Intelligent waste removal control method and system for a stamping die

Abstract

The application discloses a stamping die intelligent waste removal control method and system, and relates to the technical field of intelligent control. The method comprises the following steps: modeling the waste accumulation state of each monitoring area as a real-time state variable, dynamically estimating the probability distribution of the real-time state variable, and obtaining each state probability distribution; calculating the probability of waste accumulation exceeding a safety threshold in the next moment for each monitoring area, and calculating decision entropy based on all monitoring areas; when the decision entropy is greater than a preset entropy threshold, generating at least two candidate removal actions in a competitive relationship with each other, simulating the waste accumulation state evolution in the future prediction time domain after the execution of each candidate removal action by using each state probability distribution, and determining an optimal removal action by predicting the future decision entropy corresponding to each candidate removal action, and outputting a control instruction to drive an execution mechanism to execute. The technical problem of low waste cleaning efficiency in the prior art is solved, and the technical effect of improving waste cleaning efficiency is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to an intelligent waste removal control method and system for stamping dies. Background Technology

[0002] During the stamping process, stamping dies continuously generate scrap, debris, and cutting residue. If this waste is not removed in time, it can accumulate in the die's working area, leading to problems such as die jamming, stamping deviation, product damage, and even equipment shutdown. Therefore, real-time cleaning of waste inside the die is necessary. Current technologies typically employ fixed-time interval blowing, mechanical pushing, or trigger-based control based on a single threshold. These methods rely heavily on empirical parameters and lack the ability to dynamically perceive and predict the accumulation status of waste. When stamping cycle time, material thickness, or waste generation rate changes, traditional control methods struggle to adjust the cleaning strategy promptly, resulting in delayed cleaning actions, frequent false triggers, or excessive waste accumulation in localized areas. This not only affects waste cleaning efficiency but also reduces the operational stability of the stamping die and production continuity. Summary of the Invention

[0003] This application provides a method and system for intelligent waste removal control of stamping dies, which solves the technical problems of slow response, low cleaning efficiency and easy accumulation and blockage of waste in the existing stamping die waste removal process.

[0004] A first aspect of this application provides an intelligent waste removal and control method for stamping dies, the method comprising:

[0005] Real-time data collection of waste status observations in various monitoring areas within the stamping die is performed. The waste accumulation status in each monitoring area is modeled as a real-time state variable. An ensemble Kalman filter is used to dynamically estimate the probability distribution of the real-time state variable, resulting in a probability distribution for each state. Based on the probability distributions, the probability that the waste accumulation in each monitoring area will exceed a safety threshold in the next moment is calculated, and a decision entropy is calculated based on all monitoring areas. When the decision entropy is greater than a preset entropy threshold, at least two competing candidate removal actions are generated. Using the probability distributions, the evolution of the waste accumulation status in the future prediction time domain after executing each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and a control command is output to drive the actuator to execute.

[0006] A second aspect of this application provides an intelligent waste removal control system for stamping dies, the system comprising: State estimation module: Real-time acquisition of waste state observation data in each monitoring area within the stamping die, modeling the waste accumulation state in each monitoring area as a real-time state variable, and using ensemble Kalman filtering to dynamically estimate the probability distribution of the real-time state variable to obtain the probability distribution of each state; Calculation module: Calculates the probability that the waste accumulation in each monitoring area will exceed the safety threshold in the next moment based on the probability distribution of each state, and calculates the decision entropy based on all monitoring areas; Control module: When the decision entropy is greater than a preset entropy threshold, generates at least two competing candidate removal actions, uses the probability distribution of each state to simulate the evolution of the waste accumulation state in the future prediction time domain after executing each candidate removal action, and determines the optimal removal action by predicting the future decision entropy corresponding to each candidate removal action, and outputs control commands to drive the actuator to execute.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, real-time observation data of the scrap status in each monitoring area within the stamping die is collected. The scrap accumulation state in each monitoring area is modeled as a real-time state variable. An ensemble Kalman filter is used to dynamically estimate the probability distribution of the real-time state variable, obtaining the probability distribution of each state. Then, based on the probability distribution of each state, the probability that the scrap accumulation in each monitoring area will exceed the safety threshold in the next moment is calculated, and the decision entropy is calculated based on all monitoring areas. Finally, when the decision entropy is greater than a preset entropy threshold, at least two competing candidate removal actions are generated. Using the probability distribution of each state, the evolution of the scrap accumulation state in the future prediction time domain after executing each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and a control command is output to drive the actuator to execute. This solves the technical problems of slow response, low cleaning efficiency, and easy scrap accumulation and blockage in the existing stamping die scrap cleaning process, achieving the technical effects of improving scrap cleaning efficiency, enhancing the timeliness of scrap removal, and ensuring the continuous and stable operation of the stamping die. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the intelligent waste removal control method for stamping dies provided in the embodiments of this application; Figure 2 This is a schematic diagram of the intelligent waste removal control system for stamping dies provided in an embodiment of this application.

[0010] Figure labeling: State estimation module 11, calculation module 12, control module 13. Detailed Implementation

[0011] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0012] Example 1, as Figure 1 As shown, this application provides an intelligent waste removal control method for stamping dies, wherein the method includes: Real-time data collection of waste status observation in each monitoring area within the stamping die is performed. The waste accumulation status in each monitoring area is modeled as a real-time state variable. An ensemble Kalman filter is used to dynamically estimate the probability distribution of the real-time state variable to obtain the probability distribution of each state.

[0013] Visual sensors, laser rangefinders, vibration sensors, or pressure sensors are deployed in the waste material sliding channel, the edge area of ​​the die cavity, the material discharge port area, and the waste material concentration area of ​​the stamping die, respectively, to synchronously collect waste material status observation data in each monitoring area according to a preset sampling period. The waste material status observation data includes at least the waste material accumulation height, waste material coverage area, waste material movement speed, waste material sliding frequency, and regional vibration intensity.

[0014] The collected waste status observation data is processed for time synchronization, outlier removal, and normalization. A regional status vector is constructed according to the monitoring area number, and the current waste accumulation degree of each monitoring area is represented as a real-time status variable. Where i represents the monitoring area number, and k represents the current stamping time. This is used to characterize the waste accumulation state of the i-th monitoring area at time k; after defining the state variables, a waste accumulation state-space model is established: ,in, Let A represent the state vector composed of all monitored areas, and let A represent the scrap accumulation state transition matrix, which describes the natural accumulation relationship of scrap in continuous stamping cycles. B represents the control input vector consisting of stamping frequency, blanking rate, and cleaning action; B represents the input action matrix, used to describe the degree of influence of the control input on the waste accumulation state. This represents the process noise term that follows a Gaussian distribution; simultaneously, a corresponding observation model is established: in, Let H represent the real-time acquired waste state observation vector, and let H represent the mapping matrix from the state space to the observation space. This represents the observation noise term. Subsequently, N set members are initialized, each corresponding to a set of possible scrap state variable values. Centered on the initial scrap state mean, and combined with a preset covariance matrix, perturbation sampling is performed according to a multidimensional Gaussian distribution to form an initial state sample set. At the arrival of each stamping cycle, the state-space model is used to perform time updates on all set members, predicting the scrap accumulation state at the next moment based on the current stamping frequency and scrap generation rate, thus obtaining a predicted state set. Then, the real-time acquired observation data is input into the observation model, compared with each predicted state in the predicted state set, and the corresponding observation residuals are calculated. Finally, the Kalman gain matrix is ​​calculated based on the statistical dispersion among all set members. Where K represents the Kalman gain matrix, P represents the predicted state covariance matrix, and R represents the observation noise covariance matrix. The observation matrix is ​​transposed; the state of each set member is corrected using the Kalman gain matrix to obtain the posterior state set, and statistical analysis is performed on the posterior state set to calculate the mean, variance and covariance of the state variables of each monitoring area, forming the state probability distribution of the corresponding monitoring area. The state probability distribution is used to characterize the uncertainty of the current waste accumulation state and the future accumulation trend.

[0015] Furthermore, ensemble Kalman filtering is used to dynamically estimate the probability distribution of the real-time state variables, resulting in the probability distributions of each state, including: Initialize N set members, each representing a possible value of a set of state variables, and assign an initial covariance matrix; at a fixed time in each stamping cycle, use the stamping frequency and scrap generation rate model to update the state variables of each set member over time to obtain a predicted state set; match the real-time state variables with the predicted state set, calculate the observation residual of each set member, and correct the predicted state according to the Kalman gain to obtain a posterior state set; statistically analyze the mean and covariance of the state variables of each monitoring area from the posterior state set to obtain the probability distribution of each state.

[0016] First, an initial state set is established based on historical scrap accumulation data of each monitoring area within the stamping die. The average historical scrap accumulation amount of each monitoring area is used as the initial state center, and N set members are generated based on the historical fluctuation range. Each set member represents a possible state of current scrap accumulation. At a fixed time in each stamping cycle, the real-time stamping frequency of the current stamping machine, the average blanking amount corresponding to a single stamping, and the historical scrap distribution of each monitoring area are read to establish a scrap generation rate model. The scrap generation rate of the i-th monitoring area in the current stamping cycle is expressed as: ,in, This represents the waste generation rate of the i-th monitoring area at time k. Indicates the current stamping frequency. This indicates the average amount of scrap generated in a single stamping operation. This represents the proportion of waste falling into the i-th monitoring area; subsequently, based on the waste generation rate, a time update is performed on each set member, adding the newly added waste amount to the waste accumulation state of the previous time step, and adding a random perturbation term to simulate random changes such as waste slippage, rebound, and local blockage. The state update process is represented as follows: ,in, This represents the predicted state of the j-th set member at time k. Indicates the state at the previous moment. This represents the waste generation rate vector. This indicates the current stamping cycle time. This represents a random disturbance term; further, real-time data on actual waste observed in each monitoring area is collected, and the actual waste observed data is compared one by one with the prediction results in the prediction state set. The difference between the actual observed value and the predicted value is taken as the observation residual of the corresponding set member. ,in, Let the observed residuals of the j-th set member be denoted as . Let H represent the real-time observation data vector, and H represent the state mapping matrix, used to map the predicted state to the observation space. A large observation residual indicates a significant deviation between the current predicted state and the actual waste state. Subsequently, the prediction covariance matrix is ​​calculated based on the dispersion of the predicted state set, and the Kalman gain is calculated in conjunction with the observation noise covariance matrix. ,in, Represents the Kalman gain matrix. Let R represent the predicted state covariance matrix and R represent the observation noise covariance matrix. The predicted state is corrected using the Kalman gain to obtain the posterior state set. Statistical analysis is performed on the state variables in the posterior state set to calculate the mean and covariance of the state variables in each monitoring area, thereby obtaining the state probability distribution corresponding to each monitoring area, which is used to characterize the current degree of waste accumulation and its uncertainty.

[0017] Furthermore, initialize N collection members, including: The initial scrap state of each monitoring area when the stamping die is unloaded is obtained, the initial variance is set for each state variable, and a diagonal covariance matrix is ​​constructed; with the initial mean as the center, deterministic sampling is performed on N set members according to a multidimensional normal distribution, and the N sets of state variable values ​​obtained by sampling are taken as the N set members.

[0018] When the stamping die is stopped and in an unloaded state, initial scrap status detection is performed on each monitoring area within the die. Residual scrap status information for each monitoring area is collected using visual sensors, laser rangefinders, or pressure sensors. The scrap accumulation height, coverage area, or residual mass for each monitoring area is used as the initial state variable. For monitoring areas with no obvious residual scrap, the corresponding state variable is initialized to zero. For monitoring areas with a small amount of residual scrap, the corresponding actual detection value is used as the initial state value. Subsequently, based on the fluctuation range of scrap status in each monitoring area during historical operation, an initial variance is set for each state variable. Monitoring areas with larger scrap status fluctuations correspond to larger initial variances, while those with more stable scrap statuses correspond to smaller initial variances. The regions correspond to smaller initial variances, and the initial variances corresponding to each state variable are arranged on the main diagonal of the covariance matrix to construct an initial diagonal covariance matrix, representing the independence between the initial states of each monitoring region. Furthermore, using the initial mean vector formed by each state variable as the sampling center, deterministic sampling is performed on the state space according to a multidimensional normal distribution. This deterministic sampling employs a sampling method that is uniformly distributed in the probability space, ensuring that the generated samples cover different possible waste accumulation states and avoiding excessive concentration of samples due to random sampling. The N sets of sampled state variable values ​​are then used as N set members, each set member corresponding to a possible initial waste accumulation state, for subsequent state prediction and probability distribution updates in ensemble Kalman filtering.

[0019] Furthermore, at a fixed time point in each stamping cycle, the state variables of each set member are updated over time using a model based on stamping frequency and scrap generation rate, resulting in a predicted state set, including: For each set member, read the real-time stamping frequency of the current stamping press and the scrap status of a single feeding; based on the location of each monitoring area, determine the proportion coefficient of scrap falling into each monitoring area through historical feeding data; according to the stamping cycle step size, use a discretization form to calculate the increment of scrap status in each set member, the increment includes a deterministic generating term and a random fluctuation term following a Gaussian distribution, and add the increment to the state variable of the previous moment to obtain the predicted state set.

[0020] At the start of each stamping cycle, the real-time stamping frequency of the current stamping machine, the average blanking amount corresponding to a single stamping of the die, and the scrap removal status of the current actuator are read for each set member. The real-time stamping frequency is obtained in real time by the stamping machine encoder or controller, and the scrap status of a single blanking is obtained through historical production statistics or online visual inspection results. Subsequently, based on the positional relationship of each monitoring area inside the die, and combined with historical blanking trajectory data, the distribution of scrap falling into each monitoring area is statistically analyzed, and the scrap falling ratio coefficient corresponding to each monitoring area is calculated. Monitoring areas located on the main scrap falling path correspond to a larger ratio coefficient, while monitoring areas located in edge areas or secondary falling areas correspond to a smaller ratio coefficient, to characterize the different monitoring areas' reception of the overall scrap. The system assesses the scrap capacity; further, according to the discrete time step corresponding to the stamping cycle, it performs discretized time updates on the state variables of each set member. The scrap state increment is composed of a deterministic scrap generation term and a random fluctuation term. The deterministic scrap generation term is calculated from the real-time stamping frequency, the amount of scrap generated per cycle, and the proportional coefficient of the corresponding monitoring area, and is used to characterize the growth trend of scrap accumulation under normal stamping conditions. The random fluctuation term simulates random changes caused by stamping vibration, scrap bounce, airflow disturbance, and irregular material fracture, and is set as random noise following a Gaussian distribution. Subsequently, the scrap state increment is superimposed on the state variables of the corresponding set member at the previous time step to complete the state prediction update under the current stamping cycle. The state update process satisfies the following relationship: ,in, This represents the waste state variable of the i-th monitoring area at time k. This represents the deterministic waste increment calculated by the stamping frequency and waste generation rate model. This represents a random fluctuation term that follows a Gaussian distribution. After completing the above time update for all set members, we obtain the predicted state set corresponding to the current stamping cycle, which is used for subsequent observation correction and state probability distribution update.

[0021] Furthermore, the real-time state variables are matched with the predicted state set, the observation residuals of each set member are calculated, and the predicted states are corrected according to the Kalman gain to obtain the posterior state set, including: Each predicted state in the predicted state set is mapped to the observation space, and the residual between each state and the real-time state variable is calculated. The Kalman gain matrix is ​​estimated based on the degree of dispersion among the members of the predicted state set. The gain is multiplied by the residual and then superimposed on each predicted member. The variance of all superposition results is clipped to obtain the posterior state set.

[0022] First, each predicted state in the predicted state set is mapped to an observation space, which uniformly describes the waste state information actually detected by the sensor. During the mapping process, according to a preset observation mapping relationship, the waste accumulation height, coverage area, and regional waste density in the predicted state are converted into observations consistent with the real-time sensor output. Then, the observations corresponding to each predicted state are compared with the real-time state variables acquired in real time, and the corresponding observation residuals are calculated. The observation residuals characterize the degree of deviation between the predicted state and the actual observed state; a large residual indicates a significant difference between the current predicted state and the actual waste accumulation situation. Further, the prediction covariance matrix is ​​estimated based on the dispersion among the members of the predicted state set, and the Kalman gain matrix is ​​calculated in conjunction with the observation noise covariance matrix. The greater the dispersion among the set members, the higher the uncertainty of the current prediction, and the greater the correction weight for the real-time observation data. The Kalman gain is calculated as follows: in, This represents the Kalman gain matrix at time k. Let H represent the predicted state covariance matrix, H represent the observation mapping matrix, and R represent the observation noise covariance matrix. Then, the Kalman gain is used to perform state correction on each predicted member. The gain is multiplied by the corresponding observation residual and then superimposed onto the original predicted state to reduce the deviation between the predicted state and the actual observed state. After all predicted members are corrected, variance pruning is performed on the correction results. Variance pruning is used to limit state divergence caused by abnormal corrections. Preset upper and lower limits constrain the range of changes in the corresponding state variables of each monitoring area, and truncate or shrink abnormal set members that exceed the variance threshold to avoid excessive influence of a single abnormal observation on the overall state distribution. Finally, all set members that have been corrected and variance pruned are output as the posterior state set for subsequent state probability distribution statistics and waste accumulation trend prediction.

[0023] Calculate the probability that waste accumulation in each monitoring area will exceed the safety threshold at the next moment based on the probability distribution of each state, and calculate the decision entropy based on all monitoring areas.

[0024] Furthermore, based on the probability distribution of each state, the probability that waste accumulation in each monitoring area will exceed the safety threshold in the next moment is calculated, and the decision entropy is calculated based on all monitoring areas, including: From the probability distributions of each state, the waste accumulation state values ​​of each monitoring area are extracted to form a sample set. For each monitoring area, the ratio of the number of waste accumulation state values ​​greater than the preset safety threshold to the total number of samples is counted as the probability of exceeding the limit at the next moment. The probability of exceeding the limit of each monitoring area is normalized, and the decision entropy is calculated using the normalized probability values.

[0025] First, waste accumulation state values ​​are extracted from the state probability distributions corresponding to each monitoring area, constructing a state sample set for each monitoring area. Each state sample corresponds to a member of the Kalman filter posterior state set, representing a possible waste accumulation state in the current monitoring area at the next moment. Then, a preset safety threshold is read for each monitoring area. This safety threshold is pre-set based on the maximum allowable waste accumulation height of the mold, the waste coverage area, or the degree of blockage in the waste channel. Further, the state sample sets for the corresponding monitoring areas are traversed, and waste accumulation state values ​​greater than the preset safety threshold are counted. The number of samples for each value is calculated, and the ratio between this number and the total number of samples is used as the probability that the corresponding monitoring area will experience excessive waste accumulation in the next moment. A higher probability of exceeding the limit indicates a higher risk of waste blockage or mold malfunction in the subsequent stamping cycle. After obtaining the excess probabilities for all monitoring areas, normalization is performed on each probability to ensure the sum of probabilities for all monitoring areas is 1, thus eliminating the impact of differences in probability scales between different monitoring areas on subsequent overall decision analysis. Finally, the overall decision entropy of the system is calculated using the normalized excess probabilities of each monitoring area. Where H represents the decision entropy, and M represents the total number of monitored areas. The normalized probability of exceeding the limit for the i-th monitoring area is represented. The decision entropy is used to characterize the uncertainty of the risk distribution of waste accumulation in the current stamping die. When multiple monitoring areas have high risk of exceeding the limit at the same time and the risk distribution is relatively discrete, the decision entropy increases. When the waste risk is concentrated in a small number of monitoring areas or the overall risk is low, the decision entropy decreases, thereby providing a decision basis for the selection of subsequent waste removal actions and the optimization of control strategies.

[0026] When the decision entropy is greater than the preset entropy threshold, at least two candidate removal actions that are in competition with each other are generated. Using the probability distribution of each state, the evolution of the waste accumulation state in the future prediction time domain after each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and the control command is output to drive the execution mechanism to execute.

[0027] When the system detects that the current decision entropy exceeds the preset entropy threshold, it determines that the current waste accumulation state in the stamping die has high uncertainty or potential blockage risk, and initiates a predictive removal decision process. Subsequently, based on the current stamping cycle, the degree of waste accumulation in each monitored area, and the operating status of the actuator, at least two competing candidate removal actions are generated from the preset removal action database. Each candidate removal action includes one or more adjustable parameter combinations among the following: the start / stop mode of the removal equipment, action intensity, air knife pressure, blowing duration, pulsation frequency, or mechanical pushing speed. Different candidate removal actions correspond to different waste removal capabilities and energy consumption levels. Furthermore, the probability distribution of the current state corresponding to each monitored area is used as the initial condition for future state prediction, and the control parameters corresponding to each candidate removal action are substituted into the waste state space model to perform a rolling evolution simulation of the waste accumulation state under multiple continuous stamping cycles in the future prediction time domain. The prediction time domain length is determined based on the stamping frequency, die working cycle, and... The waste accumulation rate is preset. During the simulation, the waste accumulation state changes in each monitoring area are calculated at each prediction time, and the corresponding state probability distribution is updated synchronously to reflect the impact of different candidate removal actions on the future waste evolution trend. Subsequently, the future decision entropy is calculated for the future prediction results corresponding to each candidate removal action. The future decision entropy is used to characterize the uncertainty of the overall waste risk distribution of the system after executing the corresponding candidate removal action. If the future decision entropy corresponding to a certain candidate removal action is small, it means that the action can more effectively reduce the waste accumulation risk and improve the system state stability. Finally, the future decision entropy corresponding to all candidate removal actions is compared, and the candidate removal action with the smallest future decision entropy is selected as the optimal removal action. The corresponding control command is generated according to the optimal removal action and sent to the actuators such as air knife, blowing mechanism, mechanical pushing mechanism or vibration removal mechanism to control the waste removal, thereby realizing the predictive dynamic adjustment of the waste accumulation state of stamping die.

[0028] Furthermore, when the decision entropy is greater than a preset entropy threshold, at least two competing candidate removal actions are generated. Using the probability distributions of each state, the evolution of the waste accumulation state in the future prediction time domain after executing each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, including: Based on the current stamping cycle and waste status observation data, at least two candidate removal actions are generated. Each action includes a set of parameters from the air knife pulse cycle, opening duration, or air pressure amplitude. Using the probability distribution of each state as the initial distribution, for each candidate removal action, the corresponding control parameters are substituted into the state space model to simulate and predict a specified number of stamping cycles in advance, obtaining the predicted waste status distribution of each monitoring area at the prediction time domain endpoint. The future decision entropy corresponding to each candidate removal action is calculated from the predicted waste status distribution. The future decision entropy of each candidate removal action is compared, and the action with the smallest entropy value is selected as the optimal removal action.

[0029] When the current decision entropy exceeds the preset entropy threshold, the system first reads the real-time stamping cycle of the current stamping press, the waste status observation data of each monitored area, and the current operating status of the actuator. Based on the preset candidate action generation rules, at least two candidate removal actions are constructed. Each candidate removal action includes a set of parameters from the air knife pulse cycle, opening duration, and air pressure amplitude. There is at least one difference in control parameters between different candidate actions to form different waste removal intensities and removal rhythms. For example, one candidate action uses a short-cycle high-frequency pulse mode, and the other candidate action uses a long-cycle high-pressure continuous purging mode. Subsequently, the current state probability distribution corresponding to each monitored area is used as the initial distribution for future state evolution, and a future prediction process is executed for each candidate removal action. During the prediction process, the air knife pulse cycle, opening duration, and air pressure amplitude parameters in the corresponding candidate actions are substituted into the waste state space model to correct the waste removal efficiency and waste migration speed terms, reflecting the dynamic impact of different removal actions on the waste accumulation state. Furthermore, according to the preset prediction time domain length, the changes in the waste accumulation state of each monitoring area under multiple consecutive stamping cycles are simulated forward. The waste state distribution is updated synchronously at each prediction step, and the predicted waste state distribution for each monitoring area is output at the end of the prediction time domain. Subsequently, based on the predicted waste state distribution, the probability of waste exceeding the limit in each monitoring area at a future time is calculated, and the probability of exceeding the limit in all monitoring areas is normalized. The future decision entropy of the corresponding candidate removal action is calculated using the normalized probability values. Where H represents the future decision entropy, Let M represent the normalized out-of-limit probability corresponding to the i-th monitoring area, and M represent the total number of monitoring areas. The future decision entropy is used to characterize the uncertainty of the overall waste accumulation risk of the system after executing the corresponding candidate removal action. Finally, the future decision entropy corresponding to all candidate removal actions is compared, and the candidate removal action with the smallest future decision entropy is selected as the optimal removal action. The corresponding control command is generated and sent to the air knife actuator to control the waste removal, thereby reducing the risk of future waste accumulation exceeding the limit and improving the operating stability of the stamping die.

[0030] Furthermore, based on the current stamping cycle time and scrap condition observation data, at least two candidate removal actions are generated, including: The activity level of the waste pile is determined based on the waste status observation data; the activity level of the waste pile and the current stamping cycle are combined to select at least two candidate removal actions with the highest matching degree from the removal action database.

[0031] First, real-time waste status observation data collected from each monitoring area is read. This data includes waste accumulation height, waste coverage area, waste growth rate, waste residence time, and regional blockage frequency. Then, regional clustering analysis is performed on the waste status observation data from each monitoring area to calculate the waste accumulation trend index. This index characterizes the activity level of waste growth in the corresponding monitoring area across multiple stamping cycles. Further, the waste accumulation activity is determined based on the waste accumulation trend index. This activity characterizes the location distribution of the main waste accumulation areas. When the waste accumulation height growth rate, waste residence time, and blockage frequency of a monitoring area simultaneously exceed a preset threshold, the area is identified as a high-activity area. When multiple adjacent monitoring areas simultaneously meet the high-activity condition, they are merged into a key waste accumulation area. Finally, the current stamping cycle parameters are read, including the number of stampings per unit time, the duration of a single stamping, and the frequency of continuous stamping. The frequency is considered, and candidate actions are matched against a pre-defined cleaning action database based on the activity level of the waste pile. This database stores multiple sets of historical cleaning action parameter combinations and corresponding applicable working conditions. Each cleaning action set includes one or more parameters from the following: air knife pulse cycle, opening duration, air pressure amplitude, blowing direction, and execution sequence. Furthermore, based on the location, accumulation intensity, and current stamping cycle of the current waste pile activity area, the action matching degree between each cleaning action and the current working condition is calculated. This matching degree characterizes the adaptability of the corresponding cleaning action to the current waste accumulation state. When a candidate action is more suitable for high-frequency stamping conditions, its matching weight under high stamping cycle conditions is increased. When a candidate action is more suitable for concentrated waste accumulation in a localized area, its matching weight under high-activity area conditions is increased. Finally, the action matching degrees are sorted from highest to lowest, and at least two candidate cleaning actions with the highest matching degrees are selected as inputs for subsequent prediction of future waste state evolution and optimal action decision-making.

[0032] Furthermore, when the decision entropy is greater than a preset entropy threshold, and the future decision entropy corresponding to each predicted candidate clearing action is greater than the preset entropy threshold, an early warning signal of an unrecoverable accumulation trend is output.

[0033] When the system detects that the real-time decision entropy of the current stamping die exceeds the preset entropy threshold, it performs future waste accumulation state evolution prediction for all candidate removal actions and calculates the corresponding future decision entropy. If the future decision entropy corresponding to all candidate removal actions is consistently higher than the preset entropy threshold, it is determined that the current waste accumulation state has exceeded the normal removal capacity of the actuator and there is an irreversible accumulation trend. The irreversible accumulation trend indicates that even after performing different removal actions, each monitoring area in the future prediction time domain still has a high risk of waste exceeding limits, and the waste risk distribution remains in a state of high uncertainty. Further risk analysis is performed on the predicted waste state of each monitoring area. When multiple key monitoring areas simultaneously experience continuous growth in waste accumulation, increased probability of waste channel blockage, or continuous decline in waste removal efficiency, [further analysis is needed]. The system first raises the level of the irreversible accumulation trend assessment. Then, based on the assessment result, a warning signal is generated, including at least audible and visual alarm information, control system alarm information, and equipment linkage control information. The warning signal is then sent to the press controller, human-machine interface terminal, and upper-level monitoring system, triggering corresponding linkage measures according to a preset safety strategy. When the irreversible accumulation trend reaches a first-level risk, a warning prompt suggesting shutdown and cleaning is output. When the irreversible accumulation trend reaches a second-level risk, the stamping frequency is automatically reduced and the number of consecutive stampings is limited. When the irreversible accumulation trend continues to worsen or a mold jamming risk is predicted, a mold maintenance warning is output, and the press equipment shutdown control is triggered to prevent continuous waste accumulation from causing mold damage, product damage, or abnormal equipment shutdown.

[0034] In summary, the embodiments of this application have at least the following technical effects: First, real-time observation data of the scrap status in each monitoring area within the stamping die is collected. The scrap accumulation state in each monitoring area is modeled as a real-time state variable. An ensemble Kalman filter is used to dynamically estimate the probability distribution of the real-time state variable, obtaining the probability distribution of each state. Then, based on the probability distribution of each state, the probability that the scrap accumulation in each monitoring area will exceed the safety threshold in the next moment is calculated, and the decision entropy is calculated based on all monitoring areas. Finally, when the decision entropy is greater than a preset entropy threshold, at least two competing candidate removal actions are generated. Using the probability distribution of each state, the evolution of the scrap accumulation state in the future prediction time domain after executing each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and a control command is output to drive the actuator to execute. This solves the technical problems of slow response, low cleaning efficiency, and easy scrap accumulation and blockage in the existing stamping die scrap cleaning process, achieving the technical effects of improving scrap cleaning efficiency, enhancing the timeliness of scrap removal, and ensuring the continuous and stable operation of the stamping die.

[0035] Example 2, based on the same inventive concept as the intelligent waste removal control method for stamping dies in the foregoing examples, such as... Figure 2 As shown, this application provides an intelligent waste removal control system for stamping dies, wherein the system includes: State estimation module 11: Real-time acquisition of waste state observation data in each monitoring area within the stamping die, modeling the waste accumulation state in each monitoring area as a real-time state variable, and using ensemble Kalman filtering to dynamically estimate the probability distribution of the real-time state variable to obtain the probability distribution of each state; Calculation module 12: Calculates the probability that the waste accumulation in each monitoring area will exceed the safety threshold at the next moment based on the probability distribution of each state, and calculates the decision entropy based on all monitoring areas; Control module 13: When the decision entropy is greater than a preset entropy threshold, generates at least two competing candidate removal actions, uses the probability distribution of each state to simulate the evolution of the waste accumulation state in the future prediction time domain after executing each candidate removal action, and determines the optimal removal action by predicting the future decision entropy corresponding to each candidate removal action, and outputs control commands to drive the actuator to execute.

[0036] Furthermore, the state estimation module 11 is used to perform the following method: Initialize N set members, each representing a possible value of a set of state variables, and assign an initial covariance matrix; at a fixed time in each stamping cycle, use the stamping frequency and scrap generation rate model to update the state variables of each set member over time to obtain a predicted state set; match the real-time state variables with the predicted state set, calculate the observation residual of each set member, and correct the predicted state according to the Kalman gain to obtain a posterior state set; statistically analyze the mean and covariance of the state variables of each monitoring area from the posterior state set to obtain the probability distribution of each state.

[0037] Furthermore, the state estimation module 11 is used to perform the following method: The initial scrap state of each monitoring area when the stamping die is unloaded is obtained, the initial variance is set for each state variable, and a diagonal covariance matrix is ​​constructed; with the initial mean as the center, deterministic sampling is performed on N set members according to a multidimensional normal distribution, and the N sets of state variable values ​​obtained by sampling are taken as the N set members.

[0038] Furthermore, the state estimation module 11 is used to perform the following method: For each set member, read the real-time stamping frequency of the current stamping press and the scrap status of a single feeding; based on the location of each monitoring area, determine the proportion coefficient of scrap falling into each monitoring area through historical feeding data; according to the stamping cycle step size, use a discretization form to calculate the increment of scrap status in each set member, the increment includes a deterministic generating term and a random fluctuation term following a Gaussian distribution, and add the increment to the state variable of the previous moment to obtain the predicted state set.

[0039] Furthermore, the state estimation module 11 is used to perform the following method: Each predicted state in the predicted state set is mapped to the observation space, and the residual between each state and the real-time state variable is calculated. The Kalman gain matrix is ​​estimated based on the degree of dispersion among the members of the predicted state set. The gain is multiplied by the residual and then superimposed on each predicted member. The variance of all superposition results is clipped to obtain the posterior state set.

[0040] Furthermore, the calculation module 12 is used to perform the following method: From the probability distributions of each state, the waste accumulation state values ​​of each monitoring area are extracted to form a sample set. For each monitoring area, the ratio of the number of waste accumulation state values ​​greater than the preset safety threshold to the total number of samples is counted as the probability of exceeding the limit at the next moment. The probability of exceeding the limit of each monitoring area is normalized, and the decision entropy is calculated using the normalized probability values.

[0041] Furthermore, the control module 13 is used to perform the following methods: Based on the current stamping cycle and waste status observation data, at least two candidate removal actions are generated. Each action includes a set of parameters from the air knife pulse cycle, opening duration, or air pressure amplitude. Using the probability distribution of each state as the initial distribution, for each candidate removal action, the corresponding control parameters are substituted into the state space model to simulate and predict a specified number of stamping cycles in advance, obtaining the predicted waste status distribution of each monitoring area at the prediction time domain endpoint. The future decision entropy corresponding to each candidate removal action is calculated from the predicted waste status distribution. The future decision entropy of each candidate removal action is compared, and the action with the smallest entropy value is selected as the optimal removal action.

[0042] Furthermore, the control module 13 is used to perform the following methods: The activity level of the waste pile is determined based on the waste status observation data; the activity level of the waste pile and the current stamping cycle are combined to select at least two candidate removal actions with the highest matching degree from the removal action database.

[0043] Furthermore, the control module 13 is used to perform the following methods: When the decision entropy is greater than the preset entropy threshold, and the future decision entropy corresponding to each predicted candidate clearing action is greater than the preset entropy threshold, an early warning signal of an unrecoverable accumulation trend is output.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent waste removal and control of stamping dies, characterized in that, The method includes: Real-time data collection of waste status observation in each monitoring area within the stamping die; modeling the waste accumulation status in each monitoring area as a real-time state variable; using ensemble Kalman filtering to dynamically estimate the probability distribution of the real-time state variable to obtain the probability distribution of each state. Calculate the probability that waste accumulation in each monitoring area will exceed the safety threshold in the next moment based on the probability distribution of each state, and calculate the decision entropy based on all monitoring areas; When the decision entropy is greater than the preset entropy threshold, at least two candidate removal actions that are in competition with each other are generated. Using the probability distribution of each state, the evolution of the waste accumulation state in the future prediction time domain after each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and the control command is output to drive the execution mechanism to execute.

2. The intelligent waste removal control method for stamping dies as described in claim 1, characterized in that, An ensemble Kalman filter is used to dynamically estimate the probability distribution of the real-time state variables, resulting in the following probability distributions for each state: Initialize N set members, each member representing a set of possible values ​​for state variables, and assign an initial covariance matrix; At a fixed time in each stamping cycle, the state variables of each set member are updated over time using a model of stamping frequency and scrap generation rate to obtain a predicted state set. The real-time state variables are matched with the predicted state set, the observation residual of each set member is calculated, and the predicted state is corrected according to the Kalman gain to obtain the posterior state set. The mean and covariance of the state variables of each monitoring area are statistically analyzed from the posterior state set to obtain the probability distribution of each state.

3. The intelligent waste removal control method for stamping dies as described in claim 2, characterized in that, Initialize N collection members, including: The initial scrap state of each monitoring area when the stamping die is unloaded is obtained, the initial variance is set for each state variable, and a diagonal covariance matrix is ​​constructed. Centered on the initial mean, deterministic sampling is performed on N set members according to a multidimensional normal distribution, and the N sets of state variable values ​​obtained by sampling are taken as the N set members.

4. The intelligent waste removal control method for stamping dies as described in claim 2, characterized in that, At a fixed time point in each stamping cycle, the state variables of each set member are updated over time using a model based on stamping frequency and scrap generation rate, resulting in a predicted state set, including: For each set member, read the real-time stamping frequency of the current stamping press and the scrap status of a single feeding; Based on the location of each monitoring area, the proportion coefficient of waste falling into each monitoring area is determined by using historical material falling data; Based on the stamping cycle step size, the increment of the scrap state in each set member is calculated in a discretized form. The increment includes a deterministic generating term and a random fluctuation term that follows a Gaussian distribution. The increment is superimposed on the state variable of the previous time step to obtain the predicted state set.

5. The intelligent waste removal control method for stamping dies as described in claim 2, characterized in that, The real-time state variables are matched with the predicted state set, the observation residual of each set member is calculated, and the predicted state is corrected according to the Kalman gain to obtain the posterior state set, including: Map each predicted state in the predicted state set to the observation space, and calculate the residual between each state and the real-time state variable. The Kalman gain matrix is ​​estimated based on the degree of dispersion among the members of the predicted state set. The gain is multiplied by the residual and then superimposed on each predicted member. The variance of all superposition results is then clipped to obtain the posterior state set.

6. The intelligent waste removal control method for stamping dies as described in claim 1, characterized in that, Based on the probability distribution of each state, the probability that waste accumulation in each monitoring area will exceed the safety threshold in the next moment is calculated, and the decision entropy is calculated based on all monitoring areas, including: From the probability distribution of each state, the waste accumulation state values ​​of each monitoring area are extracted to form a sample set. For each monitoring area, the ratio of the number of waste accumulation state values ​​greater than the preset safety threshold to the total number of samples is used as the probability of exceeding the limit at the next moment. The probability of exceeding the limit in each monitoring area is normalized, and the decision entropy is calculated using the normalized probability values.

7. The intelligent waste removal control method for stamping dies as described in claim 1, characterized in that, When the decision entropy is greater than a preset entropy threshold, at least two competing candidate removal actions are generated. Using the probability distributions of each state, the evolution of the waste accumulation state in the future prediction time domain after executing each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, including: Based on the current stamping cycle and waste status observation data, at least two candidate cleaning actions are generated. Each action includes a set of parameters from the air knife pulse cycle, opening duration, or air pressure amplitude. Using the probability distributions of each state as the initial distribution, for each candidate clearing action, the corresponding control parameters are substituted into the state space model, and a specified number of stamping cycles are simulated and predicted forward to obtain the predicted waste state distribution of each monitoring area at the predicted time domain endpoint. The future decision entropy corresponding to each candidate removal action is calculated from the predicted waste state distribution; Compare the future decision entropy of each candidate clearing action, and select the action with the smallest entropy value as the optimal clearing action.

8. The intelligent waste removal control method for stamping dies as described in claim 7, characterized in that, Based on the current stamping cycle time and scrap condition observation data, at least two candidate removal actions are generated, including: The activity level of the waste pile is determined based on the waste status observation data. Based on the activity of the waste pile and the current stamping cycle, at least two candidate removal actions with the highest matching degree are selected from the removal action database.

9. The intelligent waste removal control method for stamping dies as described in claim 7, characterized in that, When the decision entropy is greater than the preset entropy threshold, and the future decision entropy corresponding to each predicted candidate clearing action is greater than the preset entropy threshold, an early warning signal of an unrecoverable accumulation trend is output.

10. A smart waste removal control system for stamping dies, characterized in that, The system is used to implement the intelligent waste removal control method for stamping dies according to any one of claims 1-9, the system comprising: State estimation module: Real-time acquisition of waste state observation data in each monitoring area within the stamping die, modeling the waste accumulation state in each monitoring area as a real-time state variable, and using ensemble Kalman filtering to dynamically estimate the probability distribution of the real-time state variable to obtain the probability distribution of each state; Calculation module: Calculates the probability that waste accumulation in each monitoring area will exceed the safety threshold in the next moment based on the probability distribution of each state, and calculates the decision entropy based on all monitoring areas; Control module: When the decision entropy is greater than the preset entropy threshold, at least two candidate removal actions that are in competition with each other are generated. Using the probability distribution of each state, the evolution of the waste accumulation state in the future prediction time domain after each candidate removal action is simulated. The optimal removal action is determined by predicting the future decision entropy corresponding to each candidate removal action, and the control command is output to drive the execution mechanism to execute.

Images