A predictive maintenance method for transformer manufacturing equipment

Abstract

This application discloses a predictive maintenance method for transformer production equipment, relating to the technical field of equipment testing. The method includes acquiring operational monitoring data and environmental load parameters of the transformer production equipment to obtain transient pumping response data and interference decoupling variables; obtaining the true pumping efficiency characteristics of the transformer production equipment; collecting vacuum pump power consumption data, performing time-series cross-validation, and obtaining fault root cause classification results; extracting the cumulative high-temperature operating time of the vacuum pump, constructing a virtual sensor model, and obtaining oil remaining life data; combining the fault root cause classification results and oil remaining life data to conduct a comprehensive evaluation of the transformer production equipment and generate an equipment degradation prediction report; acquiring factory production cycle data to obtain the equipment's natural cooling time window, and generating a non-stop maintenance scheduling plan for the production equipment. This application improves the predictability and accuracy of transformer production equipment maintenance.

Description

Technical Field

[0001] This application relates to the technical field of equipment testing, and in particular to a predictive maintenance method for transformer manufacturing equipment. Background Technology

[0002] In the transformer manufacturing process, vacuum drying and vacuum oil injection equipment are extremely critical production equipment. Their operating status and vacuuming efficiency directly affect the internal insulation performance and overall quality of the transformer. Among them, the performance and service life of the core components of the vacuuming system directly determine the operating efficiency and safety of the transformer production line. The actual operating status of this type of equipment depends not only on the wear and tear of the equipment itself, but also on a variety of factors such as the production environment and processing load.

[0003] In existing technologies, the maintenance of transformer production equipment mainly relies on manual, periodic forced shutdowns for maintenance, or automated alarm monitoring based on a single fixed threshold for vacuum level. This approach has several drawbacks. First, single-threshold monitoring cannot isolate external interference factors such as ambient temperature and humidity, and the thermal load of the transformer core and copper wires, easily misinterpreting normal physical fluctuations as equipment malfunctions. Second, when a decrease in pumping efficiency occurs, existing technologies cannot perform cross-verification of multi-dimensional parameters, making it difficult to accurately distinguish between seal aging and leakage and mechanical wear of the pump body, resulting in low troubleshooting efficiency. Third, for highly concealed internal conditions such as vacuum pump oil emulsification and contamination, there is a lack of effective dynamic prediction methods, often resulting in reactive, reactive repairs only after a serious equipment shutdown. Therefore, improving the predictability and accuracy of transformer production equipment maintenance has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a predictive maintenance method for transformer production equipment to solve the problems mentioned in the background art.

[0005] This application provides a predictive maintenance method for transformer manufacturing equipment, the method comprising: Obtain the operation monitoring data and environmental load parameters of the transformer production equipment, obtain the transient gas extraction response data during the vacuuming process based on the operation monitoring data, and obtain the disturbance decoupling variables according to the environmental load parameters; Based on the interference decoupling variables, variable decoupling analysis is performed on the transient pumping response data to obtain the true pumping efficiency characteristics of the transformer production equipment. Data on voltage drop curves during the voltage holding test of transformer production equipment and data on vacuum pump power consumption during the pumping test are collected. Time-series cross-validation is performed by combining the voltage drop curve data, the vacuum pump power consumption data, and the actual pumping efficiency characteristics to obtain the fault root cause classification results. Historical operating data of the vacuum pump is obtained, the cumulative high-temperature operating time of the vacuum pump is extracted, and a virtual sensor model is constructed by combining the actual pumping efficiency characteristics and the cumulative high-temperature operating time. Based on the virtual sensor model, the emulsification and contamination state of the vacuum pump oil is predicted to obtain the remaining life data of the oil. A comprehensive assessment of transformer production equipment is conducted by combining the fault root cause classification results and the remaining oil life data, and an equipment degradation prediction report is generated. Obtain the factory's production cycle time data, determine the equipment's natural cooling time window based on the production cycle time data, and generate a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window.

[0006] Preferably, the steps of acquiring the operation monitoring data and environmental load parameters of the transformer production equipment, obtaining transient pumping response data during the vacuuming process based on the operation monitoring data, and obtaining the disturbance decoupling variables according to the environmental load parameters are as follows: Real-time acquisition of multi-dimensional operation monitoring data of transformer production equipment throughout the entire vacuuming operation cycle. The operation monitoring data includes at least the real-time air pressure value inside the vacuum chamber, the operating frequency of the vacuum pump, and the instantaneous flow parameters of the exhaust port. A standardized time axis corresponding to the entire cycle of vacuuming operation is established, and the real-time air pressure value, the operating frequency, and the instantaneous flow rate parameter are mapped onto the standardized time axis to obtain a multi-dimensional parameter time axis; Identify the pressure drop interval on the time axis of the multidimensional parameters, and extract the transient pumping response data in the pressure drop interval; The real-time environmental load parameters of the workshop where the transformer production equipment is located are obtained. The real-time environmental load parameters include the ambient temperature and humidity fluctuation parameters in the workshop, the weight parameters of the transformer core, and the thermal capacity characteristic parameters of the transformer copper wire. The environmental temperature and humidity fluctuation parameters, the weight parameters, and the heat capacity characteristic parameters are normalized to construct an interference decoupling variable matrix, and the interference decoupling variables are obtained based on the interference decoupling variable matrix.

[0007] Preferably, the step of performing variable decoupling analysis on the transient pumping response data based on the interference decoupling variables to obtain the true pumping efficiency characteristics of the transformer production equipment specifically includes: The interference decoupling variables are input into a preset multivariate decoupling model to identify the interference weight coefficients of the environmental temperature and humidity fluctuation parameters and the heat capacity characteristic parameters on the activity of gas molecules in the vacuum chamber. Based on the transient pumping response data, the pumping rate curve within the pressure drop range is obtained, and the pumping rate curve is adaptively adjusted and fluctuation compensated according to the interference weight coefficient to obtain the target pumping rate curve. The target pumping response data is obtained based on the target pumping rate curve. Based on the target pumping response data, the instantaneous pumping rate of the production equipment within the preset high vacuum threshold range is obtained, and the net pumping rate curve of the production equipment is obtained based on the instantaneous pumping rate. The smoothness and attenuation slope of the net pumping rate curve are extracted, and the smoothness and attenuation slope are used to extract features to obtain a high-frequency feature vector. The high-frequency feature vector is then marked as the true pumping efficiency feature of the production equipment.

[0008] Preferably, the step of collecting voltage drop curve data from the transformer production equipment during the voltage holding test phase and vacuum pump power consumption data during the pumping phase, and performing time-series cross-validation by combining the voltage drop curve data, the vacuum pump power consumption data, and the actual pumping efficiency characteristics to obtain the fault root cause classification result, specifically includes: During the voltage holding test of the transformer, a sequence of micro-pressure difference changes inside the vacuum chamber is collected, and voltage drop curve data corresponding to the voltage holding test stage is generated based on the sequence of micro-pressure difference changes. Extract the vacuum pump motor operating parameters for the pumping stage corresponding to the actual pumping efficiency characteristics. Obtain active power information and reactive power information based on the vacuum pump motor operating parameters. Combine the active power information and the reactive power information to obtain the vacuum pump power consumption data for the pumping stage. Extract the descent slope from the pressure drop curve data, the abnormal peak frequency from the vacuum pump power consumption data, and the attenuation amplitude from the actual pumping efficiency characteristics; A multi-dimensional time-series cross-validation model is established, which has multiple fault logic judgment conditions preset in the model. The descent slope, the abnormal peak frequency and the attenuation amplitude are input into the multi-dimensional time-series cross-validation model for logical judgment to obtain multiple verification results. All the verification results are summarized, and the verification results that cause conflicts are weighted by confidence level. Occasional interference results with confidence levels lower than the preset fault threshold are removed, and the final fault root cause classification result is output.

[0009] Preferably, the step of constructing a virtual sensor model and predicting the emulsification and contamination state of the vacuum pump oil based on the virtual sensor model to obtain the remaining life data of the oil is as follows: Retrieve historical operating data of the vacuum pump in the factory since the last change of vacuum pump oil, and extract multiple discrete time periods in the historical operating data where the pump body exhaust temperature exceeds the critical temperature for oil oxidation resistance. The cumulative high-temperature operating time of the vacuum pump under harsh heat dissipation conditions is obtained by summing the time lengths of multiple discrete time periods. The downward slope in the pressure drop curve data is marked as a state feedback quantity, and the cumulative high-temperature operating time is used as a loss input quantity. A virtual sensor model simulating the deterioration of the physical properties of oil is constructed by combining the state feedback quantity and the loss input quantity. The newly added target high-temperature operating time and the target actual pumping efficiency characteristics of the current batch are collected in real time and input into the virtual sensor model to predict the water mixing ratio and viscosity decrease in the vacuum pump oil. The dynamic health index of the vacuum pump oil is obtained based on the water mixing ratio and viscosity decrease. The dynamic health index is linearly decay-fitted with a preset vacuum pump oil scrapping threshold to obtain the remaining working time of the oil when it reaches the scrapping state, and the remaining lifespan data of the oil is generated based on the remaining working time.

[0010] Preferably, the step of comprehensively evaluating the transformer production equipment by combining the fault root cause classification results and the remaining oil life data to generate an equipment degradation prediction report specifically includes: Extract the fault types and the fault severity values ​​corresponding to each fault type from the fault root cause classification results; Based on the fault type, the corresponding component's failure impact weight coefficient and average repair cost parameter are matched to the preset equipment maintenance database; Identify the numerical difference between the remaining lifespan data and the preset standard equipment maintenance cycle, and obtain the oil deterioration rate based on the numerical difference; A comprehensive degradation assessment model for transformer production equipment is constructed. The failure impact weight coefficient, the average repair cost parameter, and the oil deterioration rate are substituted into the comprehensive degradation assessment model to obtain the total evaluation score of the transformer production equipment. Based on the overall evaluation score, the current degradation stage of the transformer production equipment is determined, and an equipment degradation prediction report is generated based on the current degradation stage.

[0011] Preferably, the steps of acquiring the factory's production cycle time data, obtaining the equipment's natural cooling time window based on the production cycle time data, and generating a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window are as follows: Obtain the future production schedule of the factory's transformer production line, extract the expected start time and standard operation duration of the vacuuming operation of the transformer in the production schedule, and generate continuous production cycle data; The required cooling time for pump body cooling and the standard working hours for changing engine oil are obtained. The maintenance time is then calculated by combining the cooling time and the standard working hours. Scan the task idle period in the production cycle data, extract the idle time length of the task idle period, and determine whether the idle time length exceeds the maintenance time. If it is determined that the idle time length exceeds the maintenance time, mark the corresponding task idle period as a natural cooling time window. Based on the latest maintenance time node in the equipment degradation prediction report, the natural cooling time window and the latest maintenance time node are overlapped and compared on the same time axis to select the best maintenance window that is closest to and no later than the latest maintenance time node. Within the optimal maintenance window, allocate maintenance personnel and required spare parts, generate maintenance work order instructions, and generate a non-stop maintenance scheduling plan based on the maintenance work order instructions.

[0012] Preferably, before the step of allocating maintenance personnel and required spare parts within the optimal maintenance window and generating a maintenance work order instruction, the method further includes: Obtain the list of parts to be replaced from the equipment degradation prediction report and the real-time spare parts inventory data from the factory spare parts warehouse; Match the list of parts to be replaced with the real-time spare parts inventory data, and query the current inventory quantity of each item in the list of parts to be replaced; Determine whether the current inventory quantity meets the maintenance requirements. If it is determined that there are materials that do not meet the maintenance requirements, trigger the emergency procurement process and record the arrival time of the spare parts. The arrival time is compared with the start time of the optimal maintenance window. If the arrival time is later than the start time, the optimal maintenance window is marked as unavailable. In the remaining natural cooling time window, a second filtering is performed according to time order to find the extended maintenance window that is later than the arrival time and earlier than the latest maintenance time.

[0013] In summary, this application includes at least one of the following beneficial technical effects: By acquiring operational monitoring data and environmental load parameters of the transformer production equipment, transient pumping response data during the vacuuming process is extracted, and interference decoupling variables are obtained. Then, based on these interference decoupling variables, variable decoupling analysis is performed on the transient pumping response data to filter out the physical interference of ambient temperature and load heat capacity on the vacuuming rate, thereby obtaining the true pumping efficiency characteristics of the transformer production equipment. Next, pressure drop curve data during the pressure holding test and vacuum pump power consumption data during the pumping phase are collected. These are cross-validated with the true pumping efficiency characteristics over time to accurately identify and obtain fault root cause classification results. Simultaneously, historical operating data of the vacuum pump is acquired, and the cumulative high-temperature operating time is extracted. A virtual sensor model is constructed based on the decreasing trend of the true pumping efficiency characteristics to predict the emulsification and contamination state of the vacuum pump oil and obtain high-precision oil remaining life data. Finally, a comprehensive degradation assessment of the equipment is performed based on the fault root cause classification results and the oil remaining life data, generating an equipment degradation prediction report. Finally, production cycle data from the factory's manufacturing execution system was acquired, and the natural cooling time window of the equipment was accurately extracted. The equipment degradation prediction report was then overlaid, compared, and matched with this time window for scheduling, ultimately generating a zero-downtime maintenance scheduling plan for transformer production equipment. This improved the predictability and accuracy of transformer production equipment maintenance. Attached Figure Description

[0014] Figure 1 This is a flowchart of a predictive maintenance method for transformer production equipment provided in an embodiment of this application. Detailed Implementation

[0015] The following combination Figure 1 This application will be described in further detail, but the embodiments of the present invention are not limited thereto.

[0016] This application discloses a predictive maintenance method for transformer production equipment.

[0017] In this embodiment, a predictive maintenance method for transformer manufacturing equipment includes: S100: Obtain the operation monitoring data of transformer production equipment and environmental load parameters, obtain the transient gas pumping response data during the vacuuming process based on the operation monitoring data, and obtain the disturbance decoupling variables based on the environmental load parameters; S200: Perform variable decoupling analysis on transient pumping response data based on interference decoupling variables to obtain the true pumping efficiency characteristics of transformer production equipment; S300: Collect voltage drop curve data of transformer production equipment during the voltage holding test stage and vacuum pump power consumption data during the pumping stage. Combine the voltage drop curve data, vacuum pump power consumption data and actual pumping efficiency characteristics to perform time-series cross-validation and obtain the fault root cause classification results. S400: Acquire historical operating data of vacuum pump, extract the cumulative high-temperature operating time of vacuum pump, combine the actual pumping efficiency characteristics and the cumulative high-temperature operating time to construct a virtual sensor model, and predict the emulsification and contamination status of vacuum pump oil based on the virtual sensor model to obtain the remaining life data of oil. S500: Combines the root cause classification results of faults with the remaining life data of oil products to conduct a comprehensive evaluation of transformer production equipment and generate an equipment degradation prediction report. S600: Acquires the factory's production cycle data, obtains the equipment's natural cooling time window based on the production cycle data, and generates a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window.

[0018] The steps for acquiring operational monitoring data and environmental load parameters of transformer production equipment, obtaining transient pumping response data during the vacuuming process based on the operational monitoring data, and deriving disturbance decoupling variables based on the environmental load parameters are as follows: Real-time acquisition of multi-dimensional operation monitoring data of transformer production equipment throughout the entire vacuuming operation cycle. The operation monitoring data includes at least the real-time gas pressure value inside the vacuum chamber, the operating frequency of the vacuum pump, and the instantaneous flow parameters of the exhaust port. Establish a standardized time axis corresponding to the entire cycle of vacuuming operations, and map real-time air pressure values, operating frequency, and instantaneous flow parameters onto the standardized time axis to obtain a multi-dimensional parameter time axis; Identify the pressure drop intervals on the multi-dimensional parameter time axis and extract transient pumping response data within the pressure drop intervals; The real-time environmental load parameters of the workshop where the transformer production equipment is located are obtained. The real-time environmental load parameters include the temperature and humidity fluctuation parameters in the workshop, the weight parameters of the transformer core, and the thermal capacity characteristics of the transformer copper wire. The environmental temperature and humidity fluctuation parameters, weight parameters, and heat capacity characteristic parameters are normalized in terms of dimensions to construct an interference decoupling variable matrix, and the interference decoupling variables are obtained based on the interference decoupling variable matrix.

[0019] In application, taking the No. 1 vacuum drying equipment in a transformer factory as an example, the system first collects multi-dimensional operational monitoring data of the equipment in real time during the vacuuming operation cycle. The collected real-time air pressure value of the vacuum chamber is 5 kPa, the vacuum pump operating frequency is 50 Hz, and the instantaneous flow rate parameter of the exhaust port is 200 liters per second. Next, the system establishes a standardized time axis corresponding to the entire vacuuming operation cycle, mapping the real-time air pressure value, operating frequency, and instantaneous flow rate parameter onto the time axis to obtain a multi-dimensional parameter time axis. Then, the system accurately identifies the air pressure drop interval on the multi-dimensional parameter time axis and extracts the transient pumping response data within this air pressure drop interval. At the same time, the system obtains the real-time environmental load parameters of the workshop where the equipment is located, including an ambient temperature of 30 degrees Celsius, an ambient humidity of 60%, a transformer core weight of 2,000 kg, and a transformer copper wire heat capacity characteristic parameter of 300. Afterward, the system performs dimensional normalization processing on the ambient temperature and humidity fluctuation parameters, weight parameters, and heat capacity characteristic parameters to eliminate differences caused by different units. After processing, the system constructs an interference decoupling variable matrix. Finally, the system calculates and obtains the final interference decoupling variables based on the constructed interference decoupling variable matrix.

[0020] The steps for performing variable decoupling analysis on transient pumping response data based on interference decoupling variables to obtain the true pumping efficiency characteristics of transformer production equipment are as follows: The interference decoupling variables are input into a preset multivariate decoupling model to identify the interference weight coefficients of environmental temperature and humidity fluctuation parameters and heat capacity characteristic parameters on the activity of gas molecules in the vacuum chamber. Based on the transient pumping response data, the pumping rate curve within the pressure drop range is obtained. The pumping rate curve is then adaptively adjusted and fluctuation compensation is performed according to the interference weight coefficient to obtain the target pumping rate curve. The target pumping response data is then obtained based on the target pumping rate curve. Based on the target pumping response data, the instantaneous pumping rate of the production equipment within the preset high vacuum threshold range is obtained, and the net pumping rate curve of the production equipment is obtained based on the instantaneous pumping rate. The smoothness and attenuation slope of the net pumping rate curve are extracted, and the smoothness and attenuation slope are used to extract high-frequency feature vectors. These high-frequency feature vectors are then marked as the true pumping efficiency features of the production equipment.

[0021] In application, taking the No. 1 vacuum drying equipment in a transformer factory as an example, the system first inputs the interference decoupling variables obtained in the previous step into a preset multivariate decoupling model. Through model calculation, the system identifies the interference weight coefficient of parameters such as ambient temperature of 30 degrees Celsius and humidity of 60% on the activity of gas molecules in the vacuum chamber as 0.2. Next, based on the extracted transient pumping response data, the system obtains the pumping rate curve within the pressure drop range. Then, using the interference weight coefficient of 0.2, the system adaptively adjusts and compensates for fluctuations in the pumping rate curve to obtain the target pumping rate curve after eliminating interference, and obtains the target pumping response data based on this curve. Afterward, based on the target pumping response data, the system calculates the instantaneous pumping rate of the equipment within the preset high vacuum threshold range of 1000 Pa to 500 Pa as 50 liters per second. Based on these instantaneous pumping rates, the system generates the net pumping rate curve of the equipment. Next, the system extracts the smoothness value of the net pumping rate curve as 90% and the attenuation slope value as 0.5. The system extracts features from the smoothness and attenuation slope, obtaining a high-frequency feature vector composed of these values. Finally, the system labels this high-frequency feature vector as the true pumping efficiency feature of the device.

[0022] The steps for collecting voltage drop curve data from the transformer production equipment during the voltage holding test phase and vacuum pump power consumption data during the pumping phase, and then performing time-series cross-validation based on the voltage drop curve data, vacuum pump power consumption data, and actual pumping efficiency characteristics to obtain the fault root cause classification results are as follows: During the voltage holding test of the transformer, the micro-pressure difference change sequence inside the vacuum chamber is collected, and the voltage drop curve data corresponding to the voltage holding test stage is generated based on the micro-pressure difference change sequence. Extract the vacuum pump motor operating parameters corresponding to the actual pumping efficiency characteristics during the pumping stage. Obtain active power information and reactive power information based on the vacuum pump motor operating parameters. Combine the active power information and reactive power information to obtain the vacuum pump power consumption data during the pumping stage. Extract the descent slope from the pressure drop curve data, the abnormal peak frequency from the vacuum pump power consumption data, and the attenuation amplitude from the true pumping efficiency characteristics. A multi-dimensional time-series cross-validation model is established. The multi-dimensional time-series cross-validation model has multiple logical judgment conditions for faults. The descent slope, abnormal peak frequency and attenuation amplitude are input into the multi-dimensional time-series cross-validation model for logical judgment, and multiple verification results are obtained. Summarize all the verification results, perform confidence weighting on the conflicting verification results, remove occasional interference results with confidence levels below the preset fault threshold, and output the final fault root cause classification result.

[0023] In application, taking the No. 1 vacuum drying equipment in a transformer factory as an example, during the pressure holding test phase, the system collects a sequence of micro-pressure difference changes inside the vacuum chamber, such as the process of pressure dropping from 1 kPa to 900 kPa, and generates pressure drop curve data corresponding to the pressure holding test phase based on this sequence. Simultaneously, during the pumping phase, the system extracts the corresponding vacuum pump motor operating parameters, calculating the active power information as 10 kW and the reactive power information as 2 kW. Combining the active and reactive power information, the system obtains the vacuum pump power consumption data during the pumping phase. Next, the system extracts the decreasing slope value of 0.1 from the pressure drop curve data, the abnormal peak frequency of five times per day from the vacuum pump power consumption data, and the attenuation amplitude of 10% from the actual pumping efficiency characteristics. Then, the system establishes a multi-dimensional time-series cross-validation model, which presets various logical judgment conditions for faults. The system inputs the decreasing slope of 0.1, the abnormal peak frequency of five times per day, and the 10% attenuation amplitude into the model for logical judgment, obtaining multiple verification results. Finally, the system summarizes all verification results and performs a confidence-weighted judgment on conflicting results. The system removes occasional interference results with a confidence level below 50% and outputs the final root cause classification result, which identifies the equipment seal as aging.

[0024] The steps for constructing a virtual sensor model and predicting the emulsification and contamination state of the vacuum pump oil based on the virtual sensor model to obtain the remaining oil life data are as follows: Retrieve historical operating data of the vacuum pumps in the factory since the last change of vacuum pump oil, and extract multiple discrete time periods in the historical operating data where the pump body exhaust temperature exceeded the critical temperature for oil oxidation resistance. The cumulative high-temperature operating time of the vacuum pump under harsh heat dissipation conditions is obtained by summing the time lengths of multiple discrete time periods. The downward slope in the pressure drop curve data is marked as the state feedback quantity, and the cumulative high-temperature operating time is used as the loss input quantity. A virtual sensor model simulating the deterioration of the physical properties of oil is constructed by combining the state feedback quantity and the loss input quantity. Real-time acquisition of newly added target high-temperature operating time and the actual pumping efficiency characteristics of the current batch of targets is input into the virtual sensor model to predict the water mixing ratio and viscosity reduction in the vacuum pump oil. Based on the water mixing ratio and viscosity reduction, the dynamic health index of the vacuum pump oil is obtained. The dynamic health index is linearly decayed and fitted to the preset vacuum pump oil scrapping threshold to obtain the remaining working time of the oil when it reaches the scrapping state, and the remaining life data of the oil is generated based on the remaining working time.

[0025] In application, taking the No. 1 vacuum drying equipment in a transformer factory as an example, the system first retrieves historical operating data of the vacuum pumps since the last vacuum pump oil replacement. The system extracts multiple discrete time periods from the data where the exhaust temperature exceeds the 80°C critical oxidation level of the oil. The system sums up the duration of all these discrete time periods, obtaining a cumulative high-temperature operating time of 50 hours under harsh heat dissipation conditions. Next, the system marks the downward slope value in the pressure drop curve data as a state feedback quantity, and uses the 50 hours of cumulative high-temperature operating time as a loss input quantity. Combining the state feedback quantity and the loss input quantity, the system constructs a virtual sensor model simulating the deterioration of the oil's physical properties. Then, the system collects the newly added two-hour target high-temperature operating time and the target actual pumping efficiency characteristics of the current batch in real time. The system inputs these into the virtual sensor model, predicting that the water content inside the vacuum pump oil is 2%, and the viscosity decrease is 10%. Based on the water content and viscosity decrease, the system calculates the dynamic health index of the vacuum pump oil to be 75. Finally, the system linearly decays the health index of 75 with the preset scrap threshold of 30, calculates the remaining working time required for the oil to reach the scrap state as 200 hours, and generates the remaining lifespan data of the oil.

[0026] The steps for comprehensively evaluating transformer production equipment and generating an equipment degradation prediction report by combining the root cause classification results and oil remaining life data are as follows: Extract the fault types and the corresponding fault severity values ​​from the root cause classification results; Based on the fault type, the failure impact weight coefficient and average repair cost parameter of the corresponding component in the preset equipment maintenance database are matched; Identify the numerical difference between the remaining lifespan data and the preset standard equipment maintenance cycle, and determine the rate of oil deterioration based on the numerical difference; A comprehensive degradation assessment model for transformer production equipment was constructed. The failure impact weight coefficient, average repair cost parameter, and oil deterioration rate were substituted into the comprehensive degradation assessment model to obtain the total evaluation score of the transformer production equipment. Based on the overall evaluation score, the current degradation stage of the transformer production equipment is determined, and a degradation prediction report is generated based on the current degradation stage.

[0027] In this application, we take the No. 1 vacuum drying equipment in a transformer factory as an example. First, the system extracts the root cause classification results from the previous step, identifying the fault type as sealing ring aging, with a severity value of eight. Next, based on the sealing ring aging fault type, the system searches and matches in a preset equipment maintenance database, finding a failure impact weight coefficient of 0.6 and an average repair cost parameter of 500 yuan for the corresponding component. Then, the system extracts 200 hours from the remaining oil life data and identifies a preset standard equipment maintenance cycle of 1000 hours. The system calculates the difference between the remaining life and the standard maintenance cycle to be 800 hours, and based on this significant difference, calculates the current oil deterioration rate to be a decrease of two units per day. Finally, the system constructs a comprehensive degradation assessment model for the transformer production equipment. The system incorporates the 0.6 failure impact weight coefficient, the 500 yuan average repair cost parameter, and the 2-unit-per-day oil deterioration rate into the comprehensive degradation assessment model for weighted calculation, ultimately yielding a total evaluation score of 75 points for the transformer production equipment. Finally, based on the total evaluation score of 75, the system classifies the current degradation stage of the equipment as the mid-term wear and tear degradation stage, and generates a detailed equipment degradation prediction report based on this current degradation stage.

[0028] The steps for obtaining factory production cycle data, determining the natural cooling time window of the equipment based on the production cycle data, and generating a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window are as follows: Obtain the future production schedule of the factory's transformer production line, extract the expected start time and standard operation duration of the vacuuming operation of the transformer in the production schedule, and generate continuous production cycle data; Obtain the cooling time required for pump body cooling and the standard working hours for changing engine oil. Combine the cooling time and standard working hours to obtain the maintenance time. Scan the task idle period in the production cycle data, extract the idle time length of the task idle period, and determine whether the idle time length exceeds the maintenance time. If it is determined that the idle time length exceeds the maintenance time, mark the corresponding task idle period as a natural cooling time window. Based on the latest maintenance time node in the equipment degradation prediction report, the natural cooling time window and the latest maintenance time node are compared and overlapped on the same time axis to select the best maintenance window that is closest to and no later than the latest maintenance time node. Within the optimal maintenance window, allocate maintenance personnel and required spare parts, generate maintenance work order instructions, and generate a non-stop maintenance scheduling plan based on the maintenance work order instructions.

[0029] In practice, taking the No. 1 vacuum drying equipment in a transformer factory as an example, the system first obtains the production schedule for the factory's transformer production line for the coming week. From the schedule, the system extracts the estimated start time for vacuuming each transformer as 8:00 AM, with a standard operating time of four hours. This information is then linked together to generate continuous production cycle data. Next, the system obtains the required cooling time for the pump body as two hours and the standard time for changing the oil as one hour. The system adds the cooling time and standard time to calculate the total maintenance time as three hours. Then, the system scans the task gaps in the production cycle data and extracts the gap from 12:00 PM to 4:00 PM as four hours. The system determines that this four-hour gap exceeds the three-hour maintenance time and marks it as a natural cooling time window. Finally, the system checks the equipment degradation prediction report and finds the latest maintenance time to be this Friday afternoon. The system compares multiple generated natural cooling time windows and the latest maintenance time nodes on the same timeline, selecting the optimal maintenance window that is closest to and no later than Friday afternoon, which is Friday noon. Finally, within this optimal maintenance window, the system assigns maintenance workers and required parts, generates maintenance work orders, and outputs a non-stop maintenance scheduling plan.

[0030] Before the steps of assigning maintenance personnel and required spare parts within the optimal maintenance window and generating maintenance work order instructions, the following are also included: Obtain the list of parts to be replaced from the equipment degradation prediction report and the real-time spare parts inventory data from the factory spare parts warehouse; Match the list of parts to be replaced with real-time spare parts inventory data to query the current inventory quantity of each item in the list of parts to be replaced. Determine whether the current inventory quantity meets the needs of this maintenance. If it is determined that there are materials that do not meet the needs of this maintenance, trigger the emergency procurement process and record the arrival time of the spare parts. The arrival time is compared with the start time of the optimal maintenance window. If the arrival time is later than the start time, the optimal maintenance window is marked as unavailable. Within the remaining natural cooling time window, a secondary filter is performed according to time order to find the extended maintenance window that is later than the delivery time and earlier than the latest maintenance time.

[0031] In practice, taking the No. 1 vacuum drying equipment in a transformer factory as an example, before generating a maintenance work order, the system first retrieves the list of parts to be replaced from the equipment degradation prediction report, which indicates the need to replace five sealing rings. Simultaneously, it retrieves the real-time spare parts inventory data from the factory's spare parts warehouse. Next, the system compares and matches the list of parts to be replaced with the real-time inventory data, finding that the current inventory of sealing rings on the list is only two. Then, the system determines that the current inventory of two is insufficient to meet the maintenance requirement of five. Because the materials are insufficient, the system immediately triggers an emergency procurement process, places an order with the supplier, and records the estimated arrival time of this batch of new spare parts as 10:00 AM this Saturday. Afterwards, the system strictly compares the arrival time of 10:00 AM Saturday with the start time of the optimal maintenance window calculated in the previous step, which is 12:00 PM Friday. The system finds that 10:00 AM Saturday is later than 12:00 PM Friday, therefore determining that the spare parts cannot arrive in time and directly marking the optimal maintenance window of 12:00 PM Friday as unavailable. Finally, the system performs a second round of filtering in chronological order among all remaining natural cooling time windows. The system eventually found a maintenance window that would be extended if the delivery time was later than 10:00 AM on Saturday but earlier than the latest maintenance time on Monday of the following week, for example, scheduled for 2:00 PM on Sunday.

[0032] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A predictive maintenance method for transformer manufacturing equipment, characterized in that, include: Obtain the operation monitoring data and environmental load parameters of the transformer production equipment, obtain the transient gas extraction response data during the vacuuming process based on the operation monitoring data, and obtain the disturbance decoupling variables according to the environmental load parameters; Based on the interference decoupling variables, variable decoupling analysis is performed on the transient pumping response data to obtain the true pumping efficiency characteristics of the transformer production equipment. Data on voltage drop curves during the voltage holding test of transformer production equipment and data on vacuum pump power consumption during the pumping test are collected. Time-series cross-validation is performed by combining the voltage drop curve data, the vacuum pump power consumption data, and the actual pumping efficiency characteristics to obtain the fault root cause classification results. Historical operating data of the vacuum pump is obtained, the cumulative high-temperature operating time of the vacuum pump is extracted, and a virtual sensor model is constructed by combining the actual pumping efficiency characteristics and the cumulative high-temperature operating time. Based on the virtual sensor model, the emulsification and contamination state of the vacuum pump oil is predicted to obtain the remaining life data of the oil. A comprehensive assessment of transformer production equipment is conducted by combining the fault root cause classification results and the remaining oil life data, and an equipment degradation prediction report is generated. Obtain the factory's production cycle time data, determine the equipment's natural cooling time window based on the production cycle time data, and generate a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window.

2. The predictive maintenance method for transformer production equipment according to claim 1, characterized in that, The steps of acquiring operation monitoring data and environmental load parameters of transformer production equipment, obtaining transient pumping response data during the vacuuming process based on the operation monitoring data, and obtaining disturbance decoupling variables based on the environmental load parameters are as follows: Real-time acquisition of multi-dimensional operation monitoring data of transformer production equipment throughout the entire vacuuming operation cycle. The operation monitoring data includes at least the real-time air pressure value inside the vacuum chamber, the operating frequency of the vacuum pump, and the instantaneous flow parameters of the exhaust port. A standardized time axis corresponding to the entire cycle of vacuuming operation is established, and the real-time air pressure value, the operating frequency, and the instantaneous flow rate parameter are mapped onto the standardized time axis to obtain a multi-dimensional parameter time axis; Identify the pressure drop interval on the time axis of the multidimensional parameters, and extract the transient pumping response data in the pressure drop interval; The real-time environmental load parameters of the workshop where the transformer production equipment is located are obtained. The real-time environmental load parameters include the ambient temperature and humidity fluctuation parameters in the workshop, the weight parameters of the transformer core, and the thermal capacity characteristic parameters of the transformer copper wire. The environmental temperature and humidity fluctuation parameters, the weight parameters, and the heat capacity characteristic parameters are normalized to construct an interference decoupling variable matrix, and the interference decoupling variables are obtained based on the interference decoupling variable matrix.

3. The predictive maintenance method for transformer production equipment according to claim 2, characterized in that, The steps for performing variable decoupling analysis on the transient pumping response data based on the aforementioned interference decoupling variables to obtain the true pumping efficiency characteristics of the transformer production equipment are as follows: The interference decoupling variables are input into a preset multivariate decoupling model to identify the interference weight coefficients of the environmental temperature and humidity fluctuation parameters and the heat capacity characteristic parameters on the activity of gas molecules in the vacuum chamber. Based on the transient pumping response data, the pumping rate curve within the pressure drop range is obtained, and the pumping rate curve is adaptively adjusted and fluctuation compensated according to the interference weight coefficient to obtain the target pumping rate curve. The target pumping response data is obtained based on the target pumping rate curve. Based on the target pumping response data, the instantaneous pumping rate of the production equipment within the preset high vacuum threshold range is obtained, and the net pumping rate curve of the production equipment is obtained based on the instantaneous pumping rate. The smoothness and attenuation slope of the net pumping rate curve are extracted, and the smoothness and attenuation slope are used to extract features to obtain a high-frequency feature vector. The high-frequency feature vector is then marked as the true pumping efficiency feature of the production equipment.

4. The predictive maintenance method for transformer production equipment according to claim 3, characterized in that, The steps for collecting voltage drop curve data from the transformer production equipment during the voltage holding test phase and vacuum pump power consumption data during the pumping phase, and then performing time-series cross-validation based on the voltage drop curve data, the vacuum pump power consumption data, and the actual pumping efficiency characteristics to obtain the fault root cause classification results are as follows: During the voltage holding test of the transformer, a sequence of micro-pressure difference changes inside the vacuum chamber is collected, and voltage drop curve data corresponding to the voltage holding test stage is generated based on the sequence of micro-pressure difference changes. Extract the vacuum pump motor operating parameters for the pumping stage corresponding to the actual pumping efficiency characteristics. Obtain active power information and reactive power information based on the vacuum pump motor operating parameters. Combine the active power information and the reactive power information to obtain the vacuum pump power consumption data for the pumping stage. Extract the descent slope from the pressure drop curve data, the abnormal peak frequency from the vacuum pump power consumption data, and the attenuation amplitude from the actual pumping efficiency characteristics; A multi-dimensional time-series cross-validation model is established, which has multiple fault logic judgment conditions preset in the model. The descent slope, the abnormal peak frequency and the attenuation amplitude are input into the multi-dimensional time-series cross-validation model for logical judgment to obtain multiple verification results. All the verification results are summarized, and the verification results that cause conflicts are weighted by confidence level. Occasional interference results with confidence levels lower than the preset fault threshold are removed, and the final fault root cause classification result is output.

5. The predictive maintenance method for transformer production equipment according to claim 4, characterized in that, The steps of constructing a virtual sensor model and predicting the emulsification and contamination state of the vacuum pump oil based on the virtual sensor model to obtain the remaining oil life data are as follows: Retrieve historical operating data of the vacuum pump in the factory since the last change of vacuum pump oil, and extract multiple discrete time periods in the historical operating data where the pump body exhaust temperature exceeds the critical temperature for oil oxidation resistance. The cumulative high-temperature operating time of the vacuum pump under harsh heat dissipation conditions is obtained by summing the time lengths of multiple discrete time periods. The downward slope in the pressure drop curve data is marked as a state feedback quantity, and the cumulative high-temperature operating time is used as a loss input quantity. A virtual sensor model simulating the deterioration of the physical properties of oil is constructed by combining the state feedback quantity and the loss input quantity. The newly added target high-temperature operating time and the target actual pumping efficiency characteristics of the current batch are collected in real time and input into the virtual sensor model to predict the water mixing ratio and viscosity decrease in the vacuum pump oil. The dynamic health index of the vacuum pump oil is obtained based on the water mixing ratio and viscosity decrease. The dynamic health index is linearly decay-fitted with a preset vacuum pump oil scrapping threshold to obtain the remaining working time of the oil when it reaches the scrapping state, and the remaining lifespan data of the oil is generated based on the remaining working time.

6. The predictive maintenance method for transformer production equipment according to claim 5, characterized in that, The steps for comprehensively evaluating transformer production equipment and generating an equipment degradation prediction report by combining the fault root cause classification results and the remaining oil life data are as follows: Extract the fault types and the fault severity values ​​corresponding to each fault type from the fault root cause classification results; Based on the fault type, the corresponding component's failure impact weight coefficient and average repair cost parameter are matched to the preset equipment maintenance database; Identify the numerical difference between the remaining lifespan data and the preset standard equipment maintenance cycle, and obtain the oil deterioration rate based on the numerical difference; A comprehensive degradation assessment model for transformer production equipment is constructed. The failure impact weight coefficient, the average repair cost parameter, and the oil deterioration rate are substituted into the comprehensive degradation assessment model to obtain the total evaluation score of the transformer production equipment. Based on the overall evaluation score, the current degradation stage of the transformer production equipment is determined, and an equipment degradation prediction report is generated based on the current degradation stage.

7. A predictive maintenance method for transformer production equipment according to claim 6, characterized in that, The steps for obtaining factory production cycle data, determining the natural cooling time window of the equipment based on the production cycle data, and generating a non-stop maintenance scheduling plan for the production equipment by combining the equipment degradation prediction report and the natural cooling time window are as follows: Obtain the future production schedule of the factory's transformer production line, extract the expected start time and standard operation duration of the vacuuming operation of the transformer in the production schedule, and generate continuous production cycle data; The required cooling time for pump body cooling and the standard working hours for changing engine oil are obtained. The maintenance time is then calculated by combining the cooling time and the standard working hours. Scan the task idle period in the production cycle data, extract the idle time length of the task idle period, and determine whether the idle time length exceeds the maintenance time. If it is determined that the idle time length exceeds the maintenance time, mark the corresponding task idle period as a natural cooling time window. Based on the latest maintenance time node in the equipment degradation prediction report, the natural cooling time window and the latest maintenance time node are overlapped and compared on the same time axis to select the best maintenance window that is closest to and no later than the latest maintenance time node. Within the optimal maintenance window, allocate maintenance personnel and required spare parts, generate maintenance work order instructions, and generate a non-stop maintenance scheduling plan based on the maintenance work order instructions.

8. The predictive maintenance method for transformer production equipment according to claim 7, characterized in that, Before the steps of assigning maintenance personnel and required spare parts within the optimal maintenance window and generating maintenance work order instructions, the following are also included: Obtain the list of parts to be replaced from the equipment degradation prediction report and the real-time spare parts inventory data from the factory spare parts warehouse; Match the list of parts to be replaced with the real-time spare parts inventory data, and query the current inventory quantity of each item in the list of parts to be replaced; Determine whether the current inventory quantity meets the maintenance requirements. If it is determined that there are materials that do not meet the maintenance requirements, trigger the emergency procurement process and record the arrival time of the spare parts. The arrival time is compared with the start time of the optimal maintenance window. If the arrival time is later than the start time, the optimal maintenance window is marked as unavailable. In the remaining natural cooling time window, a second filtering is performed according to time order to find the extended maintenance window that is later than the arrival time and earlier than the latest maintenance time.

Images