An AI-based digital factory intelligent control method

Abstract

The application discloses an AI-based digital factory intelligent control method, which comprises the following steps: collecting data, extracting the features of the data, calculating a health index according to the data features, and giving a warning for the equipment; collecting the running data of the equipment in a healthy state, establishing a baseline based on the running data, comparing the real-time collected characteristic values with the baseline, identifying a state change critical point, tracking a deterioration track, identifying a degradation state according to the state change critical point, calculating a reversibility probability, obtaining a reversible interval and an irreversible point of equipment degradation based on the reversibility probability, implementing a corresponding intervention strategy according to the reversibility, predicting the time of failure occurrence according to the features of the data, calculating the best intervention time, and matching the means according to the means matching rule base according to the failure features; the equipment health state critical point is accurately identified, the failure occurrence time is predicted in advance, the reversibility of the equipment degradation state is reasonably judged, and the intervention strategy is targeted and economic and effective.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of artificial intelligence control, and in particular to an AI-based digital factory intelligent control method. BACKGROUND

[0002] In modern industrial production, digital factories have become an important means to improve production efficiency, ensure product quality and reduce operating costs. With the deepening of the concept of intelligent manufacturing and Industry 4.0, factory equipment is gradually developing towards high automation and intelligence. However, with the increase in equipment complexity and the diversification of operating environments, equipment failure and performance degradation problems are increasingly prominent, posing many challenges to production.

[0003] The traditional equipment maintenance mode mainly relies on periodic maintenance and after-maintenance. Although periodic maintenance can prevent equipment failure to some extent, it often has problems of over-maintenance or insufficient maintenance, resulting in resource waste or frequent failures. After-maintenance is to repair the equipment after failure, which not only causes production interruption, but also may cause chain failure, resulting in greater economic loss. Therefore, the traditional maintenance mode cannot meet the needs of modern digital factories for high reliability and high stability of equipment. SUMMARY

[0004] The application provides an AI-based digital factory intelligent control method, which accurately identifies the critical point of equipment health state, predicts the failure occurrence time in advance, reasonably judges the reversibility of equipment degradation state, and formulates targeted and economically effective intervention strategies.

[0005] The application provides an AI-based digital factory intelligent control method, which comprises:

[0006] S101, collecting data, extracting features of the data, calculating a health index according to the data features, and warning the equipment based on the health index, wherein the data includes operating data of the equipment and basic data of the equipment;

[0007] S102, collecting operating data of the equipment in the health state, establishing a baseline based on the operating data, comparing the real-time collected feature values with the baseline, identifying the state change critical point, and tracking the deterioration trajectory according to the state change critical point;

[0008] S103, identifying the degradation state according to the state change critical point, calculating the reversibility probability according to the degradation state of the equipment, obtaining the reversible interval and the irreversible point of the equipment degradation based on the reversibility probability, and implementing the corresponding intervention strategy according to the reversibility;

[0009] S104, predicting the time of failure occurrence according to the characteristics of the data, calculating the optimal intervention time based on the predicted time of failure occurrence, and matching the means according to the means matching rule base according to the characteristics of the failure.

[0010] Preferably, the calculation formula of the health index is: , wherein, is the health index, is the energy spectrum entropy, is the current imbalance degree, is the voiceprint abnormal score, is the pass rate slope, is the weight of the energy spectrum entropy, which is obtained according to the degree of mechanical wear and the vibration signal, is the weight of the current imbalance degree, which is obtained according to the electrical fault, is the weight of the voiceprint abnormality, which is obtained according to the voiceprint feature, is the weight of the pass rate slope, which is obtained according to the degree of deterioration of production quality, .

[0011] Preferably, the baseline is a range interval that is adjusted in real time according to the equipment operating state, environmental changes, and equipment aging factors, and the data characteristics include the energy spectrum entropy, the current imbalance degree, the voiceprint abnormal score, and the pass rate slope. The baseline range is the average value of the data characteristics in a period of time plus or minus twice the standard deviation.

[0012] Preferably, the state change critical point includes a first level critical point and a second level critical point. The first level critical point is a point where the health index rises and continues to rise for a period of time, but does not reach a preset risk threshold. The second level critical point is a point where the health index rises and exceeds the preset risk threshold.

[0013] Preferably, the formula for calculating the reversibility probability is: , wherein, represents the reversibility probability, n is the number of reversible evidence, N is the total number of evidence, reversibility is determined according to the calculated reversibility probability, a probability threshold range is set, the threshold range includes a highest threshold and a lowest threshold, when the reversibility probability is greater than the highest threshold, the degradation of the equipment is the highest reversibility, when the reversibility probability is less than or equal to the highest threshold and greater than the lowest threshold, the degradation of the equipment is critical reversibility, and when the reversibility probability is less than or equal to the lowest threshold, the degradation of the equipment is irreversible.

[0014] Preferably,

[0015] S201, obtaining the process relationship of the equipment according to the collected basic data of the equipment, calculating the dependence strength, and constructing the dependence relationship of the production chain based on the dependence strength and the equipment information;

[0016] S202, collect historical failure data, build a propagation model according to the historical failure data, monitor the risk source node in real time, and generate a blocking strategy based on the risk source node.

[0017] Preferably, the risk source node is divided into three risk levels of single-point high-risk, multi-node medium-risk and full-chain high-risk, the single-point high-risk is only one risk source node, the multi-node medium-risk is multiple risk source nodes, and the full-chain high-risk is multiple risk source nodes, and the multiple risk source nodes are related and interact with each other, resulting in a situation of full-line shutdown.

[0018] Preferably,

[0019] S301, obtaining equipment degradation entropy value data according to operation data and failure frequency of the equipment, calculating risk pressure value according to the equipment degradation entropy value data, calculating risk conduction intensity based on the risk pressure value, identifying risk flow direction, and generating a risk flow line graph based on the risk flow direction;

[0020] S302, monitoring the risk flow intensity in real time, generating a digital blocking barrier based on the risk flow intensity, identifying a risk flow loop according to an algorithm, and adjusting the strategy according to the risk flow loop level.

[0021] Preferably, the formula for calculating the risk conduction intensity is: wherein, is the risk conduction intensity, is the risk pressure value of the equipment i, is the risk conduction characteristic data between the equipment i and the equipment j, is the weight of the risk pressure value, is the weight of the risk conduction characteristic data, + =1.

[0022] Preferably, the risk flow direction refers to the conduction from a high-pressure equipment to a low-pressure equipment.

[0023] One or more technical solutions provided in the present application have at least the following technical effects or advantages: by accurately identifying the critical point of the equipment health state, predicting the failure occurrence time in advance, reasonably judging the reversibility of the equipment degradation state, formulating an intervention strategy with strong pertinence and economic effectiveness, realizing the precise monitoring of the equipment health state, the critical point identification, the degradation state grading judgment, the reversibility determination, and formulating a scientific and reasonable intervention strategy, through effect verification and continuous optimization, ensuring the stable operation of the equipment, improving the production quality and efficiency, and reducing the maintenance cost.

[0024] By means of the device dependency graph and the graph neural network, the dependency relationship between devices and the risk propagation path can be accurately identified, the problem of associated device chain failure caused by irreversible degradation of a single device in the production line can be inhibited, dynamic risk propagation can be blocked, the range of production stoppage and loss can be minimized, and the stability and economy of the production line operation can be improved.

[0025] By combining the digital blocking barrier and the intelligent circuit breaker, the risk conduction path can be more accurately blocked, risk rebound can be prevented, healthy devices can be effectively protected, the reliability of the overall operation of the devices can be improved, dynamic topology analysis of device risks can be realized, risk conduction can be accurately predicted and effectively blocked through risk flow simulation and digital twin blocking technology, risk rebound can be prevented, the stability and reliability of device operation can be improved, and the smooth progress of production can be ensured. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of an AI-based digital factory intelligent control method according to an embodiment of the present application;

[0027] Figure 2 is a flowchart of constructing a production chain dependency relationship according to an embodiment of the present application;

[0028] Figure 3 is a flowchart of generating a risk flow line graph according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings; the preferred embodiments of the present application are shown in the drawings, but the present application can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0030] It should be noted that the terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are only for the purpose of illustration and do not represent the only embodiment.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0032] Example one: Figure 1 is a flowchart of an AI-based digital factory intelligent control method according to an embodiment of the present application, comprising:

[0033] S101, collect data, extract the characteristics of the data, calculate the health index according to the data characteristics, and give an early warning to the equipment based on the health index;

[0034] Further, a temperature sensor, a voltage sensor, a vibration sensor, an acoustic sensor and a current clamp are installed on the equipment to monitor the temperature, voltage, vibration signal, mechanical noise and three-phase current of the equipment. The vibration sensor is installed on the main shaft of the equipment to monitor the vibration of the main shaft of the equipment in real time, including amplitude, frequency and phase. The acoustic sensor is installed on the equipment shell to capture abnormal mechanical noise during equipment operation in real time. The current clamp is installed at the input end of the motor or the output end of the power distribution cabinet to monitor the fluctuation of the three-phase current of the motor. The same kind of parts are evenly distributed to multiple equipment for production, and the number of qualified parts and the number of unqualified parts of each equipment during the production cycle are collected in real time. The qualified rate of parts is calculated according to the collected number of qualified parts and the number of unqualified parts.

[0035] The collected vibration signal is subjected to fast Fourier transform (FFT) to obtain the energy distribution of 0-5kHz frequency domain. The 0-5kHz frequency domain is divided into multiple equal or non-equal width sub-frequency bands to obtain the energy of each frequency band. The formula for calculating the energy spectrum entropy is: wherein, is the energy spectrum entropy, which quantifies the disorder degree of the frequency energy distribution of the vibration signal, N is the total number of frequency band division (0-5kHz frequency domain is divided into multiple sub-frequency bands), i is the index of the frequency band, indicating the serial number of the current calculated frequency band, is the energy proportion of the i-th frequency band, reflecting the distribution concentration degree of energy in the frequency domain. Low energy spectrum entropy indicates that energy is concentrated in a few frequency bands (such as normal equipment vibration energy mainly distributed in low frequency band), indicating that the energy distribution is orderly. High energy spectrum entropy indicates that energy is dispersed in multiple frequency bands (such as fault leading to increase of high frequency component), indicating that the energy distribution is disorderly, and the risk of mechanical wear or failure is increased. The formula for calculating the current unbalance degree according to the monitored three-phase current is: wherein, is the current unbalance degree, ranging from 0 to 1, is the maximum value in the three-phase current, is the minimum value in the three-phase current, is the average value of the three-phase current, when =0, it indicates that the three-phase current is completely balanced, which is the ideal state, the larger the value is, the more serious the current unbalance is, and the higher the risk of electrical failure is. The current unbalance degree value at the current time is obtained, and the time stamp is recorded. The frequency spectrum resolution is calculated according to the signal length and sampling rate of the collected mechanical signal, and the single-side amplitude spectrum, i.e. the frequency domain spectrum, is generated according to the frequency spectrum resolution. The measured spectrum is compared with the theoretical fault frequency library, and the formula for calculating the voiceprint anomaly score is: wherein, is the voiceprint abnormality score, A is the fault frequency amplitude, B is the noise base, the background noise level of the spectrum under normal working conditions, T is the significance threshold, is the upper limit of the dynamic range; the pass rate slope is calculated according to the obtained pass rate of the parts, and the calculation formula of the pass rate slope is: wherein, is the pass rate slope, is the current pass rate, is the pass rate of the previous moment, is the time interval, the positive value of the pass rate slope indicates that the pass rate is rising, the negative value of the pass rate slope indicates that the pass rate is falling, and the greater the absolute value of the pass rate slope, the more obvious the quality deterioration trend, the health index is calculated according to the calculated energy spectrum entropy, current unbalance degree, voiceprint abnormality score and pass rate slope, and the calculation formula of the health index is: wherein, is the health index, is the energy spectrum entropy, is the current unbalance degree, is the voiceprint abnormality score, is the pass rate slope, is the weight of the energy spectrum entropy, which is obtained according to the degree of mechanical wear and tear and the vibration signal, is the weight of the current unbalance degree, which is obtained according to the electrical fault, is the weight of the voiceprint abnormality, which is obtained according to the voiceprint feature, is the weight of the pass rate slope, which is obtained according to the deterioration degree of production quality, ;

[0036] The health index calculated is set to a first-level warning threshold, when the health index is greater than the preset first-level warning threshold, the system automatically triggers a first-level warning, indicating that the equipment has a high probability of failure risk, and the warning information is pushed through short message, email or industrial APP, and the equipment is stopped for maintenance.

[0037] S102, collecting running data of the equipment in a healthy state, establishing a dynamic baseline based on the running data, comparing the real-time collected characteristic value with the dynamic baseline, identifying a state change critical point, and tracking a deterioration trajectory according to the critical point;

[0038] Specifically, the dynamic baseline is a range interval that is adjusted in real time according to factors such as the running state of the equipment, changes in the environment, aging of the equipment, etc. The running data of the equipment in a healthy state is collected. The healthy state of the equipment is the state when the new equipment installation and debugging are completed, after major repair, or when the production enters the stable period. The data collection is started in this state. According to the collected data, the features such as temperature, voltage, vibration signal, mechanical noise, and three-phase current are extracted. The characteristic values such as energy spectrum entropy, current imbalance degree, voiceprint abnormality score, and qualified rate slope are calculated. The average value and standard deviation of the energy spectrum entropy, current imbalance degree, voiceprint abnormality score, and qualified rate slope in a period of time are calculated. According to the statistical process control (SPC) principle, the initial normal range of each characteristic value is set as the average value plus or minus twice the standard deviation. For example: the vibration energy spectrum entropy baseline range is 0.1-0.3. At the same time, the rolling window average method is used to calculate the statistics to avoid the influence of extreme values on the threshold value, so that the initial baseline range can accurately reflect the characteristics of the normal operation of the equipment. The threshold value is automatically recalculated every quarter, and the newly collected healthy data is included in the calculation range. When the equipment is maintained or the key components are replaced, the running characteristics of the equipment may change. At this time, the baseline needs to be reset immediately. The baseline is dynamically adjusted according to the qualified rate. If the sensor features are within the baseline range, but the qualified rate abnormally decreases, the decision tree model is used in combination with the qualified rate signal to sort the feature importance, recalibrate the feature weight, and preferentially retain the features that are highly correlated with the qualified rate. The threshold range is adjusted to make the baseline more accurately judge the health status of the equipment.

[0039] The real-time collected characteristic values are compared with the established dynamic baseline, for each characteristic value, it is identified whether it exceeds the baseline range, the number of times and the duration of each characteristic exceeding the limit are recorded, the characteristic with more than three times of exceeding the limit is marked as an abnormal event, when at least two characteristics exceed the limit for three times in succession and the qualified rate slope presents a negative value, this situation is marked as potential degradation, for the characteristic exceeding the limit, the deterioration speed is quantified using the simple moving average method, and the health index is calculated, the normal range and risk threshold of the health index are set, when the health index starts to rise from the normal range and continuously rises in a period of time but does not reach the risk threshold, and the qualified rate slope presents a slight negative value, it is determined that the equipment is a first-grade critical point, when the health index continuously rises and exceeds the set risk threshold, or the qualified rate slope sharply decreases, it is determined that the equipment is a second-grade critical point, from single-characteristic abnormality to multiple-characteristic continuous exceeding, the rise of the health index and the sharp decrease of the qualified rate slope, the complete deterioration trajectory of the initial abnormality, progressive degradation and critical failure of the equipment is tracked, when the first-grade critical point is identified, a first-level early warning is triggered, at this time, the load of the equipment is automatically reduced by 20% to reduce the operating pressure of the equipment and delay the deterioration speed of the equipment, and the maintenance personnel is notified to check the equipment; when the second-grade critical point is identified, a critical early warning is triggered immediately, at this time, the standby equipment is immediately switched to ensure the continuity of production, and a detailed diagnostic report is generated, which includes the abnormal characteristic combination, the degradation trajectory graph and the possible cause analysis and the like.

[0040] S103, identifying the degradation state according to the state change critical point, obtaining the reversible interval and irreversible point of the equipment degradation according to the degradation of the equipment, and implementing the corresponding intervention strategy according to the degradation level and the reversibility;

[0041] Further, other data in the operation of the equipment are collected, the other data including mechanical response characteristics (such as vibration entropy return speed after lubrication), material deformation characteristics (metal fatigue degree obtained by current waveform analysis) and recovery potential characteristics (success rate of historical maintenance of the same type), when the health index continuously rises and abnormal conditions occur, or the qualified rate slope presents a significant negative change, it is determined that the equipment may enter a degradation state, and the degradation state is determined in combination with the mechanical response characteristics, the material deformation characteristics and the recovery potential characteristics, the vibration entropy return speed after lubrication lower than the normal level indicates that the mechanical response of the equipment is abnormal; if the metal fatigue cumulative index obtained by the current waveform analysis exceeds the low risk threshold, it indicates that the deformation of the material has affected the normal operation of the equipment; if the success rate of historical maintenance of the same type is lower than a certain level, it means that the equipment may be difficult to recover under the current state through conventional maintenance; according to the above conditions, when multiple characteristics simultaneously show abnormalities, it is determined that the equipment is in a degradation state.

[0042] The reversibility is determined according to the device characteristics, the reversible threshold and the irreversible threshold are set, the real-time monitored vibration entropy falling speed is compared with the reversible and irreversible thresholds, if the falling speed is greater than the reversible threshold, it is reversible, if the falling speed is less than the irreversible threshold, it is irreversible; the historical maintenance records of the same type of device are inquired, the maintenance history success rate is calculated, if the success rate is greater than 80%, it is reversible, if the success rate is less than 30%, it is irreversible; the metal fatigue accumulation index obtained by analyzing the current waveform, if the index is less than 0.5, it is reversible tendency, if the index is greater than 0.8, it is irreversible tendency; the vibration entropy falling speed, the maintenance history success rate and the fatigue accumulation index are all evidence items for determining whether it is reversible, the formula for calculating the reversibility probability according to the evidence items is: wherein, represents the reversibility probability, n is the number of reversible evidence, N is the total number of evidence, the reversibility is determined according to the reversibility probability calculated, if >0.7, it is determined that the degradation of the device is high reversibility (corresponding to the degradation level of 0-1 level); if 0.4 ≤0.7, it is determined that the degradation of the device is critical reversibility (corresponding to the degradation level of 1 level); if ≤0.4, it is determined that the degradation of the device is irreversible (corresponding to the degradation level of 2-3 level), the degradation level 0 and 1 belong to the reversible interval, the degradation level 0 is that the device is in the elastic degradation state, such as short-term temperature overrun, at this time the device has the ability of automatic recovery, which belongs to the reversible interval, the degradation level 1 is that the device appears plastic degradation, such as insufficient lubrication, through external intervention, such as automatic lubrication or load adjustment, the device can recover the baseline performance within 72 hours, which belongs to the reversible interval; the degradation levels 2 and 3 belong to the irreversible point, the degradation level 2 is that the device appears local failure, such as bearing crack, at this time the device has been seriously damaged, and needs to be replaced to recover normal operation, if not handled in time, the device state will further deteriorate, which belongs to the state close to irreversible, the degradation level 3 is that the device is completely failed, such as main shaft fracture, the device has already been unable to continue to run, and can only be retired, which is an explicit irreversible point.

[0043] According to the intervention strategy corresponding to the identified degradation level, when the degradation level is 0-1 and is highly reversible, for the degradation caused by insufficient lubrication, the system automatically triggers the lubrication device, adds appropriate amount of lubricating oil to the equipment, improves the lubrication condition of the equipment, automatically adjusts the load of the equipment according to the performance and running state of the equipment, avoids overloading operation of the equipment, makes the equipment work within a reasonable load range, and promotes the recovery of the performance of the equipment; for the degradation level of 1 and critical reversibility, the easily damaged parts prone to failure in the equipment are replaced, such as replacing the belts and filter elements that will reach the service life, avoiding the damage of these easily damaged parts to cause the further deterioration of the equipment state, adjusting the production process parameters to compensate for the operation of the equipment, reducing the performance fluctuation of the equipment caused by degradation, for example, adjusting the machining precision, speed and other parameters to ensure the quality of the product; for the degradation level of 2-3 and irreversible, an equipment retirement warning signal is sent to the plant manager, reminding to arrange the retirement plan of the equipment in time, automatically calling the intelligent distribution module to transfer the production task of the irreversible degradation equipment to the first-level performance equipment to ensure the continuity of production. At the same time, a device retirement schedule is generated, the retirement and updating work of the equipment is reasonably arranged, a spare part procurement application (through the MES system) is triggered, and the spare parts required for equipment updating are purchased in time; at the same time, a device updating budget application is submitted to the ERP system to provide financial support for the updating of the equipment.

[0044] In S104, the time of the fault occurrence is predicted according to the characteristics of the data, and the optimal intervention time is calculated based on the predicted time of the fault occurrence, and the fault characteristics are matched with the means according to the means matching rule base;

[0045] Specifically, according to the calculated data characteristics, when the energy spectrum entropy daily increment is greater than 0.1 and the qualified rate slope decreases by less than -0.15 / min, it indicates that the equipment fault diffusion speed is relatively fast, and it is predicted that the fault will occur within 12 hours, and when the current unbalance degree is greater than 15% and the qualified rate breaks through the preset minimum qualified rate threshold, it means that the electrical system of the equipment has serious problems, and it is predicted that the machine will stop within 2 hours, and the calculation formula of the optimal intervention time is: wherein, is the optimal intervention time, is the predicted time of the fault, which is obtained according to the data characteristics, is the production gap cutoff time, which is determined according to the current production state of the equipment, The safety margin is used to ensure that the intervention operation has enough time to prepare and execute, and to avoid equipment failure caused by untimely intervention. The default setting is 1 hour. According to the data characteristics, the fault feature combination is obtained, which includes vibration dominance, voltage fluctuation and sudden drop of qualified rate, and multi-modal abnormality and health index exceeding risk threshold. The means matching is performed from the means matching rule library. The steps are as follows: for the fault feature of vibration dominance, the load is reduced and automatic lubrication is performed. Reducing the load can reduce the mechanical stress of the equipment and inhibit the further development of mechanical wear. Automatic lubrication can improve the lubrication condition of the equipment and reduce friction. For the fault feature of voltage fluctuation and sudden drop of qualified rate, the stable voltage source is switched and the defective products are reprocessed. Switching the stable voltage source can ensure the electrical stability of the equipment and avoid damage to the equipment caused by voltage fluctuation. Re-processing defective products can improve the qualified rate of products and reduce losses. For the fault feature of multi-modal abnormality and health index exceeding risk threshold, the standby equipment is switched and the maintenance work order is triggered. When the equipment has multi-modal abnormality and the health index is high, it means that the equipment may have serious failure. Switching the standby equipment can avoid the interruption of the production line and ensure the continuity of production. Triggering the maintenance work order can arrange professional personnel to repair the faulty equipment in time.

[0046] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: by accurately identifying the critical point of the equipment health state, predicting the failure time in advance, reasonably judging the reversibility of the equipment degradation state, formulating targeted and economically effective intervention strategies, realizing accurate monitoring of the equipment health state, critical point identification, degradation state classification judgment, reversibility determination, and formulating scientific and reasonable intervention strategies, through effect verification and continuous optimization, ensuring stable operation of the equipment, improving production quality and efficiency, and reducing maintenance cost.

[0047] Embodiment two: based on the irreversible equipment failure in the above embodiment one, for the problem that the single equipment irreversible degradation in the production line causes the associated equipment to fail in series (such as the main shaft fracture causing the gear box overload), this embodiment solves the limitations of traditional single-point maintenance by constructing a networked cascading risk control system, such as Figure 2

[0048] S201, obtaining the process relationship of the equipment according to the collected basic data of the equipment, calculating the dependence strength, and constructing the dependence relationship of the production chain based on the dependence strength and the equipment information;

[0049] ​Further, the collected equipment includes equipment model, number and function, through the production site, understand the process flow and connection relationship between each equipment in the production chain, for example, in the automobile manufacturing production chain, clear the sequence and cooperation mode between stamping equipment, welding equipment, coating equipment and assembly equipment, record the input and output information of each equipment in the production process, form a complete equipment process relationship list, through theoretical analysis, experimental test or actual production data monitoring, obtain the load transfer coefficient between the equipment in the production chain, collect historical fault data, including fault occurrence time, equipment name, fault type and fault conduction information, according to the historical fault data, count the historical fault conduction times between each pair of equipment, the formula for calculating the dependence intensity through the historical fault conduction times is: Wherein, is the dependence intensity, used to measure the degree of influence of one equipment failure on another equipment, is the historical fault conduction times, refers to the number of times that one equipment failure leads to another equipment failure or abnormality in a certain period of time, is the current load ratio, refers to the relative proportion relationship between the loads of two equipment under the current production state, according to the dependence intensity, set the judgment threshold, when the calculated dependence intensity is greater than the preset judgment threshold, mark the equipment connection as a red warning edge, use the visualization tool Gephi to display the equipment in the production chain in the form of nodes, the connection between the equipment is represented by edges, according to the calculated dependence intensity, set different colors and styles for the edges, such as using bright red lines to represent high dependence relationship (red warning edge), and using lines of different thickness or color to display other dependence relationships, at the same time, mark the equipment name on the node, and display the dependence intensity and other key information on the edge, so that the dependence relationship between the equipment is clear and visible, when the irreversible degradation equipment information is obtained, mark the risk source node in the equipment dependence map; when the reversible degradation equipment information is obtained, mark it as a potential conduction node.

[0050] S202, collect historical fault data, build a propagation model according to the historical fault data, real-time monitor the risk source node, and generate a blocking strategy based on the risk source node;

[0051] Specifically, from the fault records, according to the prominent features of cascading failures, such as the chain reaction of failures, the large-scale shutdown of production lines, etc., representative cases are screened out. According to the screened cases, basic information such as the time, location, equipment model, and failure phenomenon of the fault occurrence is recorded. According to the basic information, the equipment degradation state is identified, the dependency strength is calculated, and the associated equipment degradation probability is identified. For the associated equipment degradation probability, the possibility of the associated equipment appearing degradation after the failure of the risk source equipment is identified. The probability value is determined by combining expert evaluation and data analysis. The well-organized and labeled case information is stored in a unified format to construct a structured training dataset. Using the graph neural network framework PyTorch Geometric, the equipment degradation state and dependency strength data are input into the input layer. The equipment degradation state and dependency strength data are organized into a graph structure, where the equipment is the node of the graph, the dependency strength is the weight of the edge between the nodes, and the equipment degradation state is the attribute of the node. The equipment degradation state and dependency strength data are filled into the PyTorch Geometric data structure as the input of the input layer. The output layer outputs the associated equipment degradation probability and the prediction result of the paralysis time. Two independent output branches are used. One branch outputs the degradation probability (the value range is between 0 and 1), and the other branch outputs the paralysis time. For the degradation probability output branch, the Sigmoid activation function is used to limit the output value to between 0 and 1. For the paralysis time output branch, the linear activation function can be used to directly output the predicted time value. The constructed training dataset is divided into training set, validation set, and test set. The training set is used for model training. The Adam optimization algorithm is used to adjust the parameters of the model to minimize the loss function. The number of training rounds and the batch size are set. The parameters of the model are updated iteratively to make the prediction results of the model gradually approach the actual data. The validation set is used to adjust the parameters of the model and evaluate the performance of the model during training. The test set is used to evaluate the generalization ability of the model finally. The risk source node is monitored in real time. The monitoring content includes the operating parameters of the equipment (such as temperature, vibration, current, etc.), fault alarm information, etc.The state data of the equipment is acquired in real time through the sensor and the data acquisition device, a risk early warning threshold is set, when the state data of the risk source node is monitored to exceed the early warning threshold, a risk deduction process is triggered, according to the structure of the equipment dependency graph, a first-order associated equipment directly connected with the risk source node is extracted, an adjacency list traversal algorithm is used to traverse the adjacent nodes of the risk source node, and the adjacent nodes are extracted as the first-order associated equipment, the recession entropy value of the risk source node is acquired, the recession entropy value can be calculated according to the operation parameters and the historical data of the equipment, and is used to measure the recession degree and the fault risk of the equipment, the dependency strength of the risk source node and the first-order associated equipment is acquired from the equipment dependency graph, the conduction recession probability = source node recession entropy value * dependency strength, according to actual production requirements and risk bearing capacity, a conduction recession probability threshold is set, when the calculated conduction recession probability is greater than the threshold, a second-order associated equipment directly connected with the first-order associated equipment is scanned, the adjacent nodes of the first-order associated equipment are traversed, and the information of the second-order associated equipment is acquired.

[0052] According to the number, influence range and severity of the risk source nodes, the risk level is divided, if there is only one risk source node, and the influence range of the node is limited to only a few closely related devices, and the disturbance to the overall production process is small, it is defined as a single-point high-risk; if there are multiple risk source nodes, but these nodes have not formed a chain reaction to cause the whole production line to be paralyzed, it is a multi-node medium-risk; if there are multiple risk source nodes, and these nodes are interrelated and interacted, causing the whole line to be shut down, it is defined as a whole-chain high-risk, for the single-point high-risk, the risk source equipment is closed and the redundant equipment is enabled, the emergency stop button is pressed and the power is turned off, for the multi-node medium-risk, the speed is reduced and the process compensation strategy is implemented, by adjusting the control parameters of the production equipment (such as motor speed, conveyor belt speed, etc.), the production speed is reduced to 50% of the original, according to the production situation after the speed is reduced, the production process is adjusted accordingly, such as increasing the processing time, adjusting the processing parameters, etc., to ensure that the product quality is not affected, when the whole-chain high-risk occurs, the strategy of isolating the fault section and dynamically reorganizing the production flow is executed, the physical isolation device (such as isolation fence, valve, etc.) or the logical isolation measure (such as the locking function in the software control system) is set to isolate the fault section with multiple risk source nodes from other normal production sections, to prevent the risk from further spreading.

[0053] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: through the equipment dependency graph and the graph neural network, the dependency relationship between the equipment and the risk propagation path can be accurately identified, the problem of associated equipment chain failure caused by irreversible recession of a single equipment in the production line can be inhibited, the risk propagation can be dynamically blocked, the shutdown range and the loss can be minimized, and the stability and economy of the production line operation can be improved.

[0054] Embodiment three: in the above embodiments, the device risk association is mainly based on static topological relationship (such as process connection), this scheme will introduce risk flow, simulate the transmission path of fault energy in the device network, and realize virtual blocking, such as Figure 3 as shown.

[0055] S301, obtaining device degradation entropy value data according to the operation data and failure frequency of the device, calculating risk pressure value according to the device degradation entropy value data, identifying risk flow direction, and generating risk flow line graph based on the risk flow direction;

[0056] Further, all devices on the production line are converted into energy nodes, according to the collected historical operation data and device failure records, the historical operation data records the operation state and performance of the device in different time periods, understands the running condition and performance change trend of the device in the past, and the device failure record is the problem once appeared by the device and the frequency and severity of the fault, according to the historical operation data, failure record and expert evaluation result, the result obtained according to the historical operation data, failure record and expert evaluation result is weighted and averaged, finally the degradation entropy value corresponding to each device is obtained, the risk pressure value is calculated according to the device degradation entropy value, the risk pressure value = device degradation entropy value × process dependence coefficient, wherein the device degradation entropy value reflects the degradation degree of the device itself, the higher the degradation entropy value, the closer the device is to the fault state, and the greater the risk of its own failure, the process dependence coefficient embodies the mutual correlation and influence degree of the devices in the process flow of the production line, in the production line, the devices are not running in isolation, the running state of a device will have a direct or indirect influence on other devices, the larger the process dependence coefficient, the closer the process correlation between the devices, and the more significant the influence of the failure or performance degradation of a device on other devices, for example: through in-depth analysis of the historical operation data of the main shaft, detailed review of the failure record and comprehensive consideration of the expert evaluation, it is determined that the degradation entropy value of the main shaft is 0.9, which indicates that the main shaft has shown obvious signs of degradation, and the possibility of failure is relatively high. At the same time, in the analysis of the process relationship between devices, it is found that there is a close process relationship between the main shaft and the gear box, the main shaft as a key component of power transmission, its running state has an important influence on the normal operation of the gear box, after evaluation, it is determined that the process dependence coefficient of the gear box to the main shaft is 0.7, which means that the operation of the gear box depends to a large extent on the stable work of the main shaft, any failure or performance fluctuation of the main shaft may have a greater impact on the gear box, according to the risk pressure value calculation formula, the degradation entropy value 0.9 of the main shaft is multiplied by the process dependence coefficient 0.7, that is, 0.9 × 0.7 = 0.63, this calculation result 0.63 is the risk pressure value of the main shaft, the main shaft is facing a certain size of risk pressure in the current production line environment, the larger the risk pressure value, the more serious the risk situation of the main shaft, and the more attention and attention we need to pay to it.

[0057] The passive radio frequency tag is arranged at the equipment gap position. The passive radio frequency tag can sense the weak signal changes generated in the risk conduction process, for example, when equipment A fails, the abnormal vibration or electromagnetic interference generated thereby will be conducted to the surrounding equipment through the equipment gap. The passive radio frequency tag can capture these changes in time. The risk conduction characteristics are collected by the passive radio frequency tag, the risk conduction characteristics include vibration wave frequency shift and electromagnetic interference change, the vibration wave frequency shift reflects the abnormal change of the equipment running state, when the equipment fails or performance declines, the vibration frequency and amplitude will change, thereby causing the vibration wave frequency shift, the electromagnetic interference change is caused by the electrical fault inside the equipment or the interference of the external electromagnetic environment. The collected risk conduction characteristics are transmitted to the data processing center. The risk flow direction is from the high-voltage equipment (i.e. the fault source, the equipment with a higher risk pressure value) to the low-voltage equipment (i.e. the equipment with a lower risk pressure value). A risk flow line diagram is generated according to the risk conduction characteristic data, the risk pressure value and the risk flow direction. Specifically, the connection relationship between the equipment is determined according to the physical layout and the process flow of the production line. For each equipment, the risk conduction intensity of the equipment to the surrounding equipment is calculated according to the risk pressure value and the risk conduction characteristic data of the equipment. The formula is as follows: wherein, is the risk conduction intensity, is the risk pressure value of the equipment i, is the risk conduction characteristic data between the equipment i and the equipment j, is the weight of the risk pressure value, is the weight of the risk conduction characteristic data, + =1, the risk flow direction information (conducted from the high-voltage equipment to the low-voltage equipment) and the connection relationship between the equipment are input into the Dijkstra algorithm. The Dijkstra algorithm outputs the shortest path of the risk propagation. The production line equipment is displayed on the plane in a graphical manner through a two-dimensional graphical interface. Each equipment is represented by a specific graphical symbol. The size and color of the graph are initialized according to the risk pressure value of the equipment. According to the obtained risk propagation path, a line with an arrow is drawn on the risk flow line diagram to represent the risk conduction direction. The thickness and color of the line are set according to the risk conduction intensity, that is, the risk flow line diagram is generated.

[0058] S302, real-time monitoring of risk flow intensity, generating digital blocking barriers based on risk flow intensity, identifying risk flow loops according to the algorithm, adjusting the strategy according to the risk flow loop level;

[0059] Specifically, the calculation of the risk flow intensity is obtained by integrating the real-time collected equipment operation state data of various types of sensors (such as vibration, temperature, pressure sensors, etc.) deployed in advance, after pre-processing such as filtering and amplification by the data acquisition system, inputting into the risk flow intensity calculation model for analysis, the risk flow intensity calculation model establishes a physical relationship network reflecting the risk propagation path according to the analysis of the connection mode (such as mechanical transmission, pipeline transportation or electrical interconnection) between the equipment according to the production line topology structure; secondly, for different connection types, extract key sensor parameters (such as flow rate / pressure difference for pipeline connection, current harmonic / temperature rise for electrical circuit), use multi-source data fusion technology to eliminate measurement noise and standardize processing; then, combined with historical fault data and mechanism analysis, use statistical modeling or machine learning methods (such as Bayesian network, random forest) to quantify the contribution weight of each parameter to risk transmission, and construct parameter coupling equation; finally, through real-time calculation of dynamic correlation indicators (such as energy transmission efficiency, abnormal fluctuation covariance), output the comprehensive intensity value reflecting the risk propagation probability and influence degree between equipment, and continuously iterate the model parameters with sensor data flow, finally output the indicators representing the risk propagation intensity between equipment, set the risk flow intensity threshold according to the historical operation data, when the real-time monitored risk flow intensity is greater than the set threshold, a trigger signal is sent, and a digital blocking barrier is automatically generated according to the trigger signal, the digital twin is a virtual mapping of the production line equipment, which synchronizes the operation state and parameters of the equipment in real time, the generation of the barrier is dynamically constructed in the digital twin environment through software programming and algorithm, and the core is embedded with a phase cancellation array, which is an anti-frequency vibration system composed of multiple independently controllable vibration units, when the barrier is activated, the phase cancellation array analyzes the frequency and phase characteristics of the risk flow in real time, and accurately releases the reverse anti-frequency vibration wave, the anti-frequency vibration wave released in the opposite direction and the vibration wave in the risk flow superimpose each other, and the phase cancellation phenomenon occurs. By accurately controlling the frequency and phase of the anti-frequency vibration wave, the vibration wave energy in the risk flow can be weakened or completely cancelled out, thereby blocking the conduction of the risk flow from the risk source equipment to the healthy equipment.

[0060] The risk conduction path between devices is monitored in real time using a depth-first search (DFS) algorithm, recursively traversing the downstream path from each node and maintaining an access stack to record the current path; when a node is found to have been visited and exists in the current path stack, it is determined that a risk closed loop is formed (such as A→B→C→A), and the total risk intensity of the closed loop path (the cumulative weight of each edge) is calculated; if the algorithm finds that the risk flows through the path to form a closed loop (such as A→B→C→A), the loop is marked as a high-risk loop; when the algorithm detects a risk closed loop (such as D→E→F→D) but the cumulative risk intensity is lower than the preset high-risk threshold, the system marks the loop as a medium-risk loop; for a high-risk loop, first identify the key nodes in the loop through a risk analysis method, the key nodes are devices or connection points that have an important influence on risk conduction in the loop, and an intelligent circuit breaker is implanted at the identified key nodes. The intelligent circuit breaker has the functions of automatic detection and circuit interruption, and can monitor parameters such as current and voltage in the loop in real time. When the risk indicators in the loop are detected to exceed the set value, the intelligent circuit breaker will automatically cut off the circuit and cut off the weakest device connection, thereby breaking the risk loop; for a medium-risk loop, the key nodes in the loop are also determined, and a phase offset module is inserted at the key nodes. The phase offset module can change the phase of risk conduction, and by adjusting the phase relationship of the risk wave, the possibility of risk rebound is reduced, for example: in a vibration conduction loop, after inserting the phase offset module, the phase of the vibration wave will change, thereby reducing the reflection and superposition of vibration in the loop and reducing the impact of risk on the device.

[0061] The technical solutions in the embodiments of the present application have at least the following technical effects or advantages: by combining digital blocking barriers and intelligent circuit breakers, the risk conduction path can be more accurately blocked, risk rebound can be prevented, healthy devices can be effectively protected, the reliability of the overall operation of the device can be improved, dynamic topology analysis of device risk can be realized, risk conduction can be accurately predicted and effectively blocked through risk flow simulation and digital twin blocking technology, risk rebound can be prevented, the stability and reliability of device operation can be improved, and the smooth progress of production can be ensured.

[0062] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An AI-based digital plant intelligent control method, characterized by, The method comprises the following steps: S101, collecting data, extracting the characteristics of the data, calculating a health index according to the characteristics of the data, and warning the equipment based on the health index, wherein the data comprises operation data of the equipment and basic data of the equipment; S102, collecting operation data of the equipment in a healthy state, establishing a baseline based on the operation data, comparing the real-time collected characteristic value with the baseline, identifying a state change critical point, and tracking a deterioration trajectory according to the state change critical point; S103, identifying a degradation state according to the state change critical point, calculating a reversibility probability according to the degradation state of the equipment, obtaining a reversible interval and an irreversible point of the degradation of the equipment based on the reversibility probability, and implementing a corresponding intervention strategy according to the reversibility; S104, predicting the time of failure occurrence according to the characteristics of the data, calculating the best intervention time based on the predicted failure occurrence time, and matching the means according to the means matching rule base according to the failure characteristics; S201, obtaining a process relationship of the equipment according to the collected basic data of the equipment, calculating a dependence strength, and constructing a dependence relationship of a production chain based on the dependence strength and the equipment information; S202, collecting historical failure data, constructing a propagation model according to the historical failure data, real-time monitoring a risk source node, and generating a blocking strategy based on the risk source node.

2. The AI-based digital chemical plant intelligent control method of claim 1, wherein, The calculation formula of the health index is: wherein, is the health index, is the energy spectrum entropy, is the current unbalance degree, is the voiceprint anomaly score, is the qualified rate slope, is the weight of the energy spectrum entropy, which is obtained according to the mechanical wear degree and the vibration signal, is the weight of the current unbalance degree, which is obtained according to the electrical fault, is the weight of the voiceprint anomaly, which is obtained according to the voiceprint feature, is the weight of the qualified rate slope, which is obtained according to the deterioration degree of the production quality, .

3. The AI-based digital plant intelligent control method of claim 1, wherein, The baseline is a range interval that is adjusted in real time with the equipment operation state, environmental changes and equipment aging factors, the data characteristics include energy spectrum entropy, current unbalance degree, voiceprint abnormality and qualified rate slope, and the baseline range is the average value of the data characteristics in a period of time plus or minus twice the standard deviation.

4. The AI-based digital plant intelligent control method of claim 1, wherein The state change critical point comprises a first grade critical point and a second grade critical point, the first grade critical point is a point at which the health index rises and continuously rises in a period of time, but does not reach a preset risk threshold, and the second grade critical point is a point at which the health index rises and exceeds the preset risk threshold.

5. The AI-based digital plant intelligent control method of claim 1, wherein, The formula for calculating the reversibility probability is: wherein, represents the reversibility probability, n is the number of reversible evidence, N is the total number of evidence, reversibility is determined according to the reversibility probability calculated, a probability threshold range is set, the threshold range includes a highest threshold and a lowest threshold, when the reversibility probability is greater than the highest threshold, the degradation of the device is the highest reversibility, when the reversibility probability is less than or equal to the highest threshold and greater than the lowest threshold, the degradation of the device is the critical reversibility, and when the reversibility probability is less than or equal to the lowest threshold, the degradation of the device is irreversible.

6. The AI-based digital plant intelligent control method of claim 1, wherein, The risk source node is divided into three risk grades of single-point high-risk, multi-node medium-risk and whole-chain high-risk, the single-point high-risk is only one risk source node, the multi-node medium-risk is multiple risk source nodes, and the whole-chain high-risk is that there are multiple risk source nodes, and the multiple risk source nodes are related and interact with each other, resulting in a whole line shutdown.

7. The AI-based digital factory intelligent control method of claim 1, wherein S301, obtaining equipment degradation entropy value data according to the operation data of the equipment and the failure frequency, calculating a risk pressure value according to the equipment degradation entropy value data, calculating a risk conduction strength based on the risk pressure value, identifying a risk flow direction, and generating a risk flow line graph based on the risk flow direction; S302, real-time monitoring of risk flow intensity, generating a digital blocking barrier based on the risk flow intensity, identifying a risk flow loop according to the algorithm, and adjusting the strategy according to the risk flow loop grade.

8. The AI-based digital chemical plant intelligent control method of claim 7, wherein, The formula for calculating the risk propagation intensity is: wherein is the risk propagation intensity, is the risk pressure value of the device i, is the risk propagation characteristic data between the device i and the device j, is the weight of the risk pressure value, is the weight of the risk propagation characteristic data, + = 1.

9. The AI-based digital chemical plant intelligent control method of claim 7, wherein, The risk flow direction refers to the conduction from a high-pressure equipment to a low-pressure equipment.

Images