Spring steel wire cutting command automatic control system based on industrial data analysis

Abstract

The invention relates to the technical field of industrial automatic control, industrial data analysis and artificial intelligence in intelligent manufacturing, in particular to a spring steel wire cutting command automatic control system based on industrial data analysis. Comprising a data acquisition module used for acquiring production takt data; the first feature processing module is used for generating a production takt feature vector; the second feature processing module is used for generating a cutter health degree feature vector; the third feature processing module is used for generating a physical credibility feature vector; the state prediction module is used for predicting probability distribution of future states of the system; the risk quantification module is used for calculating a failure boundary safety distance; the decision generation module is used for generating a macroscopic risk avoidance action; the parameter optimization module is used for optimizing to generate a micro-process parameter vector; and the instruction generation module is used for converting the micro process parameter vector into a machine control command sent to a spring steel wire cutting system. The defect that traditional static parameter control cannot cope with hidden interference sources such as hydraulic oil emulsification is overcome.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation control, industrial data analysis, and artificial intelligence in intelligent manufacturing technology, specifically to an automatic control system for spring steel wire cutting commands based on industrial data analysis. Background Technology

[0002] In the field of industrial automation production, maintaining stable equipment operation and ensuring consistent product quality are core technical requirements in precision machining processes such as spring steel wire cutting. Currently, the control strategy of spring steel wire cutting systems generally relies on static process parameters set based on experience. This control method pre-configures fixed parameters such as cutting speed and feed rate before starting production, and keeps them unchanged within a production cycle. This traditional static parameter control method has obvious limitations when dealing with dynamic changes in the production process. It is difficult to effectively deal with hidden sources of interference caused by the evolution of the equipment's own state, such as progressive wear of cutting tools and emulsification of hydraulic oil. These factors will continue to accumulate and cannot be covered by static parameters. Ultimately, this may lead to batch quality deviations in products or even unexpected equipment downtime, which will seriously affect the continuity and economy of production. While existing technologies may include some basic equipment condition monitoring functions, they generally lack the ability to perform in-depth fusion analysis of multi-source sensor data and predict future state evolution. Specifically, existing methods are deficient in the following aspects: they lack effective modeling methods for accurately quantifying key implicit states such as tool health and oil physical reliability from multimodal data such as high-frequency vibration and oil turbidity; existing systems lack a forward-looking risk assessment mechanism that can predict the probability of equipment failure in the future; and at the decision-making level, they have failed to establish an intelligent decision-making system that can dynamically balance production cycle and equipment safety based on quantified future risks, and autonomously generate optimal macro-level risk avoidance actions and micro-level process parameters. Therefore, how to provide an automatic control method for spring steel wire cutting commands based on industrial data analysis, which can accurately predict and dynamically avoid future risks of the system by integrating multimodal sensor data, thereby transforming passive static control into active adaptive control, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention discloses an automatic control system for spring steel wire cutting commands based on industrial data analysis. Specifically, the technical solution is as follows: An automatic control system for spring steel wire cutting commands based on industrial data analysis includes: The data acquisition module is used to collect multimodal sensor data of the spring steel wire cutting system in real time and acquire production cycle data. The first feature processing module is used to generate a production cycle feature vector based on the production cycle data. The second feature processing module is used to generate a tool health feature vector based on the multimodal sensing data. The third feature processing module is used to generate a physical reliability feature vector based on the multimodal sensing data; The state prediction module is used to combine the production cycle feature vector, the tool health feature vector, and the historical sequence of the physical reliability feature vector to predict the probability distribution of the future state of the system. The risk quantification module is used to calculate the failure boundary safety distance by combining the probability distribution of the future state of the system with the preset failure boundary value. The decision generation module is used to generate macro-risk avoidance actions based on the failure boundary safety distance and the three feature vectors. The parameter optimization module is used to generate a vector of microscopic process parameters in response to the macroscopic risk avoidance action. The instruction generation module is used to convert the microscopic process parameter vector into machine control commands that are sent to the spring steel wire cutting system.

[0004] Preferably, the third feature processing module is specifically used for: Based on the oil sensor readings in the multimodal sensing data and the preset weighted normalization model, the oil emulsification index is calculated. The oil emulsification index is used as a component in generating the physical credibility feature vector.

[0005] Preferably, the second feature processing module is specifically used for: Wavelet packet transform is performed on the vibration signal in the multimodal sensing data to extract the energy proportion of each frequency band; The energy percentage is then used as a component in generating the tool health feature vector.

[0006] Preferably, the state prediction module is specifically used for: A spatiotemporal graph convolutional network is used to perform spatiotemporal dependency modeling on the historical sequences of the production cycle feature vector, the tool health feature vector, and the physical reliability feature vector, and output the hidden state of the system at the current moment. The evolution of the hidden state is modeled using neural stochastic differential equations to calculate the probability distribution of the future state of the system.

[0007] Preferably, the risk quantification module is specifically used for: The hidden state output by the state prediction module is mapped back to the observable feature space through the trained decoder network to determine the future prediction mean and standard deviation of the physical credibility feature. The failure boundary safety distance is determined based on the future prediction mean, the standard deviation, and the preset failure boundary value.

[0008] Preferably, the decision generation module is specifically used for: Determine the normalized utility term calculated from the production cycle feature vector, the tool health feature vector, and the physical reliability feature vector; Determine the exponential penalty term calculated from the failure boundary safety distance; A reward function is constructed by combining the normalized utility term and the exponential penalty term; And based on the reward function, the macro risk aversion action is generated.

[0009] Preferably, the decision generation module is further configured to: When the failure boundary safety distance is lower than the preset minimum safety threshold, the macro risk avoidance action is generated by having the exponential penalty term play a dominant role in the reward function. When the failure boundary safety distance is not lower than the preset minimum safety threshold, the macro-risk avoidance action is generated by the normalized utility term playing a dominant role in the reward function.

[0010] Preferably, the parameter optimization module is specifically used for: Based on the macro-risk avoidance actions generated by the decision generation module as constraints, a multi-objective optimization problem is constructed; The multi-objective gray wolf optimization algorithm is adopted, and the multi-objective optimization problem is solved based on a preset surrogate model to output the microscopic process parameter vector.

[0011] Preferably, an automatic control system for spring steel wire cutting commands based on industrial data analysis further includes: The feedback calibration module is used to calculate the long-term effective output index based on the physical reliability feature vector and the production cycle feature vector; The weights used to construct the reward function are periodically calibrated based on the long-term effective output indicators.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention achieves proactive risk management by predicting the probability of future equipment states. It quantifies failure risks and generates avoidance actions before failures occur, overcoming the limitations of traditional static parameter control in dealing with hidden interference sources such as hydraulic oil emulsification. 2. This invention deeply quantifies subtle, latent conditions such as tool wear and oil contamination by performing wavelet packet transformation on vibration signals and calculating the oil emulsification index. This enables precise perception of the equipment's physical health status, preventing batch defects and unexpected downtime caused by the deterioration of hidden faults. 3. This invention possesses a dual-level closed-loop control capability of tactical response and strategic adaptation. It not only optimizes micro-process parameters in real time based on current risks but also periodically calibrates its decision-making model based on long-term effective output indicators, continuously evolving the system's decision preferences and maximizing long-term benefits. 4. This invention ensures robust prediction results even under non-ideal operating conditions by validating input data, stress-testing the core prediction model, and incorporating protection mechanisms in risk quantification calculations, significantly improving the robustness of the production process. Attached Figure Description

[0013] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. Example 1

[0015] Please see Figure 1 An automatic control system for spring steel wire cutting commands based on industrial data analysis includes: The data acquisition module is used to collect multimodal sensor data of the spring steel wire cutting system in real time and acquire production cycle data. The first feature processing module is used to generate production cycle feature vectors based on production cycle data; The second feature processing module is used to generate tool health feature vectors based on multimodal sensing data; The third feature processing module is used to generate physical credibility feature vectors based on multimodal sensing data; The state prediction module is used to combine the historical sequence of production cycle feature vector, tool health feature vector, and physical reliability feature vector to predict the probability distribution of the future state of the system. The risk quantification module is used to calculate the safe distance of the failure boundary by combining the probability distribution of the future state of the system with the preset failure boundary value. The decision generation module is used to generate macro-risk avoidance actions based on the failure boundary safety distance and three feature vectors; The parameter optimization module is used to generate micro-level process parameter vectors in response to macro-risk avoidance actions. The instruction generation module is used to convert the microscopic process parameter vector into machine control commands to be sent to the spring steel wire cutting system. Specifically, the conversion includes formatting the optimized process parameter vector according to the communication protocol required by the target equipment controller and encapsulating it into directly executable binary or text instruction codes. This embodiment provides an automatic control system for spring steel wire cutting commands based on industrial data analysis. The data acquisition module is responsible for collecting multi-modal sensor data such as high-frequency vibration and oil turbidity, and simultaneously acquiring production cycle data. The first feature processing module generates a production cycle time feature vector based on production cycle time data; the second feature processing module generates a tool health feature vector based on multimodal sensor data; and the third feature processing module generates a physical reliability feature vector based on multimodal sensor data. The state prediction module receives and combines the historical sequence of three feature vectors to predict the probability distribution of the future state of the system. The risk quantification module calculates the safety distance of the failure boundary based on this probability distribution. The decision generation module generates macro-level risk avoidance actions based on the failure boundary safety distance and three feature vectors; the parameter optimization module responds to this action to generate micro-level process parameter vectors; and the instruction generation module finally converts these parameter vectors into machine control commands. This system transforms raw physical signals into precise adaptive control commands through a chain-like data processing flow. It aims to address the shortcomings of traditional static parameter control in dealing with hidden interference sources such as hydraulic oil emulsification, thereby achieving proactive risk management of the cutting process. The current model of this system mainly focuses on internal risks caused by the equipment's own state and production cycle. In future iterations, raw material batch information, environmental temperature and humidity sensor data, etc., can be incorporated as additional inputs into the modeling of the spatiotemporal graph convolutional network, thereby further improving the system's adaptability and robustness to external interference factors. Before feature processing, the data acquisition module or the third feature processing module first verifies the validity of the multimodal sensing data; the verification rules include checking whether the sensor readings are within their preset valid range. The system detects whether there are any abnormalities in the data, such as no change for a long time or severe fluctuations. For sensor data that is identified as invalid or abnormal, the system can fill it with the historical average of the reading or the estimated value based on other relevant sensor readings and generate a system alarm to ensure the robustness of the physical reliability feature vector. Example 2

[0016] The third feature processing module is specifically used for: The oil emulsification index is calculated based on the oil sensor readings in the multimodal sensing data and the preset weighted normalization model. The oil emulsification index is used as a component of the generated physical credibility feature vector. The second feature processing module is specifically used for: Wavelet packet transform is performed on the vibration signal in multimodal sensing data to extract the energy proportion of each frequency band; The energy percentage is used as a component of the generated tool health feature vector. In this embodiment, the second feature processing module and the third feature processing module work together to deeply quantify the latent state of the system. The preset weighted normalization model has weight coefficients that are calibrated based on a large amount of historical data. To clarify this calibration process, it is necessary to clearly distinguish between calibration data variables and model runtime variables. The calibration dataset consists of a set of oil samples. The composition includes a baseline emulsification value for each sample determined through offline chemical analysis. And a set of synchronously acquired sensor reading vectors, weighted It is obtained through least squares regression analysis on a calibrated dataset, with the goal of achieving the oil emulsification index calculated by the following formula. Compared with the baseline emulsification value To minimize the error between the two, the third feature processing module calculates the oil emulsification index using this model. Its formula is: The third feature processing module calculates the oil emulsification index Ie using this model, and the formula is as follows:

[0017] Where Ie is the oil emulsification index, ωi is the weight, Si is the sensor reading, Sn is the lower limit of the range, Sx is the upper limit of the range, and i is the sensor index. It should be noted that the aforementioned linear weighted model for calculating the oil emulsification index provides an efficient and effective approximation method for practical engineering applications. In future iterations, a nonlinear model can also be used to capture more complex coupling relationships between various sensing features, further improving the accuracy of physical reliability features. Meanwhile, the second feature processing module uses wavelet packet transform technology to decompose the vibration signal and extract the energy ratio. This can accurately capture the energy anomalies of the tool in a specific frequency band due to wear or chipping. By accurately quantifying the degree of oil emulsification and the micro-vibration characteristics of the tool, the system gains a fine perception of the physical health of the equipment, avoiding batch defects and unexpected downtime caused by the continuous deterioration of hidden faults. Example 3

[0018] The state prediction module is specifically used for: A spatiotemporal graph convolutional network is used to model the spatiotemporal dependencies of the historical sequences of production cycle time feature vector, tool health feature vector, and physical reliability feature vector, and output the hidden state of the system at the current moment. Neural stochastic differential equations are used to model the evolution of the hidden state in order to solve for the probability distribution of the future state of the system. The risk quantification module is specifically used for: The hidden states output by the state prediction module are mapped back to the observable feature space through the trained decoder network to determine the mean and standard deviation of the future prediction of the physical credibility feature. Based on the future predicted mean and standard deviation and the preset failure boundary value, the safety distance of the failure boundary is determined; In this embodiment, the state prediction module and the risk quantification module together constitute the prediction core unit of the system. The spatiotemporal graph convolutional network is a preset model whose graph structure is predefined according to the physical component association of the cutting system to capture the coupling influence between various features. For example, key components such as the spindle, tool, and hydraulic pump can be used as nodes of the graph, and edges are defined according to the physical connection or functional coupling relationship between them, thereby constructing a model that reflects the topology of the system. Specifically, the adjacency matrix A of the graph is defined asymmetrically based on the energy or information flow direction between components. For example, the edge weight from the hydraulic pump to the spindle is higher than the weight from the spindle to the hydraulic pump. The network employs a stacked structure of multi-layer graph convolutions and gated recurrent units (GRUs) to simultaneously capture spatial dependencies and temporal dynamics; neural stochastic differential equations are used to predict the future evolution of hidden states in the presence of random perturbations, and the specific form of the equation is as follows: The hidden state is Wiener process is Its drift term and diffusion terms Each hidden state is parameterized by an independent multilayer perceptron (MLP) neural network and trained by maximizing the evidence lower bound (ELBO) of the observation sequence. The trained decoder network functions to translate abstract hidden states back into specific physical feature predictions. This decoder network is preferably a multilayer perceptron (MLP) structure, whose input is the future hidden state output by the neural stochastic differential equation. The output is the predicted mean of the observable physical features at the corresponding time. With the predicted standard deviation

[0019] The risk quantification module uses this to predict and calculate the safe distance from the failure boundary. Its physical meaning is how many times the predicted standard deviation the distance between the predicted characteristic mean and the failure boundary is; it is a dimensionless relative index. The preset failure boundary value... These are hard engineering constraints derived from equipment safety manuals or statistical analysis of numerous destructive tests; their calculation formula is: Failure Boundary Safety Distance Failure boundary threshold Forecast Mean (forecast standard deviation) In actual calculations, to prevent the standard deviation of the prediction from being affected... If the value is too small, it will cause instability. Therefore, a very small positive lower limit should be set for it. That is, the actual calculation adopts Ensure a safe distance even when the model predicts a certain altitude. It is still a bounded and meaningful value; This technical solution upgrades the system from passive response to proactive prediction. By quantifying the probability of reaching the failure boundary within a specific future time, it provides a basis for decision-making to take avoidance actions before failure occurs, significantly improving the robustness of the production process. In addition, to ensure the robustness of the core prediction model, a system stress test is conducted on the model before deployment. The test includes injecting simulated extreme noise, long-period flat signals, or abrupt step signals into the historical sequence input, observing and verifying whether the model's hidden state output and future state probability distribution remain within a reasonable range without numerical divergence or collapse, ensuring that the model can still provide robust prediction results under non-ideal operating conditions. Example 4

[0020] The decision generation module is specifically used for: Determine the normalized utility term calculated from the production cycle time feature vector, tool health feature vector, and physical reliability feature vector; Determine the exponential penalty term calculated from the failure boundary safety distance; A reward function is constructed by combining the normalized utility term and the exponential penalty term; And based on the reward function, macro risk aversion actions are generated; The decision generation module is also used for: When the safety distance of the failure boundary is lower than the preset minimum safety threshold, a macro risk avoidance action is generated by making the exponential penalty term play a dominant role in the reward function. When the safety distance from the failure boundary is not lower than the preset minimum safety threshold, a macro-level risk avoidance action is generated by making the normalized utility term play a dominant role in the reward function. In this embodiment, the decision generation module is the system's decision-making unit. Its constructed reward function R aims to balance production efficiency and safety risk, where a preset minimum safety threshold is defined. The weights in the reward function are set based on risk acceptance criteria and historical safe operation data analysis. This is achieved by iterative optimization in a simulation environment or through inverse reinforcement learning, with the goal of maximizing long-term effective output. The normalized utility term is then defined. It is a quantification of different performance dimensions of the system, such as utility items representing production efficiency. The normalized production cycle time feature vector can be directly taken as a numerical value; the utility term representing quality. It can be defined as 1 minus the normalized tool wear or oil emulsification degree; For example, production cycle feature vector Tool health feature vector and physical reliability feature vector Where f and g are preset mapping functions, for example, the utility term can be defined through a specific normalization function. The utility term representing productivity... It can be determined by the characteristics of the production cycle time. It is calculated using the following linear mapping function f:

[0021] in and These are the maximum and minimum production cycle times allowed by the process, respectively. Utility items represent the equipment status. The tool health characteristics can be obtained from the normalized tool health features. Physical credibility characteristics Through weighted average function The calculation shows that: Among them, weight In this way, all utility terms are explicitly mapped to the interval [0,1], with higher values ​​representing higher utility; the formula is: Reward function, Weight, Normalized utility term, Index penalty item Utility item index); Index penalty item The design reflects the avoidance of asymmetric risks, and its value is within the safe distance of the failure boundary. Below the preset minimum safety threshold It grows exponentially, and its specific form can be designed as follows: in This is a positive coefficient used to adjust the severity of the penalty. Its value can be determined through simulation experiments. Under the premise of ensuring safety, a suitable value should be chosen that does not excessively inhibit normal production. For example, it can be set to [value missing]. and Penalty item when decreasing by one standard unit This increases by an order of magnitude; the conditional judgment logic ensures the system's survival priority principle: within the safe zone, the system prioritizes production efficiency; once it enters the dangerous zone, the reward function is dominated by the penalty term, and the system will forcibly execute risk-avoidance actions such as reducing the production cycle or preventative tool changes, thereby avoiding the catastrophic consequences caused by opportunistic production behavior on the verge of failure. Example 5

[0022] The parameter optimization module is specifically used for: Based on the macro-level risk avoidance actions generated by the decision generation module as constraints, a multi-objective optimization problem is constructed. The multi-objective gray wolf optimization algorithm is adopted. This algorithm has the characteristics of fast convergence speed and strong optimization ability when dealing with multi-objective optimization problems with complex constraints. It is suitable for the needs of this system to optimize process parameters in real time. The algorithm solves the multi-objective optimization problem based on the preset surrogate model to output the micro-process parameter vector. The feedback calibration module is used to calculate long-term effective output indicators based on the physical reliability feature vector and the production cycle feature vector. And based on long-term effective output indicators, the weights used to construct the reward function are periodically calibrated; In this embodiment, the parameter optimization module and the feedback calibration module implement a two-level closed-loop control. Specifically, the process of converting macro-risk avoidance actions into constraints involves: if the action is to reduce production cycle time, the upper limit of the production cycle time parameter is set to a safe value lower than the current value in the multi-objective optimization problem; if the action is a preventative tool change, the system generates a maintenance command and pauses optimization, resuming only after the tool change is completed; if the action is to adjust the cutting fluid ratio, the optimization range of the cutting fluid-related parameters is constrained to a more conservative range; the preset surrogate model is a lightweight model trained based on historical data. A neural network is used to quickly predict production results under different combinations of process parameters, supporting the multi-objective gray wolf optimization algorithm for real-time optimization. The input of this surrogate model is a vector of microscopic process parameters including cutting speed and feed rate, and the output is the predicted value of multiple performance indicators such as production cycle time, product dimensional accuracy, and surface roughness under these parameters. The feedback calibration module works on a longer time scale. Its calculated long-term effective output index L is a comprehensive evaluation of the system's long-term performance. The dimension of this index is output quantity, and its physical meaning is the total number of qualified products produced within the total time. Its calculation formula is:

[0023] Where L is the long-term effective output indicator, For defect rate, The production cycle time is t, where t is time and T is the total duration. The feedback calibration module periodically adjusts the weights in the reward function based on the results of the long-term effective output index L. This two-layer feedback mechanism enables the system to not only make the best tactical response to the current working condition, but also to continuously optimize its decision preferences based on the long-term effects, thus realizing the evolution from tactical optimization to strategic adaptation.

[0024] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention; any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0025] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automatic control system for spring steel wire cutting commands based on industrial data analysis, characterized in that, include: The data acquisition module is used to collect multimodal sensor data of the spring steel wire cutting system in real time and acquire production cycle data. The first feature processing module is used to generate a production cycle feature vector based on the production cycle data. The second feature processing module is used to generate a tool health feature vector based on the multimodal sensing data. The third feature processing module is used to generate a physical reliability feature vector based on the multimodal sensing data; The state prediction module is used to combine the production cycle feature vector, the tool health feature vector, and the historical sequence of the physical reliability feature vector to predict the probability distribution of the future state of the system. The risk quantification module is used to calculate the failure boundary safety distance by combining the probability distribution of the future state of the system with the preset failure boundary value. The decision generation module is used to generate macro-risk avoidance actions based on the failure boundary safety distance and the three feature vectors. The parameter optimization module is used to generate a vector of microscopic process parameters in response to the macroscopic risk avoidance action. The instruction generation module is used to convert the microscopic process parameter vector into machine control commands that are sent to the spring steel wire cutting system.

2. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The third feature processing module is specifically used for: Based on the oil sensor readings in the multimodal sensing data and the preset weighted normalization model, the oil emulsification index is calculated. The oil emulsification index is used as a component in generating the physical credibility feature vector.

3. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The second feature processing module is specifically used for: Wavelet packet transform is performed on the vibration signal in the multimodal sensing data to extract the energy proportion of each frequency band; The energy percentage is then used as a component in generating the tool health feature vector.

4. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The state prediction module is specifically used for: A spatiotemporal graph convolutional network is used to perform spatiotemporal dependency modeling on the historical sequences of the production cycle feature vector, the tool health feature vector, and the physical reliability feature vector, and output the hidden state of the system at the current moment. The evolution of the hidden state is modeled using neural stochastic differential equations to calculate the probability distribution of the future state of the system.

5. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The risk quantification module is specifically used for: The hidden state output by the state prediction module is mapped back to the observable feature space through the trained decoder network to determine the future prediction mean and standard deviation of the physical credibility feature. The failure boundary safety distance is determined based on the future prediction mean, the standard deviation, and the preset failure boundary value.

6. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The decision generation module is specifically used for: Determine the normalized utility term calculated from the production cycle feature vector, the tool health feature vector, and the physical reliability feature vector; Determine the exponential penalty term calculated from the failure boundary safety distance; A reward function is constructed by combining the normalized utility term and the exponential penalty term; And based on the reward function, the macro risk aversion action is generated.

7. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 6, characterized in that, The decision generation module is also used for: When the failure boundary safety distance is lower than the preset minimum safety threshold, the macro risk avoidance action is generated by having the exponential penalty term play a dominant role in the reward function. When the failure boundary safety distance is not lower than the preset minimum safety threshold, the macro-risk avoidance action is generated by the normalized utility term playing a dominant role in the reward function.

8. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 1, characterized in that, The parameter optimization module is specifically used for: Based on the macro-risk avoidance actions generated by the decision generation module as constraints, a multi-objective optimization problem is constructed; The multi-objective gray wolf optimization algorithm is adopted, and the multi-objective optimization problem is solved based on a preset surrogate model to output the microscopic process parameter vector.

9. The automatic control system for spring steel wire cutting commands based on industrial data analysis according to claim 6, characterized in that, Also includes: The feedback calibration module is used to calculate the long-term effective output index based on the physical reliability feature vector and the production cycle feature vector; The weights used to construct the reward function are periodically calibrated based on the long-term effective output indicators.

Images