A method for predicting failures in oilseed pressing production lines

Abstract

This invention relates to the field of equipment testing technology, and in particular to a fault prediction method for oilseed pressing production lines. The invention first collects multi-node operating status data and corresponding fault types within a preset statistical time period, constructs and trains a fault prediction model, and determines extraction rules and establishes logical operation links based on the linkage logic of production line components as fault-related operating parameter characteristics. The current operating data is entered to obtain the initial fault; reference links are determined based on the links, actual operating rules are obtained, invalid faults are eliminated after comparison, the fault occurrence probability is adjusted, and maintenance priorities are assigned. The method of this invention simplifies fault results, avoids subjective biases in operation and maintenance, standardizes maintenance sequences, can predict associated hidden dangers, distinguish similar anomalies, is computationally efficient, adapts to continuous production lines, saves operation and maintenance costs, reduces equipment wear, improves operation and maintenance standardization and fault identification accuracy, and ensures stable production line operation and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of equipment testing technology, and in particular to a fault prediction method for oilseed pressing production lines. Background Technology

[0002] With the rapid development of industrial automation, electronic control components are widely used in automated production lines, significantly improving production efficiency. However, automated production line equipment is highly integrated and complex in structure, making it prone to various malfunctions during long-term continuous operation. This is especially true for oilseed pressing production lines, which have strong production continuity and significant downtime losses. Component failures not only lead to oil deterioration and quality decline but also pose safety hazards under high temperature and pressure. Therefore, effective fault prediction and operation and maintenance management are crucial for ensuring stable and efficient production in oilseed pressing lines.

[0003] Traditional automated production line fault diagnosis relies on manual, component-by-component inspection, which is not only inefficient but also results in incomplete fault identification. Currently, the industry commonly utilizes real-time operational status data collected from the production line to build analytical models, which then determine the probability of component failure within a predicted timeframe. For example, the technical solution disclosed in Chinese patent application CN119538101A uses preset standard operational status data to compare and determine if there are any anomalies in the actual collected operational status data. Once anomalies are confirmed, the standard data is used to precisely pinpoint the corresponding abnormal data content. Then, based on preset scoring rules, various preset fault types are scored one by one to obtain the corresponding scores. The actual probability of each fault is calculated based on the scores, and finally, a complete production line fault diagnosis conclusion is generated based on the probability results. Maintenance personnel then repair or replace components according to the fault diagnosis conclusion. This results-oriented approach enables real-time prediction and comprehensive screening of potential faults in oilseed pressing production lines, allowing for hazard identification before production failures officially occur, effectively reducing economic losses caused by production downtime.

[0004] Due to the strong coupling and interconnectedness among components in an oilseed pressing production line, the fault propagation effect causes different types of faults to exhibit highly overlapping parameter characteristics in operational status data. When using the aforementioned fault prediction methods for analysis, the fault type set output by the analysis model will simultaneously include both actual valid faults and invalid false alarms that have not yet occurred, with similar probabilities of occurrence for both. This necessitates manual screening and judgment by maintenance personnel based on their industry experience. Furthermore, for inexperienced maintenance personnel, the similar fault probabilities further increase the difficulty of judgment, ultimately leading to insufficient accuracy in fault type identification. Summary of the Invention

[0005] This invention provides a fault prediction method for oilseed pressing production lines to solve the problem that the operating parameter characteristics corresponding to different fault types are highly similar, resulting in low accuracy of fault type identification.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution: A method for predicting failures in an oilseed pressing production line includes the following steps: S10: Collect the operating status data and corresponding fault types of multiple time nodes within a preset statistical time period; S20: Construct and train a fault prediction model using operational status data and fault types. The fault prediction model is used to output prediction results based on operational status data collected at the current time. The prediction results include at least one fault type and its corresponding probability of occurrence. Associate operational parameter features with fault types based on the linkage working logic of each component on the oilseed pressing production line. Determine the extraction rules for operational parameter features, streamline the data processing flow for deriving extraction rules from operational status data, and establish a logical operation link between operational status data and feature extraction rules. S30: Input the currently collected operating status data into the fault prediction model to obtain the current prediction result, and take the fault type in the current prediction result as the initial fault type; obtain the operating parameter features associated with the initial fault type as the reference features, obtain the extraction rules corresponding to the reference features as the reference rules, and determine the reference link of the operating status data pointing to the reference features within the preset prediction time according to the logical operation link; input the operating status data within the preset prediction time corresponding to the initial fault type into the reference link to obtain each actual operating rule; if any operating parameter feature extracted from the actual operating rule is not included in the coverage of the reference features, then remove the initial fault type corresponding to the reference link and determine the final fault type; S40: Adjust the probability of occurrence in the prediction results corresponding to each final fault type based on the number of paths in the reference link corresponding to the final fault type, the hierarchical depth of each path in the reference link, and the other final fault types mapped by the reference rule on the reference link. S50: Adjust the maintenance priority of the final fault type according to the probability of occurrence.

[0007] The principle and beneficial effects of the basic solution are as follows: the present invention first collects operating state data at multiple time nodes and corresponding fault types within a preset statistical period, uses these data to construct and train a fault prediction model, simultaneously associates operating parameter characteristics with fault types according to the linked working logic of each component on the oil pressing production line, determines the extraction rules for operating parameter characteristics, sorts out the data processing flow, and builds a logical operation link between the operating state data and the feature extraction rules. Then, the currently collected operating state data is input into the fault prediction model to obtain initial fault types, the reference features associated with the initial fault types and the corresponding reference rules are obtained, a reference link within the preset prediction duration is determined according to the logical operation link, and the operating state data of the corresponding period is input into the reference link to obtain each actual operating rule. By comparing the coverage of the operating parameter features extracted by the actual operating rules with that of the reference features, non-conforming initial fault types are eliminated, the fault judgment range is narrowed, and the interference of invalid faults is avoided. Then, the present invention adjusts the occurrence probability of each final fault type according to the number of paths, the path hierarchy depth of the reference link corresponding to the final fault types, and other final fault types associated and mapped by the reference rules on the reference link, and further adjusts the maintenance priority according to the occurrence probability. This method can clearly reflect the internal correlation between fault types, enable operation and maintenance personnel to prioritize the handling of key fault types that have greater impact on the production line and stronger correlation during maintenance, effectively eliminate deviations caused by subjective judgment of operation and maintenance personnel, provide a more scientific basis for maintenance sequence, greatly shorten the maintenance time, and ensure that fault disposal is more targeted.

[0008] The present invention uses the predicted fault type as an anchor point to select operating state data within the prediction time, and then performs fault-related analysis relying on the established logical operation link and the fixed parameter extraction and comparison logic. Compared with processing real-time operating state data, it shortens the overall operation process, eliminates many complicated deduction links, can smoothly complete the comparison and screening of various parameter characteristics and the adjustment of fault-related data, effectively reduces the time consumed by the overall analysis and processing, and can quickly provide reliable judgment results and maintenance-related reference content after the production line gives a fault prompt.

[0009] Meanwhile, its built logical operation link can adapt to the equipment operating status under different production conditions of the oilseed pressing production line. For example, when the oilseed pressing production line is running under different batches of oilseed processing and different ambient temperatures, the heat dissipation system serving the pretreatment process conveyor motor, the steaming and roasting process heating temperature control component, the pressing process hydraulic drive component, and the entire production line electrical control power module is prone to different types of faults such as fan bearing jamming, heat dissipation fin dust accumulation and blockage, and micro-leakage in cooling pipes. Although these faults differ in location, fault mechanism, and evolution rate, they all cause the corresponding process-related components to exhibit highly similar operating parameter characteristics such as abnormal temperature rise, load current fluctuation, and system energy efficiency reduction through heat conduction and load coupling effects. Conventional single parameter threshold judgment methods are difficult to effectively distinguish them. This invention relies on the multi-layer parameter linkage logic of the link and the process and component time sequence association method to capture the essential differences of different faults in parameter evolution time sequence and cross-process transmission path, more accurately distinguish these similar faults, effectively avoid misjudgment, further improve the reliability of fault identification, and provide more solid support for maintenance work.

[0010] This invention utilizes comparative features combined with parameter features obtained from actual operating rules to perform cross-verification. This allows for the elimination of fault categories that do not match the actual operating conditions of the equipment, and the identification of real operational anomalies. For example, in the past, during the operation and maintenance of oilseed pressing production lines, errors in fault prediction often led to the misclassification of normally fluctuating operating parameters as faults, resulting in ineffective maintenance work. This invention reduces useless maintenance work caused by erroneous judgments through the above-mentioned method, saving various maintenance manpower and material resources. It also prevents improper maintenance operations from adversely affecting pressing equipment, conveying components, etc., thus maintaining the equipment in good working condition for a long time, extending the equipment's service life, and reducing the overall cost of equipment replacement and repair.

[0011] This invention utilizes a constructed logical operation chain to clarify the inherent relationships between various faults. Based on this, it rationally adjusts the probability of fault occurrence and prioritizes maintenance, adapting to the industry characteristics of continuous production in oilseed pressing lines. For example, when a fault in the pressing roller of an oilseed pressing line is related to a fault in the transmission component, this invention can identify this relationship and guide staff to prioritize handling the pressing roller fault, which has a greater impact on the production line. This reduces scheduling deviations caused by subjective human intervention and shortens the overall maintenance time. Simultaneously, it can also predict in advance the cascading operational problems that a single anomaly might trigger, such as those affecting the entire production line's electrical control power modules and the core drives of each process. When the heat dissipation system of the components malfunctions under high load, it can cause abnormal temperature rise in the power module, which in turn can lead to overheating of the main pressing motor in the pressing process and overload of the heating and stirring motor in the steaming and frying process. This is usually manifested as a surge in motor stator current and excessive winding temperature. At the same time, the oil temperature of the hydraulic drive components rises, causing oil pressure fluctuations, which can further cause blockages in the raw material conveying process in the pretreatment process and the finished oil conveying process in the filtration process. This invention can identify the above-mentioned potential hazards in advance based on the component linkage logic, and take corresponding preventive measures to avoid the situation where multiple failures occur at the same time, disrupting the production rhythm and causing production line shutdowns. It can also reduce problems such as oil deterioration and quality decline during shutdowns, thereby stabilizing overall production benefits and product quality.

[0012] This invention follows logical rules that conform to operational procedures to complete a series of tasks, including fault diagnosis and priority allocation. It mitigates the differences in judgment caused by variations in human experience. For example, in the past, maintenance personnel with different experience may have differing priorities for the same fault, easily delaying the handling of critical faults. This invention allows different personnel to follow a unified standard when performing maintenance work, promoting standardized and uniform on-site equipment maintenance and comprehensively improving the standardization of the overall maintenance work of oilseed pressing production lines. Furthermore, this method can be fully implemented using only daily collected operational status data, without the need for additional data collection facilities. It is particularly suitable for the actual needs of small and medium-sized oilseed pressing production lines, which often have limited maintenance budgets. This invention simplifies on-site maintenance processes, reasonably controls maintenance-related investments, and lowers the barrier to entry for production line maintenance while ensuring the accuracy of fault prediction, making it easily applicable to oilseed pressing production lines of all sizes.

[0013] In summary, this invention collects operational status data and fault types to construct and train a fault prediction model, establishes a logical operation link, verifies and filters to narrow down the fault range, adjusts the probability of fault occurrence and maintenance priority, can quickly and accurately identify faults, avoid misjudgments and ineffective operation and maintenance, sort out fault correlations, standardize operation and maintenance processes, save costs, predict related hidden dangers, adapt to various oilseed pressing production lines, and ensure stable operation and production efficiency of the production line.

[0014] Furthermore, the operational status data comprises real-time monitoring parameters collected from the corresponding operating components of each of the pretreatment, steaming and roasting, pressing, and filtration and refining processes in the oilseed pressing production line. The pretreatment process corresponds to the conveying and cleaning components; the steaming and roasting process to the heating and stirring components; the pressing process to the pressing and hydraulic drive components; and the filtration and refining process to the filtering, material conveying, and heat exchange components. The data collection time is synchronized with the standard production cycle time of each process, and each monitoring parameter is bound to a corresponding process execution node timestamp. Simultaneously, production data is collected... The system uses real-time supply and demand matching data to align the operational status data of components across different processes, based on the supply and demand relationship within the production line and the collaborative working logic between various components. It extracts basic operational parameter features from the operational status data, including electric drive, temperature and vibration, fluid dynamics, and operating status. Derivative combined features are then obtained from these basic operational parameter features, including energy efficiency ratio and equipment health index. The operational status data is processed and transformed using preset extraction rules to obtain operational parameter features that can be used for fault diagnosis. These extraction rules include time-series rules, numerical rules, frequency domain rules, and correlation rules.

[0015] This invention establishes a one-to-one correspondence between processes and operating components based on the differences in technological functions of pretreatment, steaming, pressing, filtering, and refining. It specifically collects operational status data of dedicated components for each process, ensuring accurate correlation between fault characteristics and their location from the data source. The data collection time is synchronized with the standard production cycle of each process, and each monitoring parameter is bound to a corresponding process execution node timestamp, preserving the temporal information of fault occurrence, propagation, and evolution as completely as possible. Simultaneously, it collects real-time supply and demand matching data from the production line. Based on the supply and demand relationship of the production line and the linkage logic between components, it aligns the cross-process temporal sequence of operational status data for components in different processes, avoiding deviations in operational parameter benchmarks caused by dynamic adjustments to the production line load. Furthermore, this invention divides operational parameter features into basic features and derived combined features, and uses four types of differentiated extraction rules—time-series, numerical, frequency domain, and correlation—for data processing. This not only covers all dimensions of operational status data for oilseed pressing equipment, improving the data basis for fault judgment, but also allows for the selection of appropriate extraction methods based on the changing patterns of different data, enhancing the scenario adaptability of feature extraction. By relying on frequency domain rules, background data interference generated during steady-state equipment operation can be effectively reduced. Compared with analysis models that output the probability of fault occurrence, this invention can more accurately extract latent loss characteristics caused by vibration, thus overcoming the shortcomings of conventional time-domain data monitoring in identifying early, slow-deterioration faults. In complex operating conditions such as multi-grade oilseed blending and pressing, and frequent fluctuations in production load, this invention can quickly distinguish between obvious faults such as equipment malfunctions and latent faults such as long-term energy efficiency degradation by relying on a hierarchical feature system, meeting the initial judgment needs for different fault levels and different stages of evolution.

[0016] Furthermore, before determining the reference link in step S30, corresponding observation periods are configured for different feature extraction rules. The real-time operating load of the production line is obtained by collecting the average current data of the main pressing motor. The preset observation period is dynamically adjusted according to the real-time operating load to obtain the preset prediction duration.

[0017] This invention relies on the operating current of the main pressing motor to determine the real-time operating load of the production line. It dynamically adjusts the characteristic observation period and determines the prediction duration according to the load magnitude, effectively solving the data imbalance problem caused by fixed statistical durations. It avoids collecting redundant and invalid data during light-load operation and incomplete collection of key operating data during heavy-load, high-intensity pressing operations. The equipment operating load directly changes the temperature rise and mechanical deformation rate of components; the matched duration adjustment ensures that the collected data closely matches the actual wear and tear. In intermittent pressing production with frequent fluctuations in feed volume, this invention can flexibly switch the data statistical duration, ensuring that the data basis for fault prediction always aligns with the actual production conditions on site.

[0018] Furthermore, in step S40, during the adjustment of the fault occurrence probability, the occurrence probability corresponding to the fault type in the prediction result is obtained as the initial occurrence probability. The initial occurrence probability is then multiplied and corrected sequentially using the evidence abundance coefficient, the intersection feature correction coefficient, and the hierarchy depth attenuation factor to obtain the preliminary corrected probability. Then, the preliminary corrected probabilities of all initial fault types are normalized to obtain the final occurrence probability of each initial fault type. The evidence abundance coefficient is determined based on the number of reference paths corresponding to the fault. The intersection feature correction coefficient is determined based on the number of shared comparison rules, the topological position of the reference link, and the derivation hierarchy of the operating parameter features within the reference link. The hierarchy depth attenuation factor is calculated based on the average topological hierarchy depth of the reference link corresponding to the fault type.

[0019] In this invention, the evidence abundance coefficient is set based on the number of reference paths to measure the completeness of supporting information for fault determination; the intersection feature correction coefficient is determined by combining the number of shared rules, link positions, and feature derivation levels to offset the judgment interference caused by shared features of multiple types of faults; the level depth attenuation factor is determined according to the average topological depth of the links to reduce the judgment uncertainty caused by multi-layer indirect derivation. This invention performs probability multiplication correction sequentially and then normalization processing to calibrate probability values ​​from three dimensions: evidence volume, feature interference, and inference credibility. In production where multiple potential faults such as sealing leakage, oil pressure deviation, and shell deformation occur simultaneously in the pressing chamber, and various faults share pressure monitoring features, this method can divide the fault credibility basis according to the amount of evidence, avoid probabilistic misjudgment caused by homogeneous features, and weaken the reference weight of long-term inference results, ultimately making the probability distribution of various coexisting faults conform to the actual hidden danger state of the equipment.

[0020] Furthermore, when determining the intersection feature correction coefficient, the remaining initial fault types that share the same comparison extraction rules with the initial fault type are first screened out, and an associated fault set is constructed. Then, the derivation level of the shared operating parameter features in the reference link is obtained. The strength of feature dependence is determined based on the central proportion of the shared operating parameter features in the reference link topology. Combined with the preset smoothing adjustment parameters, a combination operation is performed. All initial fault types in the associated fault set are traversed to complete the summary operation, and finally, the intersection feature correction coefficient is obtained.

[0021] This invention provides a complete and standardized method for calculating the intersection feature correction coefficient, unifying the interference quantification standard for faults with shared characteristics, and quantifying the interference intensity of fault judgment caused by shared operating parameter features. By dividing the influence weights based on the feature derivation hierarchy and the proportion of the topological center, the influence degree of primary and secondary features on the judgment result can be clearly distinguished. In the case of shared temperature and vibration monitoring features for two types of faults—bearing wear and shaft wear—in the transmission parts of pressing equipment, this invention can accurately distinguish the influence priority of the two types of fault features, avoiding the suppression of the probability judgment result of deep, hidden faults by shallow, simple faults, and ensuring that the judgment of multiple faults under the same source features does not conflict.

[0022] Furthermore, after step S50, the following steps are also included: S61: Based on the operating parameter characteristics of the component to be repaired within a preset time period before the repair, a comprehensive evaluation is performed to obtain the original aging degree of the component to be repaired before the repair, and the historical running time and cumulative fault frequency of the component to be repaired are retrieved simultaneously. S62: Obtain and parse the maintenance operation information of the maintenance components entered on site, and determine the corresponding maintenance execution type, wherein the maintenance execution type includes repair type and replacement type; S63: When the maintenance execution type is maintenance type, the performance recovery coefficient of the maintenance component is obtained from the maintenance operation information, and the original aging degree is corrected and updated through smooth decay calculation; when the maintenance execution type is replacement type, the original aging degree is directly reset to the preset basic loss value. S64: Collect real-time operating status parameters after maintenance is completed, obtain the operating parameter features associated with the fault type of the maintenance component as verification parameter features, comprehensively evaluate the maintenance component based on the verification parameter features, obtain the aging degree of the maintenance component as the verification aging degree, and based on the difference between the verification aging degree and the basic loss value, combined with the operating status parameters of the maintenance component within the preset prediction time after maintenance, obtain the fault type and the corresponding probability of occurrence of the verification parameter features, and generate re-maintenance information.

[0023] After prioritizing faults, this invention differentiates between maintenance and component replacement to update component aging levels. It can also utilize post-maintenance operational data to perform aging verification and generate maintenance prompts, forming a closed-loop process for maintenance operations and condition assessment. Routine maintenance only restores some equipment performance and cannot eliminate inherent fatigue wear on some materials. Component replacement can reset accumulated physical damage, and the differentiated calculations closely reflect the actual physical wear patterns of the equipment. For example, when only surface cleaning maintenance is performed on the pressing cage, the aging value is appropriately lowered; after replacing the entire pressing shaft, the wear baseline is reset. Based on the updated aging state, various wear-related faults that are prone to occur during continuous oilseed pressing operations can be predicted in advance.

[0024] Furthermore, when assessing the aging degree through comprehensive evaluation of operating parameter characteristics, the effective operating parameter characteristics reflecting the component's wear status are first selected based on the structural attributes of the component itself, actual operating conditions, and the inherent correlation between operating status parameters and component aging loss. Then, the effective operating parameter characteristics are classified by status attributes, and a corresponding correlation between each effective operating parameter characteristic and the component's aging change trend is established as an aging relationship. Evaluation weights are assigned to different effective operating parameter characteristics according to the aging relationship, and status analysis is carried out for each effective operating parameter characteristic based on the evaluation weight ratio. The analysis results of all effective operating parameter characteristics are integrated for comprehensive deduction to obtain the corresponding aging degree of the component.

[0025] This invention selects effective loss monitoring parameters based on equipment structure and operating conditions. Through classification and weighted allocation, it comprehensively infers the aging degree of components. This not only eliminates assessment errors caused by environmentally irrelevant parameters but also differentiates assessment weights based on the degree of parameter influence on aging, significantly improving the accuracy of aging level assessment. Different operating parameters affect equipment aging through different mechanisms, and the weighting ratio accurately reflects the trend of loss evolution. For example, when assessing the aging state of hydraulic oil supply components in oil pressing, this invention can discard data interfering with environmental temperature and humidity, focusing on core parameters such as oil pressure fluctuations and oil supply frequency to complete the assessment, eliminating invalid data that could lead to incorrect aging level determination.

[0026] Furthermore, when obtaining the fault type corresponding to the verification parameter features, the state difference relationship is first obtained by comparing the verification aging degree with the preset basic loss value. Based on the state difference relationship, the range of the current overall loss of the maintenance component is defined as the loss range. Then, from the operating status parameters collected within the preset prediction time after maintenance, the operating parameter features associated with the fault type of the maintenance component are obtained. The operating parameter features that do not match the defined current loss range are eliminated to obtain the target parameter features that match the current loss range. The fault type associated with the target parameter features and the corresponding occurrence probability are obtained to generate the re-maintenance information.

[0027] This invention defines component wear ranges based on aging degree verification, thereby simplifying the screening range of operating parameter features. By matching corresponding fault types within the narrowed range, the amount of data computation for fault matching can be significantly reduced, improving fault identification efficiency. Different aging ranges can only induce wear faults at their corresponding levels; faults outside the range do not have the actual conditions to occur. For low-aging range cooling fan components that have been replaced, this invention can automatically eliminate parameter features corresponding to long-term wear faults, and only specifically match faults with instantaneous abnormal operating conditions, simplifying the fault comparison process.

[0028] Furthermore, a monitoring time window is set, and the actual aging rate of the component under maintenance is calculated differentially based on the ratio of the change in the degree of aging of the component under maintenance within the current monitoring time window to the time span; the current operating condition parameters and ambient temperature of the component under maintenance are obtained, and a preset calibration standard aging rate is obtained; the absolute deviation of the actual aging rate from the calibration standard aging rate is calculated, and the absolute deviation is used as the coupled accelerated aging rate; the calibration standard aging rate and the coupled accelerated aging rate are weighted and fused to generate a composite real-time aging rate that reflects the environmental interference of the whole machine, and the physical maintenance cycle of the component under maintenance is dynamically adjusted according to the composite real-time aging rate.

[0029] This invention calculates the inherent aging rate of components using a time window differential method, and combines this with environmental data to calculate the coupled accelerated aging rate of related components. By fusing these two rates, a composite real-time aging rate is obtained. This allows for accurate monitoring of the equipment's natural wear and tear, while also comprehensively incorporating the external force-induced wear from surrounding interconnected equipment. The coordinated operation of multiple devices on a production line can easily lead to operational interference, thereby accelerating the wear rate of individual components. The fusion calculation can accurately reflect the most realistic equipment wear rhythm. By incorporating the accelerated wear caused by the vibration of adjacent feed rollers into the natural aging of the main pressing roller, this invention, based on a comprehensive aging progress, initially achieves flexible adjustment of equipment maintenance time.

[0030] Furthermore, the preset aging limit safety red line value of the maintenance component is retrieved, and the safety degradation margin between the current verification aging degree of the maintenance component and the aging limit safety red line value is calculated. The ratio of the safety degradation margin to the composite real-time aging rate is taken as the remaining physical life time. A coupling hazard weight is generated based on the ratio of the coupled accelerated aging rate to the composite real-time aging rate. A safety adjustment coefficient is obtained based on the coupling hazard weight. The safety adjustment coefficient has a non-linear negative correlation with the coupling hazard weight. When the coupling hazard weight increases, the safety adjustment coefficient decreases monotonically and non-linearly. The remaining physical life time is proportionally contracted and corrected using the safety adjustment coefficient to obtain the corrected safe time span as the physical maintenance cycle of the maintenance component.

[0031] This invention combines the calculation of the remaining service life of equipment with an aging safety red line, and uses a nonlinear adjustment coefficient based on coupled aging hazard weights to correct the maintenance cycle. It can intelligently adjust the maintenance interval according to the level of equipment wear risk, balancing equipment lifespan and production operation safety. The higher the accelerated aging hazard caused by external equipment, the greater the risk of sudden equipment failure; shortening the maintenance interval can mitigate operational risks in advance. Under the high-temperature, high-load production conditions of oilseed pressing, when the proportion of accelerated aging caused by external component interference continues to increase, this invention automatically compresses the predetermined maintenance time span, carrying out maintenance work on vulnerable structures such as seals and transmission components in advance, effectively reducing the probability of sudden production line shutdowns. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating a fault prediction method for an oilseed pressing production line. Figure 2 A flowchart illustrating the maintenance and control procedures; Figure 3 This is a flowchart illustrating the process of providing information for further maintenance. Detailed Implementation

[0033] The following will describe the concept and technical effects of the present invention clearly and completely with reference to embodiments, so as to fully understand the purpose, features and effects of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention. Example 1 like Figure 1 As shown, a fault prediction method for an oilseed pressing production line includes the following steps: S10: Collect the operating status data and corresponding fault types of multiple time nodes within a preset statistical time period.

[0034] The preset statistical time is set by the operation and maintenance personnel and is used to collect historical operating status data of oilseed pressing production line failures. It is generally 1 month to 6 months, and the default value in this embodiment is 1 month.

[0035] S20: Construct and train a fault prediction model using operational status data and fault types. This model outputs prediction results based on operational status data collected at the current time. The prediction results include at least one fault type and its corresponding probability of occurrence. A lightweight gradient boosting decision tree model, commonly used in industrial fault prediction, is adopted. The model structure includes an input layer (input operational status data), three hidden layers (feature weighting layer, association matching layer, and probability output layer), and an output layer (outputting fault type and corresponding probability of occurrence). Using a one-month sample set collected in S10 as training data, the model is divided into training and validation sets in an 8:2 ratio. Training is completed through iterative optimization of the model's loss function. After training, the model can output at least one fault type and its corresponding probability of occurrence in the 0-1 interval based on the input real-time operational status data.

[0036] Based on the linkage working logic of each component on the oilseed pressing production line, the fault type is associated with the characteristics of the operating parameters; the extraction rules of the operating parameter characteristics are determined, the data processing flow of the extraction rules derived from the operating status data is sorted out, and the logical operation link between the operating status data and the feature extraction rules is established.

[0037] The operational status data is collected from the pretreatment, steaming and roasting, pressing, and filtration and refining processes of the oilseed pressing production line. Real-time monitoring parameters of the corresponding dedicated operating components for each process are collected, and the specific relationship between processes and operating components is set by maintenance personnel according to the production line. The pretreatment process corresponds to the conveying and cleaning components; the steaming and roasting process to the heating and stirring components; the pressing process to the pressing and hydraulic drive components; and the filtration and refining process to the filtering, material conveying, and heat exchange components. The data collection time is synchronized with the standard production cycle time of each process (set by maintenance personnel), and each monitoring parameter is bound to the corresponding process execution node timestamp. Simultaneously, real-time supply and demand matching data of the production line is collected. Based on the supply and demand relationship of the production line and the linkage logic between components, cross-process time-series alignment of the operational status data of different process components is achieved.

[0038] The corresponding physical quantities are extracted from the operational status data as basic operational parameter features. These basic operational parameter features include electric drive, temperature and vibration, fluid dynamics, and operating status. Derivative combined features are then obtained from these basic operational parameter features, including energy efficiency ratio and equipment health index. Specifically, the energy efficiency ratio is obtained by calculating the ratio of the actual fuel output to the total power consumption of the motor. The deviations of the four basic features (electric drive, temperature and vibration, fluid dynamics, and operating status) from the corresponding rated standard values ​​set by the maintenance personnel are calculated. Weights are then assigned according to the degree of impact of each basic feature on equipment health (the degree of impact and the method of weighting are set by the maintenance personnel based on the actual operating status of each component on the production line). The weighted average of the deviation values ​​of each feature yields the equipment health index.

[0039] By processing and transforming operational status data through preset extraction rules, operational parameter features that can be used for fault diagnosis are obtained. These extraction rules include time-series rules, numerical rules, frequency domain rules, and correlation rules. Based on the electromechanical-hydraulic integrated equipment structure of the oilseed pressing production line, continuous production operation conditions, the working characteristics of component linkage transmission, the equipment's factory rated parameters and safety thresholds, and combined with the evolution law of progressive aging faults, the frequency domain vibration characteristics of latent mechanical faults, and practical experience in industrial field operation and maintenance, maintenance personnel set four types of preset extraction rules: time-series, numerical, frequency domain, and correlation rules. This is adapted to the characteristics of large load fluctuations and the fact that most faults are progressive and linkage-based, ensuring that the operational parameter features obtained after rule processing and transformation can effectively support fault diagnosis.

[0040] Time-series rules are used to analyze the continuous changing trend of parameters, numerical rules are used to compare the difference between parameters and the rated threshold, frequency domain rules are used to filter and extract latent vibration features, and association rules are used to match multi-parameter linkage anomalies. The data processing flow from operational status data to basic features, then to derived features, and finally to extraction rules is sorted out, and a logical operation link between operational status data and feature extraction rules is established.

[0041] The interconnected operating logic of the components on the oilseed pressing production line is a closed-loop collaborative operation rule formed by the five core components of power drive, core pressing, fluid hydraulics, temperature control and heat dissipation, and monitoring feedback, based on time-series coordination, load coupling, parameter interlocking, and fault propagation. The operating status, load, and parameters of each component are mutually constrained and respond in a coordinated manner, which is the core basis for constructing the fault-characteristic mapping relationship. The main pressing motor provides power to the entire production line, and its output current and speed directly determine the pressing load intensity. Load changes synchronously link the hydraulic pump output pressure and the cooling fan operating speed. The hydraulic system provides pressing pressure and lubrication cooling for the pressing cage and pressing shaft. Pressure fluctuations directly affect the sealing status and motor load. The cooling components are dynamically matched with the motor load. Cooling failure will cause high temperature, which will trigger a chain of abnormalities such as motor overheating and hydraulic oil aging. The gaps, seals, and shell status of pressing components such as the pressing cage and pressing bars will inversely disturb the power and hydraulic system load, causing vibration and pressure deviation. The monitoring components collect electric drive, temperature vibration, and fluid parameters in real time. For example, high-load pressing by the motor can cause a sudden increase in hydraulic pressure, leading to leakage in the press cage seal; a malfunction in the cooling fan causing it to slow down can result in high temperatures, which in turn can lead to motor overheating and oil pressure deviation. Based on the linkage working logic, a fixed mapping between the combination of operating parameter characteristics and fault types can be established, providing technical support for link construction, fault screening, and probability correction.

[0042] S30: Input the currently collected operating status data into the fault prediction model to obtain the current prediction result, and use the fault type in the current prediction result as the initial fault type.

[0043] The operating parameter features associated with the initial fault type are obtained as reference features, and the extraction rules corresponding to these reference features are obtained as reference rules. Based on the linkage logic of the five major components of the oilseed pressing production line—power drive, core pressing, fluid hydraulics, temperature control and heat dissipation, and monitoring feedback—the system first retrieves the fixed mapping relationship between fault types and operating parameter features established in advance according to the linkage logic, based on the fault causes and influencing components of the initial fault type. This mapping relationship is then used to match the operating parameter features corresponding to the initial fault type, which are then used as reference features. Next, based on the four types of extraction rules (time-series, numerical, frequency domain, and correlation) pre-set by maintenance personnel and their exclusive adaptation binding relationships with each operating parameter feature, the system queries and determines the extraction rule corresponding to the reference feature, which is then used as the reference rule.

[0044] Before determining the reference link, corresponding observation periods are configured for different feature extraction rules. The real-time operating load of the production line is obtained by collecting the average current data of the main pressing motor. The preset observation period is dynamically adjusted based on the real-time operating load to obtain the preset prediction duration. Specifically, maintenance personnel pre-configure initial observation periods for four types of feature extraction rules: time-series rules, numerical rules, frequency domain rules, and correlation rules. Real-time current data of the main pressing motor is collected within a short-term statistical window set by the maintenance personnel. The average current of the main pressing motor within the short-term statistical window is calculated. The ratio of the average current to the motor's rated current (determined according to the motor's factory settings) is calculated, and the ratio result is matched with the preset ratio range for different load states to obtain the current load state. This is used to obtain the real-time operating load of the production line. Maintenance personnel set an extension factor for the initial observation period for different load states, and dynamically adjust the preset observation period based on the extension factor. The dynamically adjusted observation period is used as the preset prediction duration.

[0045] For example, maintenance personnel define the load state as light load when the ratio of average current to rated current is no greater than 0.6, medium load when it is between 0.6 and 0.85, and heavy load when it is greater than or equal to 0.85. The maintenance personnel extend the initial observation period of each rule by 1.5 times under light load conditions, keep the initial observation period unchanged under medium load conditions, and shorten the initial observation period of each rule to 0.7 times under heavy load conditions. Assuming the current ratio of average current to motor rated current is 0.56 (no greater than 0.6), the current load state is light load, and the resulting extension factor is 1.5 times.

[0046] Based on the logical operation chain, a reference chain is determined for the operation status data within a preset prediction period (set by maintenance personnel according to maintenance needs; in this embodiment, the default is 24 hours) pointing to the comparison feature. The operation status data within the preset prediction period corresponding to the initial fault type is input into the reference chain to obtain each actual operation rule. If any operation parameter feature extracted from the actual operation rule is not included in the coverage of the comparison feature, the initial fault type corresponding to the reference chain is removed, and the final fault type is determined.

[0047] Assuming the fault prediction model outputs three initial fault types—sealing leakage, bearing wear, and hydraulic pressure deviation—the following steps are taken: First, based on component linkage logic and mapping relationships, the reference characteristics for sealing leakage are determined to be fluid pressure and fluid flow rate, with reference rules of numerical and correlation rules; the reference characteristics for bearing wear are vibration amplitude and housing temperature, with reference rules of time-series and frequency-domain rules; and the reference characteristics for hydraulic pressure deviation are hydraulic oil pressure and pipeline pressure, with reference rules of numerical and time-series rules. Then, according to the logical operation links between operating status data, basic characteristics, derived characteristics, and extraction rules, three independent reference links are determined for each of the three sets of reference characteristics based on the operating data within 24 hours. The 24-hour operating status data corresponding to the initial fault is input into the three reference links to obtain the actual operating rules for each link and extract the actual features. The actual characteristics corresponding to sealing leakage and bearing wear all fall within their respective reference characteristic ranges. However, the abnormal vibration amplitude characteristic was extracted from the actual operating rules corresponding to oil pressure deviation. Since the abnormal vibration amplitude characteristic is not covered by the reference characteristic of oil pressure deviation, the initial fault type of oil pressure deviation was directly eliminated, and sealing leakage and bearing wear were finally determined as the final fault types.

[0048] S40: Adjust the probability of occurrence in the prediction results corresponding to each final fault type based on the number of paths in the reference link corresponding to the final fault type, the hierarchical depth of each path in the reference link, and the other final fault types mapped by the reference rule on the reference link.

[0049] In adjusting the probability of fault occurrence, the probability of occurrence corresponding to the fault type in the prediction results is obtained as the initial probability of occurrence. The initial probability of occurrence is then multiplied and corrected sequentially using the evidence abundance coefficient, intersection feature correction coefficient, and hierarchy depth attenuation factor to obtain the preliminary corrected probability. Finally, the preliminary corrected probabilities of all initial fault types are normalized to obtain the final probability of occurrence for each initial fault type. The evidence abundance coefficient is determined based on the number of reference paths corresponding to the fault. The intersection feature correction coefficient is determined based on the number of shared comparison rules, the topological location of the reference link, and the derivation hierarchy of the operating parameter features within the reference link. The hierarchy depth attenuation factor is calculated based on the average topological hierarchy depth of the reference link corresponding to the fault type.

[0050] Specifically, the preliminary correction probability calculation process is shown in the following formula (1): (1), This represents the initial probability of occurrence. To initially correct the probability, we integrate three dimensions: sufficiency of evidence, feature interference, and credibility of reasoning to correct the original bias in the initial probability of occurrence.

[0051] The normalization calculation process is shown in the following formula (2): (2), For the first The final probability of occurrence of each fault can be directly used to determine the reliable fault probability for maintenance priority. It can also be used to eliminate the difference in correction magnitude between different fault types, so that the corresponding preliminary correction probabilities can be directly compared horizontally, and intuitively reflect the true risk of the fault type. The sum of the initial corrected probabilities of all faults is used as the denominator for normalization, thus achieving probability standardization. The final occurrence probability of the i-th fault is the credible fault probability after standardization, with a value range of 0 to 1 and the sum of the final probabilities of all faults being 1. It can be directly used to determine the maintenance priority.

[0052] The evidence abundance coefficient is used to quantify the sufficiency of evidence in fault determination; the more reference paths there are, the larger the coefficient becomes. This is the intersection feature correction coefficient, used to offset the probabilistic interference caused by shared features of multiple faults and to quantify the degree of influence of feature crossover. This is a hierarchy depth attenuation factor used to reduce the uncertainty of multi-layer indirect derivation faults; the deeper the average topology hierarchy, the smaller the factor.

[0053] Evidence Abundance Coefficient The calculation formula is shown in formula (3) below: (3), To determine the number of valid reference links involved in the i-th fault, the more paths there are, the more sufficient the evidence for the fault. This refers to the number of matching rules that were actually successfully matched on the reference link for this fault. This represents the maximum number of corresponding rules for this fault; The weighting coefficients set for maintenance personnel are determined by them by combining the operating characteristics of the oilseed pressing production line equipment, historical fault statistics, and on-site maintenance experience.

[0054] Path gain factor This is used to characterize the increase in the reliability of a fault by a single valid reference link. Faults in oilseed pressing equipment are mostly induced by component linkages; adding a new valid reference link only slightly increases the strength of the fault evidence. If the value is too high, even a small number of newly added paths can cause an excessive amplification of the probability, easily leading to an artificially high probability of low-risk faults; if the value is too low, it cannot reflect the reference value of the multi-evidence link. In this embodiment, based on the statistical analysis of historical fault samples from the production line, the objective contribution of a single reference path to the fault confidence level is stable in the range of 0.3. Therefore, the maintenance personnel selected... The value is 0.3, which balances the path gain effect with the stability of probability changes.

[0055] Rule Coverage Factor This is used to characterize the adjustment weight of the coefficients based on the completeness of the rule matching in the faulty link. The degree of rule matching directly reflects the fit between the features and the fault mechanism. Compared with the number of newly added paths, the completeness of rule matching has a stronger impact on the authenticity of the fault. Values ​​higher than In this embodiment, the maintenance personnel combined the electromechanical-hydraulic coupling fault patterns of the pressing production line. The reliability of the fault caused by full rule matching is about 0.5 times higher than that caused by a single path. The value of 0.5 can highlight the positive gain of full rule matching, and can also avoid the abnormal surge of coefficients caused by individual full rule matching, thus adapting to the probability correction requirements of progressive and linked faults in the production line.

[0056] Intersection feature correction coefficient The calculation formula is shown in formula (4) below: (4), The set of remaining final fault types that share at least one comparison extraction rule with the i-th fault; is the coefficient of the dependence of the fault on the shared feature, and is the ratio of the betweenness centrality of the shared feature in the reference link topology. The shallower the feature derivation level, the higher the value. To prevent overflow smoothing constants, the operation and maintenance personnel adjust the constants by combining measured operating data shared by multiple fault characteristics of the oilseed pressing production line, historical fault statistics, and on-site operation and maintenance experience.

[0057] Benchmark Guarantee Factor This is used to define the lower limit benchmark value of the formula corresponding to a single set of associated faults. In this embodiment, the electromechanical and hydraulic components of the oilseed pressing production line are highly interconnected and coupled, and many faults share fluid and temperature / vibration monitoring features. When there is no feature dependency between two types of faults... Approaching zero; if If the value is too low, the intersection feature correction coefficient after multiplication of multiple faults can easily approach 0 infinitely, causing the target fault probability to be unreasonably suppressed and high-risk faults to be incorrectly downgraded. In this embodiment, based on the statistical analysis of fault samples in historical operating status data, the baseline correction coefficient in the scenario without feature cross-interference is stable at 0.85. The maintenance personnel selected... The value is 0.85, which retains a reasonable baseline coefficient when there are no shared features, and avoids excessive reduction of the correction coefficient.

[0058] Feature linkage correction coefficient This is used to characterize the coefficient fluctuation caused by the dependence strength of a single set of shared features. In this embodiment, the higher the topological proportion of the shared feature and the earlier its derivation level, the more obvious the probability interference to different faults. However, the feature crossover in the pressing production line is mostly a gradual coupling interference, and a single shared feature only produces a small probability shift; if When the value is too high, and multiple faults share multiple features, the multiplication operation will cause... A surge in numerical values ​​can lead to an artificially high probability of failure; conversely, a value that is too low cannot offset the probability overlap error caused by shared features. In this embodiment, based on actual production line data, a reasonable correction margin of 0.15 is determined for a single set of shared features, and this is selected by the maintenance personnel. The value is 0.15, which slightly corrects the cross-feature deviation and constrains the upper limit of coefficient fluctuation to prevent abnormal expansion and distortion of the coefficient.

[0059] When determining the intersection feature correction coefficient, firstly, the remaining initial fault types that share the same comparison extraction rules with the initial fault type are screened out, and an associated fault set is constructed. Then, the derivation level of the shared operating parameter features in the reference link is obtained. Based on the central proportion of the shared operating parameter features in the reference link topology, the strength of feature dependence is determined. Combined with the preset smoothing adjustment parameters, a combination operation is performed. All initial fault types in the associated fault set are traversed to complete the summary operation, and finally, the intersection feature correction coefficient is obtained.

[0060] Hierarchical Depth Attenuation Factor The calculation formula is shown in formula (5) below: (5), The average topology layer depth of the reference link corresponding to the i-th fault is the deeper the link derivation layer, the more indirect the fault determination. The single-layer inference attenuation and dilution factor is used to characterize the reliability attenuation caused by each downward extension of the reference link's topology level. In this embodiment, the oilseed pressing production line fault exhibits a hierarchical propagation characteristic, with faults induced layer by layer by front-end component anomalies leading to back-end parameter variations. With each additional inference link level, the uncertainty in fault determination increases slightly; if If the value is too large, even a slight increase in the level will cause a sharp drop in the attenuation coefficient, directly and excessively suppressing the probability of deep-level derivation faults, making it easy to miss cascading faults; if If the value is too small, it will be difficult to reflect the uncertainty of long-link inference and will not be able to reasonably reduce the weight of multi-layer indirect failures. Based on historical failure evolution data from the production line, maintenance personnel determined that for each additional level of the inference link in the pressing equipment, the reasonable reduction in failure reliability should be 20%. Therefore, the value was selected... With a value of 0.2, it achieves slight attenuation in short links and gradual and reasonable weight reduction in long links, adapting to the working conditions of cascading fault propagation in production lines.

[0061] S50: Adjust the maintenance priority of the final failure type based on the probability of occurrence. Specifically, sort the final failure types from high to low according to the probability of occurrence. The higher the probability of the final failure, the greater the actual risk of the failure and the more serious the impact on the production line, and the higher the corresponding maintenance priority.

[0062] Example 2 The only difference between this embodiment and Embodiment 1 is that, based on the fault repair priority determination in step S50 of Embodiment 1, this embodiment adds a closed-loop verification and aging status update step for equipment repair, realizing a full-process closed-loop repair control including repair status quantification, aging parameter correction, repair effect verification, and secondary fault early warning. The repair control process is as follows: Figure 2 As shown, it includes the following steps: S61: Based on the operating parameter characteristics of the component to be repaired within a preset time period before the repair, a comprehensive evaluation is performed to obtain the original aging degree of the component before the repair, and the historical running time and cumulative fault frequency of the component to be repaired are retrieved simultaneously.

[0063] When assessing the aging degree through comprehensive evaluation of operating parameter characteristics, the effective operating parameter characteristics reflecting the component's wear status are first selected based on the structural attributes of the component itself, actual operating conditions, and the inherent correlation between operating status parameters and component aging loss. Then, the effective operating parameter characteristics are classified by status attributes, and a corresponding correlation between each effective operating parameter characteristic and the component's aging change trend is established as an aging relationship. Evaluation weights are assigned to different effective operating parameter characteristics according to the aging relationship, and status judgments are carried out for each effective operating parameter characteristic based on the evaluation weight ratio. The judgment results of all effective operating parameter characteristics are integrated for comprehensive deduction to obtain the corresponding aging degree of the component.

[0064] Specifically, the process begins by combining the structural attributes of the target component under maintenance, actual on-site operating conditions, and the inherent correlation between deviations in equipment operating parameters and mechanical aging losses commonly observed in the industrial field. Valid parameters are then screened using a database of equipment failure mechanisms. The screening criteria are threefold: first, based on the mechanical structure principles of the components, determining the physical parameters that change when the component experiences wear, fatigue, or performance degradation; second, based on the component's constant load conditions on the pressing production line, eliminating parameters affected by short-term feed fluctuations and start-stop disturbances; and third, based on publicly available aging test results for similar equipment in the industry, identifying parameters that exhibit a long-term positive correlation with aging losses. The screening process employs a mechanism matching and continuous degradation verification method, where continuous degradation verification relies on trend identification using time-series sampling data of the parameters.

[0065] More specifically, multiple sets of continuous time-series sampling data are taken within the pre-set statistical period before maintenance, and the complete statistical period is divided into several continuous sub-periods of equal length. The average value, relative baseline value offset, and frequency of out-of-limit anomalies of the corresponding parameters are extracted for each sub-period, and the trend of parameter changes in each segment is compared longitudinally. Only when the parameter offset amplitude and the frequency of anomalies show a steady unidirectional deterioration pattern with the equipment operating time are the parameters judged to meet the continuous deterioration requirements and retained as valid parameters. If the parameter only has a sudden jump in a single period, has no fixed change pattern, and oscillates randomly up and down, and does not have long-term gradual deterioration characteristics, it is judged to be an invalid parameter caused by environmental disturbances and instantaneous operating condition fluctuations, and is directly removed. Through the above dual screening logic, the parameter characteristics that can truly reflect the mechanical wear, fatigue aging, and performance degradation of the components are screened from the operating status parameters of the pre-set statistical period before maintenance, and these are defined as valid operating parameter characteristics. At the same time, invalid parameters that only change temporarily and cannot accumulate to reflect aging losses, such as single instantaneous jumps, accidental environmental disturbances, and short-term operating condition fluctuations, are removed.

[0066] Secondly, all the selected effective operating parameter characteristics are classified according to the actual failure mechanism and loss manifestation of the parameters, conforming to the aging patterns screened above, and uniformly divided into four categories: wear parameters, fatigue parameters, leakage parameters, and temperature rise aging parameters. Among them, wear parameters are used to characterize component friction loss and structural clearance changes; fatigue parameters are used to characterize performance degradation caused by long-term alternating loads; leakage parameters are used to characterize abnormal wear of fluid sealing structures; and temperature rise aging parameters are used to characterize material aging and performance degradation caused by high temperatures. A unique correlation is established between effective operating parameter characteristics and component aging change trends. The greater the deviation of equipment operating parameters, the higher the frequency of abnormal fluctuations, and the longer the duration of abnormalities, the higher the degree of aging loss of the corresponding components, thus forming a quantifiable and standardized aging correspondence rule.

[0067] Furthermore, maintenance personnel, based on the wear category of each effective parameter and its contribution to component aging and failure, and relying on historical fault statistics of the pressing production line and practical maintenance experience, assign fixed evaluation weights to each parameter, with all parameter weights added together to reach 100%. Wear and fatigue parameters, which are prone to inducing cascading equipment failures, are assigned higher weights, while leakage and temperature rise aging parameters are matched with corresponding weights according to the scope of the failure's impact. The weight allocation is tailored to the deterioration mechanism of each parameter.

[0068] Finally, based on the deviation of the measured data of each parameter from the benchmark and the level of continuous deterioration, and combined with fixed evaluation weights, the individual loss score is calculated. After weighted summation of all scores, the values ​​are converted to the range of zero to one to obtain the original aging degree of the component before maintenance. The closer the value is to one, the more severe the equipment loss. The historical running time and cumulative failure frequency of the component are retrieved simultaneously to verify the evaluation results and ensure that the aging values ​​are consistent with the actual working conditions on site.

[0069] S62: Acquire and parse the maintenance operation information of the components entered on-site. The maintenance operation information includes key information such as maintenance location, maintenance operation content, operation method, spare parts replacement record, and maintenance items. Through preset keyword matching and operation attribute judgment rules, all maintenance operations (corresponding maintenance execution types) are uniformly divided into repair type and replacement type.

[0070] S63: When the maintenance execution type is repair, the performance recovery coefficient of the maintenance component is obtained from the maintenance operation information, and the original aging degree is corrected and updated through smooth attenuation calculation; when the maintenance execution type is replacement, the original aging degree is directly reset to the preset basic loss value. Specifically, local repair and maintenance operations can only alleviate local losses and eliminate obvious faults, but cannot completely eliminate the cumulative effect of historical aging of the equipment. First, the performance recovery coefficient corresponding to this maintenance is read from the maintenance operation information. This coefficient is preset by the maintenance personnel according to the depth of maintenance operations, the completeness of fault repair, and the equipment maintenance level, and the value range is from 0 to 1. The better the repair effect and the higher the maintenance level, the larger the corresponding coefficient value. Using the industry-standard smooth attenuation calculation method, the original aging degree before maintenance is used as the benchmark, and the aging value is appropriately attenuated and corrected by combining the performance recovery coefficient, weakening the cumulative effect of historical aging of the equipment, retaining the inherent aging attributes of the equipment, and completing the aging degree update.

[0071] When the maintenance execution type is replacement, the brand-new spare parts have no historical aging accumulation or fatigue wear, and directly reset the original aging level of the components to the system's preset basic wear value. This basic wear value corresponds to the inherent slight wear of brand-new equipment at the factory, and the value is close to zero. It can completely clear the historical aging data of the old equipment and match the brand-new operating state after the spare parts are replaced.

[0072] S64: Collect real-time operating status parameters after maintenance is completed, obtain the operating parameter features associated with the fault type of the maintenance component as verification parameter features, comprehensively evaluate the maintenance component based on the verification parameter features, obtain the aging degree of the maintenance component as the verification aging degree, and based on the difference between the verification aging degree and the basic loss value, combined with the operating status parameters of the maintenance component within the preset prediction time after maintenance, obtain the fault type and the corresponding probability of occurrence of the verification parameter features, and generate re-maintenance information.

[0073] When obtaining the fault type corresponding to the verification parameter features, first compare the verification aging degree with the preset basic loss value to obtain the state difference relationship. Based on the state difference relationship, define the current overall loss range of the maintenance component as the loss range. Then, from the operating status parameters collected within the preset prediction time after maintenance, obtain the operating parameter features associated with the fault type of the maintenance component. Remove the operating parameter features that do not match the defined current loss range to obtain the target parameter features that match the current loss range. Obtain the fault type and corresponding occurrence probability associated with the target parameter features to generate the re-maintenance information.

[0074] Specifically, such as Figure 3 As shown, the re-inspection information includes the following steps: S64-1: After the equipment maintenance work is completed, collect the full-dimensional operating status parameters of the maintained components in real time, retrieve the operating parameter characteristics specifically associated with the fault type of the maintained component, and define them as verification parameter characteristics. Referring to the comprehensive evaluation logic in step S61, re-determine the real-time aging degree of the maintained components based on the verification parameter characteristics, and define it as the verification aging degree.

[0075] S64-2: A fixed correspondence is pre-established between the verification aging degree value and the equipment wear range. By calculating the difference between the verification aging degree and the preset basic wear value, the current overall wear level of the equipment is defined. In this embodiment, a wear difference of no more than 0.1 corresponds to a slight wear range, a wear difference greater than 0.1 and no more than 0.4 corresponds to a moderate wear range, and a wear difference greater than 0.4 corresponds to a severe wear range, thereby quantifying the current aging and wear status of the equipment. The thresholds 0.1 and 0.4 are both determined by maintenance personnel based on the characteristics of the oilseed pressing production line equipment and historical fault data. In this embodiment, the threshold 0.1 is the allowable deviation threshold for normal equipment break-in. A wear difference within 0.1 represents only the inherent minor wear of brand-new spare parts and normal operation break-in wear, with no hidden deterioration risk and no risk of inducing abnormal equipment failure; therefore, it serves as the dividing point for the slight wear range. The threshold of 0.4 is the critical value for the inflection point of accelerated equipment aging. Based on the statistics of a large number of historical failure samples from the production line, when the loss difference exceeds 0.4, the wear, temperature rise and sealing deterioration of the equipment will enter an irreversible accelerated deterioration stage, which is very likely to induce the gradual failure and multi-component linkage failure common in the pressing production line. Therefore, it serves as the dividing point between moderate and severe wear.

[0076] S64-3: Collect the continuous operating status parameters of the equipment within the preset predicted time after maintenance, initially extract all parameter features related to the fault type of the maintenance component, combine with the defined equipment wear range, eliminate abnormal and invalid features that do not match the current equipment wear level, retain parameter features that are highly adapted to the actual aging and wear state of the equipment, obtain target parameter features, and avoid interference from invalid parameters on fault determination.

[0077] S64-4: A fixed mapping relationship is pre-established between target parameter feature combinations, fault types, and fault occurrence probabilities. Specifically, different types of wear-related target parameter feature combinations uniquely correspond to specific fault types in the oilseed pressing production line. Furthermore, the greater the deterioration of the target parameter features, the longer the duration of the abnormality, and the higher the deterioration level, the higher the probability of the corresponding fault type occurring. Abnormal combinations of wear-related parameter features correspond to bearing wear and component jamming faults; continuously deteriorating combinations of fatigue-related parameter features correspond to abnormal motor load and component fatigue failure faults; deviation combinations of leakage-related parameter features correspond to oil circuit seal leakage and abnormal hydraulic pressure faults; and excessive combinations of temperature rise and aging-related parameter features correspond to high-temperature aging of equipment and hydraulic oil deterioration faults.

[0078] Based on the effective target parameter features obtained through screening, the potential fault types and corresponding fault occurrence probabilities of the equipment are matched and acquired. Combining the equipment aging and wear range level, fault occurrence probability, and historical maintenance records, standardized re-inspection information is generated. This re-inspection information includes the inspection warning level, potential fault type, recommended inspection time, and maintenance rectification measures, achieving closed-loop early warning and control after equipment maintenance.

[0079] Example 3 The only difference between this embodiment and embodiments 1-2 is that a monitoring time window is set, and the actual aging rate of the component under maintenance is calculated differentially based on the ratio of the change in the degree of aging of the component under maintenance within the current monitoring time window to the time span.

[0080] The monitoring time window is set based on the continuous production characteristics of oilseed pressing equipment, the aging sequence of equipment, and industry aging standards, taking into account both data validity and real-time performance, and providing stable time-series data support for differential calculation of aging rate.

[0081] The aging degree values ​​at the start and end times of the monitoring window are obtained to determine the change in the aging degree of the component under maintenance within the window. This change in aging degree is then compared to the fixed time span corresponding to the monitoring window, and the actual aging rate of the component under maintenance is calculated using a time-series difference method. The change in aging degree and the monitoring time window span together determine the actual aging rate. Given a fixed time span, a larger change in aging degree indicates a faster actual aging rate and more severe aging degradation.

[0082] The system acquires the current operating parameters and ambient temperature of the component under maintenance, and obtains the preset calibration standard aging rate. Specifically, it collects two core influencing parameters in real time: the operating parameters corresponding to the current operating state of the component under maintenance and the ambient temperature. The real-time operating conditions and ambient temperature are matched with the standard operating conditions and standard temperatures in the equipment's factory calibration database. The theoretical aging rate corresponding to the current environmental conditions is retrieved and defined as the calibration standard aging rate. The calibration standard aging rate is the inherent theoretical aging parameter under the equipment's rated operating conditions and standard environment, used as a benchmark to identify additional aging increments caused by fluctuations in external operating conditions.

[0083] The absolute deviation of the actual aging rate from the calibrated standard aging rate is calculated, and this absolute deviation is taken as the coupled accelerated aging rate. Specifically, the actual aging rate of the equipment during actual operation is compared with the theoretical benchmark calibrated standard aging rate, and the absolute deviation between the two is calculated. This absolute deviation represents the additional aging increment caused by environmental fluctuations and is defined as the coupled accelerated aging rate. The coupled accelerated aging rate is used to quantify the accelerated aging losses caused by abnormal operating conditions and environmental temperature disturbances, rather than the natural aging of the equipment itself. The deviation between the actual aging rate and the calibrated standard aging rate uniquely corresponds to the coupled accelerated aging rate; the larger the deviation, the more significant the accelerated impact of external operating conditions and the environment on the aging of the equipment.

[0084] The calibrated standard aging rate and the coupled accelerated aging rate are weighted and fused to generate a composite real-time aging rate that reflects the environmental interference of the entire machine. The physical maintenance cycle of the maintenance components is then dynamically adjusted based on this composite real-time aging rate. Specifically, to balance the inherent natural aging properties of the equipment with the accelerated aging effects of external environmental conditions, the calibrated standard aging rate and the coupled accelerated aging rate are weighted and fused. A fixed fusion weight is configured based on the operating characteristics of the oilseed pressing equipment, with the total weight being 100%. Weights are allocated according to the inherent aging stability of the equipment and the proportion of environmental interference, ultimately generating a composite real-time aging rate. This composite real-time aging rate comprehensively and accurately reflects the overall aging and deterioration state of the equipment under complex on-site conditions, providing an accurate benchmark for subsequent lifespan calculation and maintenance adjustments.

[0085] Retrieve the preset aging limit safety red line value of the maintenance component, calculate the safety degradation margin between the current verification aging degree of the maintenance component and the aging limit safety red line value; take the ratio of the safety degradation margin to the composite real-time aging rate as the remaining physical life time.

[0086] Specifically, pre-configured aging limit safety red line values ​​for maintenance components are established. These values ​​represent the critical safety threshold for equipment failure and disposal, determined by maintenance personnel based on the fatigue limits of component materials, mechanical structural strength thresholds, failure test data of similar equipment in the industry, and statistical samples of production line failures over many years. When the aging degree of a component reaches this red line value, the probability of equipment failures such as wear, fatigue, and seal failure approaches 100%, making safe and stable operation impossible. The equipment must be shut down for maintenance or replacement to ensure safe production on the production line.

[0087] The aging limit safety red line value is retrieved in real time, and the difference between the current verified aging degree of the component under maintenance and the aging limit safety red line value is calculated. The resulting difference is the safety degradation margin, which is used to characterize the remaining safe loss space of the equipment before the failure threshold. Further, a variable correspondence operation relationship is established, and the safety degradation margin is ratioed to the composite real-time aging rate. The result is the initial remaining physical life time of the component under maintenance, which is used to characterize the theoretical remaining safe operating time of the equipment under the current comprehensive aging rate.

[0088] The coupling hazard weight is generated based on the ratio of the coupled accelerated aging rate to the composite real-time aging rate; the safety adjustment coefficient is obtained based on the coupling hazard weight. The safety adjustment coefficient has a non-linear negative correlation with the coupling hazard weight. When the coupling hazard weight increases, the safety adjustment coefficient decreases monotonically and non-linearly.

[0089] Specifically, the proportion of the coupled accelerated aging rate in the composite real-time aging rate is calculated, and this proportion is defined as the coupled hazard weight, which is used to quantify the proportion and risk level of external environmental interference on equipment aging failure. In this embodiment, a fixed nonlinear correspondence rule is preset to establish a unique mapping relationship between the coupled hazard weight and the safety adjustment coefficient. That is, the safety adjustment coefficient and the coupled hazard weight have a nonlinear negative correlation. The larger the coupled hazard weight, the higher the proportion of external interference aging, the stronger the uncertainty of equipment aging, and the higher the operational safety risk. The safety adjustment coefficient decreases monotonically and nonlinearly accordingly, thereby realizing a safety protection logic of more conservative life verification for higher risks.

[0090] The remaining physical lifespan is proportionally reduced using the aforementioned safety reduction coefficient to obtain a corrected safe time span, which serves as the physical maintenance cycle for the maintenance component. Specifically, based on the obtained safety reduction coefficient, the initial remaining physical lifespan is proportionally reduced. Through matching calculations between the coefficient and the initial remaining physical lifespan, the corrected safe time span is obtained. This corrected safe time span is directly determined as the real-time physical maintenance cycle for the maintenance component, enabling adaptive dynamic adjustment of the maintenance cycle according to equipment aging, operating condition fluctuations, and environmental interference. This completely solves the problems of poor adaptability and insufficient risk prediction in traditional fixed maintenance cycles.

[0091] This embodiment accurately distinguishes between inherent natural aging and accelerated aging due to external interference through a complete closed-loop logic of time-series window differential aging measurement, environmental condition coupling correction, hazard weight nonlinear verification, and dynamic correction of remaining lifespan. It can calculate the overall aging rate and remaining safe lifespan of the equipment in real time. By adaptively correcting the maintenance cycle with nonlinear coefficients, it achieves proactive contraction of the maintenance cycle under high-risk conditions and reasonable adaptation of the maintenance cycle under normal conditions. This significantly improves the accuracy of maintenance timing and operational safety of oilseed pressing equipment, avoids the risks of premature equipment deterioration and operation beyond its service life, effectively adapts to complex and variable production scenarios, and comprehensively improves the level of intelligent operation and maintenance and continuous operation stability of production line equipment.

[0092] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for predicting failures in an oilseed pressing production line, characterized in that, Includes the following steps: S10: Collect the operating status data and corresponding fault types of multiple time nodes within a preset statistical time period; S20: Construct and train a fault prediction model using operating status data and fault types. The fault prediction model is used to output prediction results based on the operating status data collected at the current time. The prediction results include at least one fault type and its corresponding probability of occurrence. Associate the operating parameter features of the fault type with the linkage working logic of each component on the oilseed pressing production line. Determine the extraction rules for the features of the operating parameters, sort out the data processing flow of deriving the extraction rules from the operating status data, and build the logical operation link between the operating status data and the feature extraction rules; S30: Input the currently collected operating status data into the fault prediction model to obtain the current prediction result, and take the fault type in the current prediction result as the initial fault type; obtain the operating parameter features associated with the initial fault type as the reference features, obtain the extraction rules corresponding to the reference features as the reference rules, and determine the reference link of the operating status data pointing to the reference features within the preset prediction time according to the logical operation link; input the operating status data within the preset prediction time corresponding to the initial fault type into the reference link to obtain each actual operating rule; if any operating parameter feature extracted from the actual operating rule is not included in the coverage of the reference features, then remove the initial fault type corresponding to the reference link and determine the final fault type; S40: Adjust the probability of occurrence in the prediction results corresponding to each final fault type based on the number of paths in the reference link corresponding to the final fault type, the hierarchical depth of each path in the reference link, and the other final fault types mapped by the reference rule on the reference link. S50: Adjust the maintenance priority of the final fault type according to the probability of occurrence.

2. The fault prediction method for an oilseed pressing production line according to claim 1, characterized in that: The operational status data is collected from real-time monitoring parameters of the corresponding operating components in the pretreatment, steaming and roasting, pressing, and filtration and refining processes of the oilseed pressing production line. The pretreatment process corresponds to the conveying and cleaning components; the steaming and roasting process corresponds to the heating and stirring components; the pressing process corresponds to the pressing and hydraulic drive components; and the filtration and refining process corresponds to the filtering, material conveying, and heat exchange components. The acquisition time of the operational status data is synchronized with the standard production cycle of each process, and each monitoring parameter is bound to the corresponding process execution node timestamp. At the same time, real-time supply and demand matching data of the production line is collected, and cross-process time sequence alignment of the operational status data of different process components is completed based on the supply and demand relationship of the production line and the linkage working logic between the components. Basic operating parameter features are extracted from the operating status data, including electric drive, temperature and vibration, fluid, and working status. Derivative combined features are obtained from the basic operating parameter features, including energy efficiency ratio and equipment health index. The operating status data is processed and transformed by preset extraction rules to obtain operating parameter features that can be used for fault determination. The extraction rules include time-series rules, numerical rules, frequency domain rules, and correlation rules.

3. The fault prediction method for an oilseed pressing production line according to claim 2, characterized in that: Before determining the reference link in step S30, the corresponding observation period is configured for different feature extraction rules. The real-time operating load of the production line is obtained by collecting the average current data of the main pressing motor. The preset observation period is dynamically adjusted according to the real-time operating load to obtain the preset prediction duration.

4. The fault prediction method for an oilseed pressing production line according to claim 3, characterized in that: In step S40, during the process of adjusting the probability of fault occurrence, the probability of occurrence corresponding to the fault type in the prediction result is obtained as the initial probability of occurrence. The initial probability of occurrence is then multiplied and corrected by the evidence abundance coefficient, the intersection feature correction coefficient, and the hierarchical depth attenuation factor to obtain the preliminary corrected probability. Then, the preliminary corrected probabilities of all initial fault types are normalized to obtain the final probability of occurrence of each initial fault type. The evidence abundance coefficient is determined based on the number of reference paths corresponding to the fault; the intersection feature correction coefficient is determined based on the number of shared control rules, the topological location of the reference link, and the derivation level of the operating parameter characteristics within the reference link. The layer depth attenuation factor is calculated based on the average topology layer depth of the reference link corresponding to the fault type.

5. The fault prediction method for an oilseed pressing production line according to claim 4, characterized in that: When determining the intersection feature correction coefficient, firstly, the remaining initial fault types that share the same comparison extraction rules with the initial fault type are screened out, and an associated fault set is constructed. Then, the derivation level of the shared operating parameter features in the reference link is obtained. Based on the central proportion of the shared operating parameter features in the reference link topology, the strength of feature dependence is determined. Combined with the preset smoothing adjustment parameters, a combination operation is performed. All initial fault types in the associated fault set are traversed to complete the summary operation, and finally, the intersection feature correction coefficient is obtained.

6. The fault prediction method for an oilseed pressing production line according to claim 1, characterized in that: Following step S50, the following steps are also included: S61: Based on the operating parameter characteristics of the component to be repaired within a preset time period before the repair, a comprehensive evaluation is performed to obtain the original aging degree of the component to be repaired before the repair, and the historical running time and cumulative fault frequency of the component to be repaired are retrieved simultaneously. S62: Obtain and parse the maintenance operation information of the maintenance components entered on site, and determine the corresponding maintenance execution type, wherein the maintenance execution type includes repair type and replacement type; S63: When the maintenance execution type is maintenance type, the performance recovery coefficient of the maintenance component is obtained from the maintenance operation information, and the original aging degree is corrected and updated through smooth decay calculation; when the maintenance execution type is replacement type, the original aging degree is directly reset to the preset basic loss value. S64: Collect real-time operating status parameters after maintenance is completed, obtain the operating parameter features associated with the fault type of the maintenance component as verification parameter features, comprehensively evaluate the maintenance component based on the verification parameter features, obtain the aging degree of the maintenance component as the verification aging degree, and based on the difference between the verification aging degree and the basic loss value, combined with the operating status parameters of the maintenance component within the preset prediction time after maintenance, obtain the fault type and the corresponding probability of occurrence of the verification parameter features, and generate re-maintenance information.

7. A fault prediction method for an oilseed pressing production line according to claim 6, characterized in that: When the degree of aging is obtained through comprehensive evaluation of operating parameter characteristics, effective operating parameter characteristics that reflect the aging and loss status of the component are first selected based on the structural attributes of the component itself, the actual operating conditions, and the inherent correlation logic between operating status parameters and component aging and loss. Next, the effective operating parameter characteristics are classified into state attributes, and the corresponding correlation between each effective operating parameter characteristic and the aging trend of the component is established as the aging relationship. Evaluation weights are assigned to different effective operating parameter characteristics according to the aging relationship. The state judgment of each effective operating parameter characteristic is carried out according to the evaluation weight ratio. The judgment results of all effective operating parameter characteristics are integrated for comprehensive deduction to obtain the aging degree of the component.

8. A fault prediction method for an oilseed pressing production line according to claim 7, characterized in that: When obtaining the fault type corresponding to the verification parameter features, first compare the verification aging degree with the preset basic loss value to obtain the state difference relationship. Based on the state difference relationship, define the current overall loss range of the maintenance component as the loss range. Then, from the operating status parameters collected within the preset prediction time after maintenance, obtain the operating parameter features associated with the fault type of the maintenance component. Eliminate the operating parameter features that do not match the defined current loss range to obtain the target parameter features that match the current loss range. Obtain the fault types and their corresponding occurrence probabilities associated with the target parameter features, and generate re-maintenance information.

9. A fault prediction method for an oilseed pressing production line according to claim 6, characterized in that: Set a monitoring time window, and calculate the actual aging rate of the maintenance component by differential calculation based on the ratio of the change in the degree of aging of the maintenance component within the current monitoring time window to the time span. Obtain the current operating parameters and ambient temperature of the component under maintenance, and obtain the preset calibration standard aging rate; Calculate the absolute deviation of the actual aging rate from the calibrated standard aging rate, and use the absolute deviation as the coupled accelerated aging rate; The calibration standard aging rate and the coupled accelerated aging rate are weighted and fused to generate a composite real-time aging rate that reflects the environmental interference of the whole machine, and the physical maintenance cycle of the maintenance component is dynamically adjusted according to the composite real-time aging rate.

10. A fault prediction method for an oilseed pressing production line according to claim 9, characterized in that: Retrieve the preset aging limit safety red line value of the maintenance component, calculate the safety degradation margin between the current verification aging degree of the maintenance component and the aging limit safety red line value; take the ratio of the safety degradation margin to the composite real-time aging rate as the remaining physical life time. The coupling hazard weight is generated based on the ratio of the coupled accelerated aging rate to the composite real-time aging rate; the safety adjustment coefficient is obtained based on the coupling hazard weight, and the safety adjustment coefficient has a non-linear negative correlation with the coupling hazard weight. When the coupling hazard weight increases, the safety adjustment coefficient decreases monotonically and non-linearly. The remaining physical lifespan is proportionally reduced and corrected using the aforementioned safety adjustment coefficient, and the corrected safety time span is used as the physical maintenance cycle for the maintenance components.

Images