Production line equipment fault prediction method and system

Abstract

The invention provides a production line equipment fault prediction method and system, and relates to the technical field of fault prediction.The method comprises the steps that multi-source time sequence signals are synchronously collected, a multi-dimensional health feature sequence is extracted, the comprehensive sensing capacity covering multiple states of equipment is established, and further, the fault prediction efficiency is improved. By selecting a key health feature sequence for each device and performing linear fitting to quantify the performance degradation severity, accurate description and early recognition of the performance degradation trend of the device are realized, and in addition, the causal influence intensity between the devices is quantified by adopting a transfer entropy algorithm, so that the accuracy of the performance degradation trend is improved. Scientific modeling and visual analysis of a fault propagation path and a linkage effect in a production line are realized, and a comprehensive risk index is generated through weighted fusion of performance degradation severity and a global impact factor, so that a comprehensive risk assessment index is formed. Based on the cumulative failure probability function, the comprehensive risk index is converted into a fault probability curve changing along with time, a visual mathematical expression of the risk trend is formed, and a scientific basis is provided for decision making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction, in particular to a production line equipment fault prediction method and system. BACKGROUND

[0002] In the copper-containing etching solution industry, the three-effect evaporation wastewater automatic treatment control system is the core link for modern industry to achieve environmental protection standards and resource recycling. The system realizes efficient removal of pollutants such as ammonia nitrogen and heavy metal ions in wastewater through precise control of the collection pool, reaction pool, discharge pool, and supporting automatic feeding machine, conveying pump, online monitoring instrument and other equipment working in coordination. However, the stable operation of the system highly depends on the reliability of each key equipment. The traditional maintenance strategy mainly adopts the periodic maintenance or after-service maintenance mode, which cannot early perceive and predict the performance degradation of rotating equipment such as evaporator circulating pumps, dosing pumps and conveying pumps. At the same time, due to the strong process coupling and dynamic correlation between production line equipment, the hidden failure of a certain equipment can be transmitted to the downstream link through material flow and energy flow, leading to chain reaction and even system paralysis. The more serious challenge is that the existing technology lacks quantitative analysis capability of the dynamic correlation between equipment states, and cannot trace the performance degradation of specific equipment before equipment anomaly. The operation and maintenance personnel can only rely on manual regular inspection to judge the equipment health status. This method not only has strong lag and high labor cost, but also cannot avoid unplanned shutdown due to sudden failure, which seriously restricts the continuous and stable operation of the production line and economic benefits.

[0003] Therefore, it is necessary to provide a production line equipment fault prediction method and system to solve the above technical problems. SUMMARY

[0004] To solve the above technical problems, the present application provides a production line equipment fault prediction method and system, which can early warn of fault occurrence when the equipment performance in the production line degrades.

[0005] The present application provides a production line equipment fault prediction method, comprising: S1: synchronously collecting multi-source original time sequence signals of each equipment on the production line, segmenting each original time sequence signal based on a preset time window, and extracting a health feature value in each time window to obtain a health feature sequence of each equipment in multiple dimensions; S2: selecting a unique key health feature sequence for each equipment based on a preset selection rule, and linearly fitting the key health feature sequences of each equipment to obtain the performance degradation severity of each equipment; S3: based on the key health feature sequences of all equipment, using a transfer entropy algorithm to sequentially calculate the causal influence intensity of all upstream equipment on the current equipment for each equipment on the production line; S4: for each current device, multiply its all upstream device's causal influence strength on the current device by the performance degradation severity of the upstream device, sum up to obtain a global influence factor of each device, and weight and fuse the performance degradation severity of each device with the global influence factor to obtain a comprehensive risk index of each device; S5: multiply the comprehensive risk index of each device by a preset device failure rate coefficient to obtain a risk growth rate coefficient, and substitute the risk growth rate coefficient into a preset cumulative failure probability function to generate a failure change probability curve of each device; S6: analyze the failure risk change probability curve of each device, determine the future time point when the failure probability exceeds the preset warning threshold, and generate predictive warning information and targeted maintenance decisions within the remaining effective time window for each device accordingly.

[0006] Preferably, in step S2, the performance degradation severity obtaining step comprises: obtaining the slope value of the fitting straight line obtained by linear fitting as the real-time performance degradation rate of the device; dividing the real-time performance degradation rate by the preset health benchmark degradation rate threshold of the device to obtain the performance degradation severity.

[0007] Preferably, in step S2, the linear fitting adopts robust regression fitting based on least squares method.

[0008] Preferably, in step S3, only the transfer entropy value with statistical significance exceeding 95% is retained as the causal influence strength.

[0009] Preferably, in step S4, the calculation formula of the comprehensive risk index is: wherein, is the comprehensive risk index, is the performance degradation severity, is the global influence factor, is the weight coefficient of the performance degradation severity, is the weight coefficient of the global influence factor.

[0010] Preferably, the weight coefficient of the performance degradation severity and the weight coefficient of the global influence factor are determined by the following steps: assign a unique position index number to each device on the production line, the index number increasing from the most upstream device to the most downstream device of the process chain of the production line; divide the position index number of the current device by the maximum index number of the devices on the production line to obtain the relative position proportion of the current device; The relative position ratio is multiplied by a preset proportion adjustment factor, and a preset basic weight value is added to obtain a weight coefficient of the performance degradation severity of the current device , 1 minus the weight coefficient , to obtain a weight coefficient of the global influence factor .

[0011] Preferably, in step S5, the preset cumulative failure probability function is: wherein, is a risk growth rate coefficient, is a future time period from the current time, is a failure probability of the device in the future time unit, is a natural exponential function symbol.

[0012] Preferably, in step S5, the preset device failure rate coefficient obtaining step comprises the following steps: During a preset time period when the device is in a healthy running state, the integrated risk index of the device is continuously recorded to obtain a healthy risk index sample set; The statistical quantile of the healthy risk index sample set is calculated, and the value of the 95th quantile is set as the health warning threshold of the integrated risk index of the device; When the integrated risk index reaches the health warning threshold, the instantaneous failure probability of the device is defined as a preset risk reference probability; Based on the risk reference probability and the preset cumulative failure probability function, the reference risk growth rate coefficient corresponding to the risk reference probability in a unit time span is calculated by back calculation; The reference risk growth rate coefficient is divided by the health warning threshold to obtain the device failure rate coefficient.

[0013] Preferably, in step S6, the preset warning threshold is obtained in the following manner: According to the criticality of the device in the production line, a criticality level is assigned to each device, and a preset basic warning probability value is assigned according to the criticality level; The standard deviation of the historical sequence of the integrated risk index of each device in the healthy running state is calculated; The preset basic warning probability value and the reciprocal of the standard deviation of the historical sequence of the integrated risk index are weighted and fused according to a preset weighting weight to obtain a preset warning threshold of each device.

[0014] The application also provides a production line device failure prediction system applied to the production line device failure prediction method, comprising: The health feature extraction module is configured to synchronously collect multi-source original time sequence signals of each device on the production line, segment each original time sequence signal based on a preset time window, and extract a health feature value in each time window to obtain a health feature sequence of each device in multiple dimensions; The device performance degradation quantification module is configured to select a unique key health feature sequence for each device based on a preset selection rule, and perform linear fitting on the key health feature sequence of each device to obtain a performance degradation severity of each device; The system causal influence analysis module is configured to calculate the causal influence intensity of all upstream devices on each device on the production line based on the key health feature sequences of all devices and using a transfer entropy algorithm. The comprehensive risk calculation module is configured to multiply the causal influence intensity of all upstream devices on each device by the performance degradation severity of the upstream device, sum the results, and obtain a global influence factor of each device, and then weight and fuse the performance degradation severity and the global influence factor to obtain a comprehensive risk index of each device. The fault risk probability prediction module is configured to multiply the comprehensive risk index of each device by a preset device failure rate coefficient to obtain a risk growth rate coefficient, and substitute the risk growth rate coefficient into a preset cumulative failure probability function to generate a fault change probability curve of each device. The predictive decision generation module is configured to analyze the fault risk change probability curve of each device, determine a future time point at which the fault probability exceeds a preset warning threshold, and generate predictive warning information and a targeted maintenance decision within a remaining valid time window for each device based on the future time point.

[0015] Compared with related technologies, the production line device fault prediction method and system provided by the present application has the following beneficial effects: The application establishes comprehensive perception ability covering various states of the equipment by synchronously collecting multi-source time sequence signals and extracting multi-dimensional health feature sequences, lays a data foundation for accurately evaluating the health state of the equipment, further, realizes accurate characterization and early identification of the performance degradation trend of the equipment by dynamically selecting key health feature sequences for each piece of equipment and linearly fitting the performance degradation severity, in addition, realizes scientific modeling and visual analysis of the fault propagation path and chain effect in the production line by quantifying the causal influence strength between the equipment using the transfer entropy algorithm, thereby improving the prediction dimension from a single point of equipment to a system-level network, generating a comprehensive risk index by weighting and fusing the performance degradation severity and the global influence factor, organically combining the health state of the equipment itself and the vulnerability affected by the outside world, forming a comprehensive and dynamic risk evaluation index, converting the comprehensive risk index into a fault probability curve changing with time based on the cumulative failure probability function, so that the abstract "risk" is quantified into the "probability" that can be directly interpreted, providing a scientific basis for decision-making, finally, determining the fault overrun time point and generating targeted maintenance decisions through the analysis of the probability curve, not only providing early warning, but also indicating the maintenance window and specific measures, converting the prediction result into an executable action guide, and finally realizing accurate and efficient prediction of the production line equipment failure. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flow chart of a production line equipment failure prediction method of the application; Figure 2 A module structure diagram of a production line equipment failure prediction system of the application. DETAILED DESCRIPTION

[0017] The application will be described in further detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the application, and not to limit the application. In addition, it should be noted that, for the sake of convenience, only the parts related to the application are shown in the drawings, not all structures. Furthermore, the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0018] In addition, it should be noted that, for the sake of convenience, only the parts related to the application are shown in the drawings, not all contents. Before discussing the example embodiments in more detail, it should be mentioned that some example embodiments are described as processes or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operations are completed, but can also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0019] Embodiment one A production line equipment failure prediction method, in the specific implementation process, as shown in the figure, it shows a flow chart of a production line equipment failure prediction method of the application, comprising: Figure 1 Step S1: synchronously collecting multi-source original time sequence signals of each equipment on the production line, segmenting each original time sequence signal based on a preset time window, and extracting health feature values in each time window respectively to obtain health feature sequences of each equipment in multiple dimensions. In the specific implementation process, for example, a plurality of sensor data acquisition nodes are independently deployed for each equipment on the production line. For a specific pump equipment on the production line, the multi-source signal acquisition includes but is not limited to obtaining the vibration original waveform signal representing the mechanical operating state through the vibration acceleration sensor installed on the driving end of the pump body, obtaining the temperature original signal representing the thermodynamic state through the temperature sensor attached to the pump bearing seat, and obtaining the current original signal representing the electrical load characteristics through the current transformer connected in series in the pump motor power supply circuit. All sensor data are synchronously triggered and sampled by a unified time data acquisition unit to ensure that the data across physical quantities are strictly aligned in time domain. Segmenting each original time sequence signal based on a preset time window means that the continuous vibration, temperature and current signal streams of the pump equipment are sliced according to the set time length to form independent data analysis windows. Extracting health feature values in each time window respectively means that for a single data window of the pump equipment, the kurtosis value of the vibration signal is calculated in parallel to capture the instantaneous impact, the first order differential slope of the temperature signal is calculated to quantify the temperature rise rate, and the fluctuation variance of the current signal is calculated to evaluate the running stability, so as to generate health feature values in multiple index dimensions for the single time window of the equipment. When segmenting each original time sequence signal based on a preset time window, the length of the time window is set according to the characteristics of the equipment. For example, for high-speed rotating equipment, at least two hundred complete rotation periods are required to capture the characteristic frequency, and for heating equipment with large thermal inertia, the window needs to be extended to observe the effective temperature rise trend. Arranging the feature values of multiple windows in time sequence, i.e. obtaining the set of health feature sequences of the equipment in multiple dimensions, performing the above operation for each equipment in the production line, and finally obtaining the health feature sequence library of all equipment in multiple dimensions to establish a rich data basis for subsequent analysis.

[0020] Step S2: selecting a unique key health feature sequence for each equipment based on a preset selection rule, and linearly fitting the key health feature sequences of each equipment to obtain the performance degradation severity of each equipment.

[0021]

[0022] Specifically, in step S2, the obtaining step of the performance degradation severity comprises:​ a slope value of the fitting straight line obtained by the linear fitting is taken as a real-time performance degradation rate of the equipment; the real-time performance degradation rate is divided by a health benchmark degradation rate threshold preset for the equipment to obtain a performance degradation severity.

[0023] Specifically, in step S2, the linear fitting adopts a robust regression fitting based on a least square method.

[0024] In the implementation process, first, a feature mapping rule is established according to the physical correlation between the equipment type and the failure mode, a unique key health feature sequence is selected for each equipment on the production line, for example, for a rotating machinery type equipment, a vibration kurtosis sequence is selected as the key health feature sequence by default, because it has high sensitivity to mechanical failures such as bearing pitting and gear wear, for a thermal equipment, a temperature change rate sequence is selected as the key health feature sequence by default, because it directly reflects the heat exchange efficiency decline and overheating failure trend, for a fluid conveying equipment, a current fluctuation sequence is selected as the key health feature sequence by default, because it can effectively capture cavitation phenomenon and abnormal load change; then, the key health feature sequence is linearly fitted, and a robust regression algorithm based on a least square method is adopted in the linear fitting, which reduces the interference of abnormal values in the sequence on the fitting result through iterative weighting, so that reliable performance degradation trend estimation can be obtained even in the case of pulse noise in the data; a slope value of the straight line obtained by the fitting is taken as a real-time performance degradation rate, which quantifies the change amplitude of the key health feature per unit time, and a positive value increase indicates that the equipment state is accelerating deterioration; the real-time performance degradation rate is divided by a health benchmark degradation rate threshold preset for the equipment to obtain a performance degradation severity, the benchmark threshold is obtained by linear fitting of the historical feature sequence of the equipment during the health running verification period, and represents the natural degradation speed of the equipment in the normal aging process; the performance degradation severity as a dimensionless ratio directly reflects the multiple relationship of the current degradation speed relative to the normal aging speed, when the value is close to 1, it indicates that the equipment is in normal aging state, and when the value is significantly greater than 1, it indicates that the equipment enters an abnormal degradation stage and needs attention; in addition, the logical relationship of the health feature sequence extraction of each equipment in multiple dimensions and the key health feature sequence used, the initial multi-dimensional feature collection establishes a complete equipment health state information library, although a single key health feature sequence is selected for trend fitting in subsequent analysis, but the multi-dimensional feature reserve provides data basis for diagnosis of different failure modes, when the equipment fails, the system can determine the most relevant feature dimension through backtracking analysis, and then dynamically update the key feature selection rule, so as to ensure that the system has continuous learning and optimization ability, and at the same time provides multi-angle data support for subsequent fault root cause analysis.

[0025] Step S3: Based on the sequence of key health features of all devices, the transfer entropy algorithm is used to calculate the causal influence strength of all upstream devices on the current device in sequence.

[0026] Specifically, in step S3, only the transfer entropy values with statistical significance exceeding 95% are retained as the causal influence strength.

[0027] In the implementation process, first, the transfer entropy algorithm is used to calculate the causal influence strength. For example, for any current device on the production line, all devices located upstream of the process chain form a device pair with the current device; for any device pair, the calculation of the causal influence strength is based on the respective key health feature sequences of the upstream device and the current device in the device pair; here, the key health feature sequence is a sequence formed by arranging the health feature values calculated on a plurality of continuous time windows in chronological order, and each feature value in the sequence is called a data point, representing the quantitative result of the health status of the device in the corresponding time window; the algorithm processes each data point in the key health feature sequence of the current device in the device pair in chronological order; when processing the data point at a specific time position in the key health feature sequence of the current device, the algorithm extracts the values of a plurality of consecutive data points in the key health feature sequence of the upstream device before the time position to form a representation of the past state of the upstream device, and extracts the value of the next data point in the key health feature sequence of the current device after the time position to form a representation of the next state of the current device; through the conditional mutual information calculation method in information theory, the uncertainty that can be reduced for predicting the next state of the current device is quantified after introducing the past state sequence of the upstream device under the condition of knowing the past state sequence of the current device itself, and the reduction of this uncertainty is the transfer entropy value at the specific time position; for all data points in the key health feature sequence of the current device that can simultaneously satisfy the two conditions of successfully extracting the complete past state sequence of the upstream device and successfully extracting the next state of the current device, the transfer entropy values thereof are calculated respectively, and all these transfer entropy values are arithmetically averaged to finally obtain the causal influence strength of the device on the current device of the upstream device pair; in the same way, the causal influence strength of any device in the production line on the device is obtained. In addition, statistical significance test is performed on each calculated transfer entropy value to distinguish between real causal relationship and random correlation. The time series randomization test method is used to generate alternative data sequences by disturbing the time order of the feature sequence of the upstream device and recalculate a large number of alternative transfer entropy values. These alternative values are constructed into an empirical distribution under the assumption of no causal relationship. The observed transfer entropy value is compared with the empirical distribution to calculate the p value. Only when the p value is less than or equal to 0.05, the transfer entropy value is considered statistically significant, indicating that the causal influence strength of the upstream device on the current device is real and not a random coincidence. Finally, all transfer entropy values that pass the significance test are retained as valid causal influence strength values for subsequent calculations, and the values that do not pass the test are considered as zero to avoid introducing false correlation noise, providing high-quality and reliable causal relationship input data for subsequent global risk fusion.

[0028] Step S4: for each current device, multiply the causal influence strength of all its upstream devices on the current device by the performance degradation severity of each upstream device, sum up the results to obtain the global influence factor of each device, and weight and fuse the performance degradation severity of each device with the global influence factor to obtain the comprehensive risk index of each device.

[0029] Specifically, in step S4, the calculation formula of the comprehensive risk index is: wherein, is the comprehensive risk index, is the performance degradation severity, is the global influence factor, is the weight coefficient of the performance degradation severity, is the weight coefficient of the global influence factor.

[0030] Specifically, the weight coefficient of the performance degradation severity and the weight coefficient of the global influence factor are determined by the following steps: assign a unique position index number to each device on the production line, the index number increasing from the most upstream device to the most downstream device in the process chain of the production line; divide the position index number of the current device by the maximum index number of the devices on the production line to obtain the relative position proportion of the current device; multiply the relative position proportion by a preset proportion adjustment factor and add a preset basic weight value to obtain the weight coefficient of the performance degradation severity of the current device , subtract the weight coefficient of the performance degradation severity of the current device from 1 to obtain the weight coefficient of the global influence factor .

[0031] In the specific implementation process, for each current device, first multiply the causal influence strength of all its upstream devices on the current device by the performance degradation severity of each upstream device, the essence of this multiplication operation is to weight and calculate the causal influence strength by taking the performance degradation severity as the weight coefficient, so as to quantify the actual influence degree of the upstream device on the current device due to its own performance degradation level; then sum up all the weighted results to obtain the global influence factor of the current device, which represents the comprehensive external risk pressure caused by all upstream devices on the current device through the fault propagation path; then weight and fuse the performance degradation severity of the current device with the global influence factor, wherein the calculation formula of the weight and fusion is that the comprehensive risk index is equal to the performance degradation severity multiplied by the weight coefficient plus the global influence factor multiplied by the weight coefficient Weighting coefficient and The determination method includes assigning a unique position index number to each piece of equipment on the production line, starting from the upstream equipment in the process chain and incrementing sequentially; dividing the current equipment's position index number by the maximum index number of equipment on the production line to obtain the relative position ratio; multiplying this relative position ratio by a preset ratio adjustment factor and adding a preset base weight value to obtain a weight coefficient. Then subtract the weighting coefficient from 1. Obtain the weighting coefficients This weighting strategy ensures that the weighting coefficients for the severity of performance degradation of equipment located downstream in the process chain are applied accordingly. It increases with the increase of the position index number, while the weight coefficient of the global influence factor... The corresponding decrease reflects the characteristic that downstream equipment is more dependent on its own state, while upstream equipment is more susceptible to the overall system impact. Through this weighted fusion, a comprehensive risk index for each piece of equipment is finally obtained. This index organically integrates the internal factors of equipment performance degradation and the external factors affected by the system, providing a unified risk quantification indicator for subsequent accurate prediction.

[0032] Step S5: Multiply the comprehensive risk index of each device by the preset device failure rate coefficient to obtain the risk growth rate coefficient. Substitute the risk growth rate coefficient into the preset cumulative failure probability function to generate the failure change probability curve of each device.

[0033] Specifically, in step S5, the preset cumulative failure probability function is: in, This represents the risk growth rate coefficient. A future time period starting from the current moment. For the device in the future The probability of a failure occurring within a unit of time. This is the symbol for the natural exponential function.

[0034] Specifically, in step S5, the step of obtaining the preset equipment failure rate coefficient includes the following steps: During a preset time period when the equipment is in a healthy operating state, its comprehensive risk index is continuously recorded to obtain a health risk index sample set. Calculate the statistical quantile of the health risk index sample set, and set the value of the 95th quantile as the health warning threshold of the comprehensive risk index of the device; When the comprehensive risk index reaches the health warning threshold, the instantaneous failure probability of the equipment is defined as the preset risk benchmark probability; based on the risk benchmark probability and a preset cumulative failure probability function, a benchmark risk growth rate coefficient corresponding to the risk benchmark probability reached by the equipment in a unit time span is reversely calculated; The benchmark risk growth rate coefficient is divided by the health warning threshold to obtain an equipment failure rate coefficient.

[0035] In the implementation process, the comprehensive risk index of each equipment is multiplied by a preset equipment failure rate coefficient to obtain a risk growth rate coefficient, wherein the preset value of the equipment failure rate coefficient is obtained in the following manner: a health risk index sample set is formed by continuously recording the comprehensive risk index of the equipment in a preset time period when the equipment is in a healthy running state, the statistical quantile of the sample set is calculated, and the 95th quantile value is set as the health warning threshold of the equipment comprehensive risk index. The instantaneous failure probability of the equipment when the comprehensive risk index reaches the health warning threshold is defined as a preset risk benchmark probability. Based on the risk benchmark probability and a preset cumulative failure probability function, a benchmark risk growth rate coefficient corresponding to the risk benchmark probability reached by the equipment in a unit time span is reversely calculated, wherein the unit time span refers to a predefined time benchmark unit for standardized calculation, for example, 1 hour or 1 minute, so that it can be correctly substituted into the subsequent cumulative failure probability function for mathematical operation. Then, the risk growth rate coefficient is substituted into the preset cumulative failure probability function, which is in the form of the failure probability of the equipment in the future unit time span, which is equal to 1 minus the calculation result of the negative risk growth rate coefficient of the natural exponential function e multiplied by the unit time span power. The cumulative failure probability value of the equipment at different future time points starting from the current time is calculated through the function, wherein is a continuous unit time span that gradually increases from zero. By calculating the failure probability value corresponding to different values, a series of data points corresponding to future time points and probabilities are obtained, and these data points are connected to form a smooth curve of probability change over time, i.e., the failure change probability curve of the equipment. The curve directly shows the change trend of the cumulative growth of the equipment failure risk over time, and provides a visual basis for subsequent warning decisions.

[0036] Step S6: Analyze the failure risk change probability curve of each equipment, determine the future time point at which the failure probability exceeds the preset warning threshold, and generate predictive warning information and targeted maintenance decisions within the remaining effective time window for each equipment accordingly.

[0037] Specifically, in step S6, the preset warning threshold is obtained in the following manner: According to the criticality of the equipment in the production line, a criticality level is assigned to each equipment, and a preset basic warning probability value is assigned according to the criticality level; The standard deviation of the historical sequence of the comprehensive risk index of each equipment in the healthy running state is calculated; The preset basic warning probability value and the reciprocal of the standard deviation of the historical sequence of the comprehensive risk index are weighted and fused according to a preset weighting weight to obtain a preset warning threshold of each equipment.

[0038] In the implementation process, the failure probability value corresponding to each future time point on the failure risk change probability curve of each equipment is calculated by traversal, and it is compared with the warning threshold set individually for the equipment in real time. The warning threshold is obtained by, for example, according to the criticality level of the equipment in the whole production line process, a basic warning probability value is assigned to the equipment, the basic warning probability value decreases with the increase of the criticality of the equipment to realize earlier warning of the critical equipment, and the standard deviation of the sequence of the comprehensive risk index of the equipment in the healthy running history period is calculated to quantify its inherent fluctuation characteristics. Then, the basic warning probability value and the reciprocal of the standard deviation are weighted and fused according to a preset weight. The reciprocal of the standard deviation plays a role in automatically adjusting the threshold adaptability. For the equipment with large historical fluctuations, the warning threshold is appropriately increased to avoid false alarms, and for the equipment with stable operation, the threshold is reduced to improve the warning sensitivity. Through this weighted fusion, a personalized warning threshold dynamically adapting to the characteristics of the equipment and the process requirements is finally generated. When the failure probability value on the probability curve first exceeds the warning threshold, the system records the time point as the warning trigger point, and calculates the time interval from the current time to the trigger point as the remaining effective time window. Finally, the system generates predictive warning information containing the specific equipment number, the warning trigger time point, the remaining effective time window length and the recommended maintenance measures. The information provides clear failure expectations and sufficient response time for equipment maintenance personnel, thereby realizing closed-loop management from prediction to execution and efficient failure prediction of production line equipment.

[0039] The working principle of the production line equipment failure prediction method provided by the application is as follows: The present application synchronously collects multi-source original time sequence signals of each device on the production line, extracts multi-dimensional health characteristic values, forms a device health state panoramic image, dynamically selects the most sensitive key health characteristic sequence for each device based on a preset rule, quantifies the performance degradation severity through linear fitting to capture the device's own performance degradation trend, analyzes the information flow among the key health characteristic sequences of all devices using a transfer entropy algorithm, and only retains the transfer entropy values with statistical significance exceeding 95% to accurately quantify the causal influence strength of the upstream device on the current device, multiplies the causal influence strength and the performance degradation severity of the corresponding upstream device, and then sums them up to obtain a global influence factor to represent the system correlation fault propagation effect, dynamically weights and fuses the performance degradation severity and the global influence factor according to the device process position to obtain a comprehensive risk index, so as to realize unified evaluation of the device's own degradation and external influence, multiplies the comprehensive risk index and the device failure rate coefficient obtained by calibrating the health history data to obtain a risk growth rate coefficient with a time dimension, and substitutes it into an exponential cumulative failure probability function to calculate the fault probability value at different future time points to generate a fault risk change probability curve, and finally determines the time point at which the fault probability exceeds the warning threshold set based on the device criticality and historical volatility through analysis of the probability curve and generates a predictive maintenance decision, forming a cross from device individual health state monitoring to production line system risk linkage prediction realized through a data-driven manner.

[0040] Embodiment two A production line device fault prediction system, in the specific implementation process, as shown in Figure 2 , it shows a module structure diagram of a production line device fault prediction system of the present application, which comprises: The health characteristic extraction module 100 is used for synchronously collecting multi-source original time sequence signals of each device on the production line, segmenting each original time sequence signal based on a preset time window, and extracting health characteristic values in each time window to obtain health characteristic sequences of each device in multiple dimensions; The device performance degradation quantification module 200 is used for selecting a unique key health characteristic sequence for each device based on a preset selection rule, and performing linear fitting on the key health characteristic sequences of each device to obtain the performance degradation severity of each device; The system causal influence analysis module 300 is used for calculating the causal influence strength of all upstream devices on the current device for each device on the production line in turn based on the key health characteristic sequences of all devices using a transfer entropy algorithm; The comprehensive risk calculation module 400 is used for multiplying the causal influence strength of all upstream devices on the current device by the performance degradation severity of the upstream device for each current device, summing them up to obtain the global influence factor of each device, and weighting and fusing the performance degradation severity and the global influence factor of each device to obtain the comprehensive risk index of each device; The fault risk probability prediction module 500 is used for multiplying the comprehensive risk index of each device by a preset device failure rate coefficient to obtain a risk growth rate coefficient, and substituting the risk growth rate coefficient into a preset cumulative failure probability function to generate a fault change probability curve of each device. The predictive decision generation module 600 is used for analyzing the fault risk change probability curve of each device, determining a future time point at which the fault probability exceeds a preset early warning threshold, and generating predictive early warning information and a targeted maintenance decision within a remaining valid time window of each device according to the future time point.

[0041] The working principle of the production line device fault prediction system provided by the application is as follows: First, the health feature extraction module 100 synchronously collects multi-source original signals of the production line device, and divides a time window to extract multi-dimensional health feature values, thereby forming a device health state basic data set. The device performance degradation quantification module 200 selects a key health feature sequence based on a preset rule, and performs linear fitting to calculate a performance degradation severity, so as to quantify the degree of performance decline of the device. The system causal influence analysis module 300 analyzes the information flow among all key health feature sequences of the devices by using a transfer entropy algorithm, and calculates statistically significant causal influence strength, so as to construct a system-level fault propagation network. The comprehensive risk calculation module 400 multiplies the causal influence strength of each upstream device by the performance degradation severity of the device to obtain a global influence factor, and fuses the factor with the performance degradation severity of the device according to a dynamic weight to generate a comprehensive risk index, so as to realize unified evaluation of internal and external risk elements. The fault risk probability prediction module 500 multiplies the comprehensive risk index by a preset device failure rate coefficient to obtain a risk growth rate coefficient, and substitutes the risk growth rate coefficient into an exponential cumulative failure probability function to calculate fault probability values at different future time points, thereby generating a probability time curve. The predictive decision generation module 600 analyzes the curve and compares the curve with an individualized early warning threshold to determine a fault overrun time point and generate an early warning decision including a remaining valid time window and a targeted measure. Each module is connected through a data stream to form a fault prediction system of the closed-loop automated production line device from multi-source perception to a prediction decision.

[0042] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 Each flow or multiple flows and / or blocks Figure 1means for performing the function specified by the block or blocks.

[0043] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiments can be completed by instructing relevant hardware through programs, and the programs can be stored in a computer readable storage medium, including Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM) or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other medium that can be used to carry or store data in a computer readable manner.

[0044] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the statement "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

Claims

1. A method for predicting equipment failures on a production line, characterized in that, The equipment failure prediction method includes the following steps: S1: Synchronously collect multi-source raw time-series signals from various devices on the production line, segment each raw time-series signal based on a preset time window, and extract the health feature values ​​within each time window to obtain the health feature sequence of each device in multiple dimensions. S2: Based on preset selection rules, a unique key health feature sequence is selected for each device, and the key health feature sequence of each device is linearly fitted to obtain the performance degradation severity of each device; S3: Based on the key health feature sequences of all equipment, the transfer entropy algorithm is used to calculate the causal influence intensity of all upstream equipment on the current equipment for each equipment on the production line in turn; S4: For each current device, multiply the causal influence strength of all its upstream devices on the current device by the performance degradation severity of that upstream device and sum them to obtain the global influence factor of each device. Then, weight and fuse the performance degradation severity of each device with the global influence factor to obtain the comprehensive risk index of each device. S5: Multiply the comprehensive risk index of each device by the preset device failure rate coefficient to obtain the risk growth rate coefficient. Substitute the risk growth rate coefficient into the preset cumulative failure probability function to generate the failure change probability curve of each device. S6: Analyze the failure risk change probability curve of each device, determine the future time point when the failure probability exceeds the preset warning threshold, and generate predictive warning information for each device and targeted maintenance decisions within the remaining effective time window.

2. The method for predicting production line equipment failure according to claim 1, characterized in that, In step S2, the steps for obtaining the severity of performance degradation include: The slope of the fitted line obtained by linear fitting is used as the real-time performance degradation rate of the device. The severity of performance degradation is obtained by dividing the real-time performance degradation rate by the device's preset health baseline degradation rate threshold.

3. The method for predicting production line equipment failure according to claim 2, characterized in that, In step S2, the linear fitting employs robust regression fitting based on the least squares method.

4. The method for predicting production line equipment failure according to claim 3, characterized in that, In step S3, only the transfer entropy values ​​with a statistical significance exceeding 95% are retained as the causal influence strength.

5. The method for predicting production line equipment failure according to claim 4, characterized in that, In step S4, the formula for calculating the comprehensive risk index is: in, As a comprehensive risk index, For the severity of performance degradation, As a global impact factor, The weighting coefficients for the severity of performance degradation. This represents the weighting coefficient of the global impact factor.

6. The method for predicting production line equipment failure according to claim 5, characterized in that, Weighting coefficients for the severity of performance degradation Weighting coefficients of global impact factors The method for determining it includes the following steps: Each piece of equipment on the production line is assigned a unique location index number, which increases from the upstream equipment in the production line's process chain to the downstream equipment. Divide the current device's location index number by the maximum device index number on the production line to obtain the relative position ratio of the current device. Multiply the relative position ratio by a preset ratio adjustment factor, and add a preset base weight value to obtain the weighting coefficient for the severity of the current device's performance degradation. 1 minus the weighting coefficient Obtain the weight coefficients of the global impact factor. .

7. The method for predicting production line equipment failure according to claim 6, characterized in that, In step S5, the preset cumulative failure probability function is: in, This represents the risk growth rate coefficient. A future time period starting from the current moment. For the device in the future The probability of a failure occurring within a unit of time. This is the symbol for the natural exponential function.

8. The method for predicting production line equipment failure according to claim 7, characterized in that, In step S5, the step of obtaining the preset equipment failure rate coefficient includes the following steps: During a preset time period when the equipment is in a healthy operating state, its comprehensive risk index is continuously recorded to obtain a health risk index sample set. Calculate the statistical quantile of the health risk index sample set, and set the value of the 95th quantile as the health warning threshold of the comprehensive risk index of the device; When the comprehensive risk index reaches the health warning threshold, the instantaneous failure probability of the equipment is defined as the preset risk benchmark probability; Based on the risk baseline probability and the preset cumulative failure probability function, the baseline risk growth rate coefficient corresponding to when the equipment reaches the risk baseline probability within a unit time span is calculated. Divide the baseline risk growth rate coefficient by the health warning threshold to obtain the equipment failure rate coefficient.

9. A method for predicting production line equipment failures according to claim 8, characterized in that, In step S6, the preset warning threshold is obtained through the following methods: Based on the criticality of the equipment in the production line, each piece of equipment is assigned a criticality level, and a preset basic early warning probability value is assigned according to the criticality level; Calculate the standard deviation of the historical series of the comprehensive risk index for each device under healthy operating conditions; The preset basic warning probability value and the reciprocal of the standard deviation of the historical series of the comprehensive risk index are weighted and fused according to the preset weighting weight to obtain the preset warning threshold for each device.

10. A production line equipment fault prediction system, characterized in that, The equipment failure prediction system, applied to a production line equipment failure prediction method as described in any one of claims 1 to 9, comprises: The health feature extraction module is used to synchronously collect multi-source raw time-series signals from various devices on the production line, segment each raw time-series signal based on a preset time window, and extract the health feature values ​​within each time window to obtain the health feature sequence of each device in multiple dimensions. The equipment performance degradation quantification module is used to select a unique key health feature sequence for each device based on preset selection rules, and to perform linear fitting on the key health feature sequence of each device to obtain the performance degradation severity of each device. The system causal impact analysis module is used to calculate the causal impact intensity of all upstream devices on the current device for each device on the production line based on the key health feature sequences of all devices and using the transfer entropy algorithm. The comprehensive risk calculation module is used to sum the causal impact strength of all upstream devices on each current device by the performance degradation severity of that upstream device, and obtain the global impact factor of each device. The performance degradation severity of each device is then weighted and fused with the global impact factor to obtain the comprehensive risk index of each device. The fault risk probability prediction module is used to multiply the comprehensive risk index of each device by the preset device failure rate coefficient to obtain the risk growth rate coefficient. The risk growth rate coefficient is then substituted into the preset cumulative failure probability function to generate the fault change probability curve of each device. The predictive decision generation module is used to analyze the failure risk change probability curve of each device, determine the future time point when the failure probability exceeds the preset warning threshold, and generate predictive warning information for each device and targeted maintenance decisions within the remaining effective time window.

Images