor is further used to obtain the first control parameter, determine a second control parameter based on the first control parameter, and control the battery manufacturing equipment based on the second control parameter.

2. The control system of the battery manufacturing equipment according to claim 1, characterized in that: The first processor includes a first data acquisition module; The first data acquisition module acquires first information at a sampling period that is less than a preset threshold.

3. The control system of the battery manufacturing equipment according to claim 1, characterized in that: The first processor also includes a determination module and a control module; The determination module is used to obtain a first control parameter obtained by the neural network processor, and determine a second control parameter based on the first control parameter; The control module is used to control the battery manufacturing equipment based on the second control parameter.

4. The control system of the battery manufacturing equipment according to claim 1, characterized in that: The first processor also includes a data preprocessing module; The data preprocessing module is used to perform filtering processing on the first information.

5. The control system of the battery manufacturing equipment according to claim 1, characterized in that: The second processor includes a second data acquisition module; The second data acquisition module acquires second information at a sampling period greater than or equal to a preset threshold.

6. The control system of the battery manufacturing equipment according to claim 1, characterized in that: The second processor also includes a feature extraction module; The feature extraction module is used to extract features from the first information and the second information to obtain a feature vector set.

7. The control system of battery manufacturing equipment according to claim 1, characterized in that: The second processor also includes a conversion module; The conversion module is used to convert the first information from the time domain to the frequency domain.

8. The control system of battery manufacturing equipment according to claim 1, characterized in that: The second processor also includes a data reporting module. The data reporting module is used to upload training data to the cloud to train the neural network model in the cloud; wherein the neural network model outputs a first control parameter, and the training data is associated with the first information, the second information and the preset first control parameter.

9. The control system of battery manufacturing equipment according to claim 1, characterized in that: The control system also includes a first storage module and a second storage module; The first storage module is used to store the first information and the first control parameter acquired by the first processor; The second storage module is used to store the second information, the feature information extracted by the second processor, and the first control parameter acquired by the neural network processor.

10. The control system of battery manufacturing equipment according to claim 1, characterized in that: The first operating system is an RTOS system; the second operating system is a Linux system.

Images