Intelligent prediction and data tracing management method for lubricating system

Abstract

The invention provides an intelligent prediction and data traceability management method for a lubricating system, and the method comprises the steps: obtaining the multi-source operation data of the input, operation and disposal stages of a lubricating medium in the whole life cycle, building the cross-stage data association based on a unique batch identifier and space-time metadata, and forming a traceable operation data chain. According to the method, dynamic correlation characteristics among equipment operation efficiency, lubricating medium states and environmental conditions are quantified through a multi-dimensional correlation analysis module, and a state evolution decision model is generated in combination with a knowledge rule base. Based on the model, the system calls a risk assessment knowledge base to generate oil replacement prediction suggestions, a component risk priority list and a maintenance strategy report. A maintenance result is fed back to the operation data chain, decision model updating and strategy optimization are driven through a knowledge base self-adaptive calibration mechanism, closed-loop optimization of intelligent lubrication management is achieved, and the fault tracing efficiency, prediction reliability and the maintenance cost-effectiveness ratio are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment lubrication system management technology, and in particular to an intelligent prediction and data traceability management method for lubrication systems. Background Technology

[0002] In industrial production lines, ensuring the efficient operation of lubrication systems is a core management objective for maintaining the stability and extending the service life of critical moving components. Achieving accurate predictive maintenance planning and comprehensive lifecycle data traceability management constitutes a key path to improving system reliability and optimizing management decision-making efficiency. By integrating multi-source operational data, a solid basis can be provided for developing effective predictive maintenance plans, improving the reliability of condition prediction and data traceability, optimizing supply chain and inventory management efficiency, and ultimately driving overall production and operational efficiency improvements.

[0003] However, existing methods primarily focus on localized monitoring during the lubricant usage phase, lacking comprehensive lifecycle data management for lubricants from input and use to disposal. When a failure occurs, root cause analysis becomes difficult, and cross-stage impact analysis is lacking. Furthermore, existing methods generally rely on pre-set and rigid management rules. The system struggles to dynamically adjust and continuously optimize maintenance strategies based on equipment status evolution, new operational data, and dynamically changing operating conditions.

[0004] To address this, a method for intelligent prediction and data traceability management of lubrication systems is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent prediction and data traceability management method for lubrication systems, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent prediction and data traceability management method for lubrication systems, the method comprising:

[0007] Acquire multi-source operational data of the lubricating medium throughout its entire life cycle, and establish cross-stage data associations based on unique batch identifiers and spatiotemporal metadata to form a traceable operational data chain;

[0008] Based on the multi-source operation data, the equipment operation efficiency index, lubrication medium status index and environmental condition index are extracted. Combined with oil chemical deterioration parameters and real-time contamination data, the dynamic correlation characteristics between them are quantified through a multi-dimensional correlation analysis module. Combining the dynamic correlation characteristics with the system health assessment results, a state evolution decision model is generated based on a preset knowledge rule base.

[0009] The state evolution decision model calls a preset risk assessment knowledge base to generate oil replacement prediction suggestions, component risk priority lists and maintenance strategy reports; and the maintenance results are fed back to the operation data chain according to the maintenance strategy reports, and the decision model is updated and the strategy is optimized through the knowledge base self-adaptive calibration mechanism.

[0010] Preferably, the full-life-cycle multi-source operation data comprises:

[0011] In the input stage, the unique batch identification of the lubricating medium, the actual filling process parameters and the initial performance indicators of the lubricating medium are obtained;

[0012] In the operation stage, the multi-type sensor network deployed on the key components and data collection points of the equipment is used to collect the equipment operation efficiency indicators, lubricating medium state indicators and environmental condition indicators in quasi-real time. The sensor deployment should follow the principles of key points, fault-prone points and difficult-to-detect points. For example, for rotating machinery, vibration sensors can be installed vertically, horizontally and axially, and temperature sensors can be installed near the bearing seat and data collection points.

[0013] In the disposal stage, the key degradation parameters of the waste oil and the component wear state data are recorded.

[0014] Preferably, the establishment of the correlation between the data of different stages forms a traceable operation data chain, and the specific steps include:

[0015] Taking the unique batch identification of the lubricating medium as the core traceability key, the operation performance of the batch lubricating medium in different equipment and different time periods, all related abnormal events, and even the final scrap analysis and the component wear data caused thereby are logically correlated to ensure the longitudinal continuity of the data flow. Through the space-time metadata attached to all multi-source operation data, the correlation index between the environmental influence intensity data and the oil product degradation characteristic data is established, so as to analyze the specific influence of environmental factors on the degradation rate of the lubricating medium in a time period and a space region, and to identify the influence mechanism of the external environment on the health of the lubricating system.

[0016] Preferably, the construction of the state evolution decision model comprises the following specific steps:

[0017] The update rule of the decision model comprises the following specific steps: based on the device physical connection topology, the dynamic correlation characteristics identified by the multi-dimensional correlation analysis module are matched with the failure correlation rules stored in the preset knowledge rule base, and the weights and activation conditions of the corresponding paths in the model are activated and updated accordingly; the update trigger of the decision model comprises the following specific steps: when the oil pollution degree breaks through the preset warning threshold, a maintenance action is prompted immediately and a knowledge rule base calibration process is started simultaneously, the calibration process involves fine-tuning of the rule weights and re-evaluation of the trigger conditions, and the data verification and manual confirmation are performed.

[0018] Preferably, the risk assessment knowledge base performs, the specific steps include:

[0019] According to the state evolution decision model, the nodes meeting the preset failure feature conditions are identified as risk sources; according to the device physical connection topology, a conduction path template is retrieved from the associated knowledge rule base, the template describes in detail how the fault is transmitted from one risk source to other related components through the physical connection relationship and predicts the potential impact range; the path association weight factor, the node impact factor and the environment adaptation correction factor are loaded; based on the factor superposition value, a risk level evaluation is generated through a comprehensive risk assessment algorithm, the multi-model combination method includes but is not limited to weighted summation, fuzzy comprehensive evaluation and Bayesian network reasoning, and a detailed risk report is output, the report includes risk source distribution, high-risk path identification and a maintenance priority list sorted according to risk level and impact degree.

[0020] Preferably, the maintenance strategy report generation includes:

[0021] The oil replacement prediction suggestion is based on the chemical degradation parameters of the lubricating medium and the real-time pollution data, calls a preset replacement knowledge rule base to match the device load working condition, combines the actual degradation degree of the lubricating medium and the actual operation load of the device, predicts the remaining effective life of the lubricating medium and gives an optimized replacement opportunity suggestion; the component risk priority list clearly marks the path and high-risk identification whose risk level exceeds a preset threshold in a visual form; the maintenance strategy report also includes specific maintenance suggestions, covering recommended lubricating medium models, detailed replacement and maintenance steps and preventive measures.

[0022] Preferably, the decision model updating and strategy optimization include:

[0023] The device state recovery data in a preset period after maintenance is collected, including whether the data collection point and the component operation performance index are restored to the health baseline; the data feature difference before and after maintenance is compared to identify the actual influence of the maintenance operation on the device state and the lubricating medium performance, and the associated factor weight in the knowledge rule base is dynamically adjusted, the iterative optimization algorithm based on error feedback is used for adjustment, the iterative optimization algorithm can adopt gradient descent method, genetic algorithm and reinforcement learning algorithm, to more accurately reflect the actual causal relationship; when the prediction error of the lubricating medium replacement period and the component failure exceeds the allowable deviation, a knowledge rule base parameter revision process is triggered, the process includes adjusting the parameters of the multi-dimensional association analysis module, updating the threshold of the failure association rule and reconstructing the model parameters, to realize the continuous optimization of the entire intelligent lubrication management system.

[0024] Compared with the prior art, the present application has the following advantages:

[0025] The application takes the unique batch number throughout the whole lubricating medium life cycle as the core traceability key. The multi-source operation data of the lubricating medium in the input, operation and disposal stages are associated through the unique batch number to build a whole life cycle traceable data chain; when an abnormality occurs in the equipment, the maintenance personnel can quickly trace the historical use track of the problem batch lubricating medium, and accurately obtain the operation state, abnormal record and degradation evolution process of the lubricating medium in the multi-equipment environment.

[0026] Compared with the traditional method relying on static rules, the state evolution decision model adopts a multi-dimensional analysis module to deeply mine the associated characteristics between the multi-source operation data; by integrating the dynamic associated characteristics into the equipment state model, the system can accurately represent the real-time health state and degradation trend of the equipment; the application significantly improves the adaptability in complex production environment, effectively reduces the false alarm rate caused by environmental fluctuations, and enhances the practical value of state prediction.

[0027] The system integrates the state model and the risk assessment knowledge base to accurately identify high-risk components and potential fault propagation paths; through a multi-factor risk quantitative evaluation mechanism, a detailed risk report is generated and a customized maintenance strategy is output; based on the risk level, the maintenance resource allocation priority is optimized, so that the maintenance team can focus on high-consequence risk points and avoid unnecessary inspection of healthy components; thereby reducing the labor cost, spare parts consumption and downtime loss, and improving the comprehensive maintenance efficiency.

[0028] Through the comparative analysis of the post-maintenance equipment recovery data and the historical data, the system continuously evaluates the maintenance effect; when the prediction result deviates from the actual state, the adaptive algorithm is relied on to dynamically adjust the analysis module parameters, update the fault judgment rules and optimize the model structure; this closed-loop feedback mechanism ensures that the prediction model and the knowledge rule base can continuously evolve, constantly improving the prediction accuracy and optimizing the maintenance strategy, and maintaining the efficient operation of the intelligent lubrication management system. BRIEF DESCRIPTION OF DRAWINGS

[0029] Fig. 1 A flowchart of an intelligent prediction and data traceability management method for a lubrication system is provided for the embodiments of the application;

[0030] Fig. 2 A running data chain construction schematic diagram is provided for the embodiments of the application;

[0031] Fig. 3 A structure schematic diagram of a state evolution decision model is provided for the embodiments of the application. DETAILED DESCRIPTION

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figs. 1-3 The intelligent prediction and data traceability management method for lubrication systems involved in this invention has the following specific implementation steps:

[0034] Acquire multi-source operational data of the lubricating medium throughout its entire life cycle, and establish cross-stage data associations based on unique batch identifiers and spatiotemporal metadata to form a traceable operational data chain;

[0035] Based on the multi-source operational data, the equipment operating efficiency index, lubricating medium status index, and environmental condition index are extracted, combined with oil chemical degradation parameters and real-time contamination data. The dynamic correlation characteristics among the three are quantified through a multi-dimensional correlation analysis module. Combining the dynamic correlation characteristics with the system health assessment results, a state evolution decision model is generated based on a preset knowledge rule base.

[0036] Based on the state evolution decision model, a preset risk assessment knowledge base is invoked to generate oil replacement prediction suggestions, component risk priority lists, and maintenance strategy reports; operations are performed according to the maintenance strategy reports, and the maintenance results are fed back to the operation data chain, driving decision model updates and strategy optimization through the knowledge base adaptive calibration mechanism.

[0037] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0038] Example 1:

[0039] This embodiment describes a method for obtaining multi-source operational data of a lubricating medium throughout its entire life cycle, and establishes cross-stage data associations based on unique batch identifiers and spatiotemporal metadata to form a traceable operational data chain.

[0040] Specifically, in the lubricating medium input stage, the system collects the unique batch identification of the lubricating medium and records the process parameters at the time of filling; these parameters cover the filling amount, filling pressure, filling time, and filling point; at the same time, the initial performance indicators of the lubricating medium are recorded, including the viscosity, acid value, cleanliness level, and main chemical components of the new oil; when the lubricating medium is put into storage, the above information is bound with the batch number and recorded; during actual filling, the system records the filling environmental conditions, such as temperature and humidity, and supplements them to the initial information; in order to realize the correlation of multi-source operation data, the present method uses a relational database for storage, establishes core data tables of batch information, filling events, and initial performance data, and correlates them by taking the batch number as the primary key, thereby ensuring the traceability of the data.

[0041] In the running stage, the present method collects the equipment running efficiency indicators, lubricating medium state indicators, and environmental condition indicators in quasi-real time through a multi-type sensor network deployed on key components of the equipment and data collection points; the key points refer to the core transmission components or high load bearing components of the equipment, such as main bearings, gearboxes, and pumps; the points prone to failure refer to the positions where historical failure records frequently occur or the known stress concentration areas; the points difficult to detect manually refer to the dangerous areas with complex structure, limited space, or high temperature and high pressure.

[0042] Specifically, the sensor network includes an online laser particle counter, an online capacitive moisture sensor, a three-axis acceleration sensor, a resistance temperature detector, and a temperature and humidity composite sensor; all sensor data transmission and edge computing mainly use the Modbus TCP / IP protocol; for vibration data that needs to be transmitted at high speed, the system considers using the EtherCAT or OPCUA protocol.

[0043] The system uses an industrial-grade edge computing gateway equipped with a multi-core processor, at least 8 GB of RAM, and 128 GB of SSD storage, and carries an embedded Linux operating system, to realize effective transmission of data in the sensor network. The built-in data preprocessing module in the gateway realizes timestamp calibration, unit conversion, preliminary filtering, and outlier detection of the raw data. Subsequently, the processed data is uploaded to the cloud platform in JSON or Protobuf format, so that the transmitted data is standardized and universal. The data collection frequency varies according to the data type, including oil liquid parameters, which are usually collected once a minute; temperature, humidity, and power consumption data are collected every 5 to 10 seconds; vibration data are sampled at high frequency as needed, and the feature values are uploaded at a lower frequency after feature extraction on the edge side. The specific transmission protocol configuration is as follows: Modbus TCP / IP is configured in the master-slave mode with a polling period of 1 second; EtherCAT is used for vibration data to realize microsecond-level synchronization; OPCUA is used for unified data model and secure communication.

[0044] The gateway built-in data preprocessing module adopts a sliding average filtering algorithm with a window size of 5 for preliminary filtering. The Z-score-based method is used for anomaly value detection, and the threshold is set to 3. After ensuring the stability of data transmission, the system obtains vibration characteristic spectrum such as root mean square value, peak value, and kurtosis through the accelerometer; obtains operating temperature through the thermocouple and infrared sensor; obtains power consumption data through the current transformer and power meter. The lubricating medium state parameters are obtained by means of including laser light resistance technology for particle counting; capacitance method and infrared spectrum absorption method for obtaining moisture content; dielectric constant sensor for obtaining dielectric constant. The environmental monitoring sensor collects environmental condition index including on-site environmental temperature, humidity, corrosive gas concentration, and dust concentration.

[0045] The key degradation parameters of waste oil in the disposal stage system and the component wear state data are recorded. The wear element concentration and additive loss are obtained through offline laboratory spectral analysis. The waste oil degradation parameters include the acid value and viscosity change of the waste oil. The component wear state data includes non-invasive detection results, wear type identification after disassembly of the faulty component, and macroscopic wear degree evaluation. The multi-source operation data all contain accurate spatio-temporal metadata including timestamp, device universal identifier, component unique identifier, and data acquisition point spatial location information; the rich and multi-source operation data ensure the traceability and correlation of the data.

[0046] The method takes the unique batch identification of the lubricating medium as the core traceability key. Based on the batch identification, the running performance of the batch lubricating medium in different devices and different time periods, all related abnormal events, and the final scrap analysis are logically associated. The association of multi-source operation data is stored in a relational database, and the batch number is used as the primary key for association.

[0047] Based on the spatio-temporal metadata of the multi-source operation data, an association index between the environmental impact intensity data and the oil product degradation characteristic data is established. The system analyzes the specific influence of environmental factors on the degradation rate of the lubricating medium under specific time periods and spatial regions, i.e., specific devices and data acquisition points, to identify the potential influence mechanism of external environment on lubrication health.

[0048] Based on the traceable operation data chain, the method enters the intelligent analysis and prediction link, and the specific implementation is as follows:

[0049] Firstly, the system extracts the device operation efficiency index, lubricating medium state index, and environmental condition index from the data chain, as well as the oil product chemical degradation parameters and real-time pollution degree data from the disposal stage; then, the multi-dimensional correlation analysis module deeply mines the correlation between these multi-source data, and quantifies the dynamic correlation characteristics between the device operation efficiency index, lubricating medium state index, and environmental condition index.

[0050] The multi-dimensional correlation analysis module uses a long short-term memory network (LSTM) to perform time series analysis on sensor data within a sliding time window, to predict future component health index or lubricating medium key parameter change rate; applies SMOTE technology to process unbalanced fault samples; combines Bayesian networks for structure learning, incorporates known physical and chemical mechanisms as prior knowledge through PC algorithm, to identify causal relationships between variables; a variational autoencoder (VAE) is used for feature fusion and dimensionality reduction of multi-source operating data, to map high-dimensional data to a low-dimensional latent space, providing optimized input for the LSTM model. A ridge regression model is used to quantify the degree of specific influence of equipment load on bearing temperature. The state evolution decision model improves the prediction accuracy and fault analysis capability of the model by inputting quantified dynamic correlation features.

[0051] Further, the multi-dimensional correlation analysis module uses a time series analysis model based on a long short-term memory network to analyze and quantify the time series relationship between multi-source operating data. The input features are a sliding time window, such as original sensor data within the past 24 hours, statistical features such as mean and variance, and trend features such as slope; the output layer of the model predicts the component health index within the next 72 hours, which is a continuous value from 0 to 100, or the change rate of the lubricating medium key parameter; to solve the problem of unbalanced fault samples in historical data, the SMOTE oversampling technology is used to expand the minority class samples during model training.

[0052] The model uses a structure learning algorithm based on Bayesian networks, such as the PC algorithm, to construct a causal graph with known physical and chemical mechanisms of lubrication system wear, oxidation, and pollution as prior knowledge; to identify causal tendencies between variables, such as the causal relationship strength between high temperature and lubricating medium oxidation rate.

[0053] The model uses a variational autoencoder VAE to perform feature fusion and dimensionality reduction on multi-source operating data. The original high-dimensional features are mapped to a low-dimensional latent space to preserve key information. The reduced features can be used as input for the long short-term memory network model.

[0054] The model uses a ridge regression model to quantify the influence of equipment load on bearing temperature by the size of the coefficient, to quantify the degree of influence.

[0055] Preferably, the long short-term memory network model of the embodiment generally comprises 2 to 3 layers of LSTM layers, each layer having 256 units; a ReLU activation function is used and a Dropout rate of 0.2 is set to prevent overfitting. The model uses an Adam optimizer during training. The learning rate is 0.001 during training; the batch size is 64; and 200 epochs are trained. The model uses the Min-Max normalization method to scale the data to the range of 0 to 1 during data preprocessing. The SMOTE oversampling technique sets the k value to 5 and the sampling ratio to 200% to effectively balance the fault samples. In the causal inference auxiliary model, the PC algorithm is implemented using a related open source library and the significance level is set to 0.05. The encoder and decoder of the variational autoencoder VAE are both neural networks comprising two fully connected layers; the number of neurons in each layer is 128 and 64, respectively; and the loss function combines the reconstruction loss and the KL divergence term, with the KL divergence weight being 0.001. The regularization parameter of the ridge regression is determined by cross-validation to be a preset value of 1.0.

[0056] For example, when the equipment load continuously exceeds a certain threshold X% and at the same time the moisture content of the lubricating medium exceeds a certain threshold Yppm, the wear rate of a certain component may show an accelerating trend; an increase in ambient temperature by a certain number of degrees may cause the oxidation rate of the lubricating medium to increase by a certain amplitude. The multi-dimensional correlation analysis module integrates these quantified dynamic correlation features with the preset equipment baseline and the historical fault mode library to realize real-time and refined evaluation of the current health status of each component and the overall lubrication of the equipment.

[0057] The system establishes a preset equipment baseline by collecting long-term stable data of the equipment under normal operating conditions as the equipment baseline and automatically learning its normal parameter range using statistical analysis and anomaly detection algorithms; and builds a historical fault mode library by inducting historical maintenance records, expert experience and fault tree analysis. The historical fault mode library stores typical fault types and associated abnormal data features and thresholds. The evaluation results are usually presented in the form of a numerical health index and discrete health levels, including four levels of very high, high, medium and low. These evaluation results are dynamically updated according to the changes in the latest data and associated features.

[0058] The state evolution decision model comprises a system health assessment module; the system health assessment module collects long-term stable data of the equipment in a normal operating state and takes the long-term stable data as a baseline of the equipment; the system health assessment module applies statistical analysis and abnormality detection algorithms to construct a normal parameter range model; a historical failure mode library is constructed through historical maintenance records, expert experience and fault tree analysis; the historical failure mode library stores typical failure types and associated abnormal data features and thresholds. The system health assessment module combines the quantified dynamic correlation features, the preset equipment baseline and the historical failure mode library. Finally, real-time and refined assessment of the current health status of each component and overall lubrication of the equipment is realized.

[0059] Further, the system health assessment module combines the quantified dynamic correlation features, the preset equipment baseline including parameter standards during running of a new equipment and the historical failure mode library; real-time and refined assessment of the current health status of each component and overall lubrication of the equipment is realized. Generation of the health index is realized through a support vector machine (SVM) based classifier. The classifier takes multi-dimensional correlation features as input, such as principal components of vibration signals, a pollution degree change rate and an accumulated value of temperature abnormalities. Meanwhile, a continuous health index of 0 to 100 is output.

[0060] For example, the PCA reduced features are input into a pre-trained SVM model, and the output value of the model is the health index. The discrete health levels of the health index are divided into, 0 to 25 as extremely high risk, red warning; 26 to 50 as high risk, orange warning; 51 to 75 as medium risk, yellow warning; and 76 to 100 as low risk, green normal. The initial values of the above thresholds are set based on historical failure data analysis and expert experience. The SVM classifier adopts an RBF kernel function, a regularization parameter C is 1.0, and a Gamma value is 0.1. The model is trained on historical data containing normal, slight wear, moderate pollution, early failure and other states through supervised learning, and the label is a health index of 0 to 100; the training data set extracts 30 key features such as vibration, temperature and oil analysis through feature engineering; the number of principal components of PCA reduction is selected as components retaining 95% variance. The threshold of the health level is set based on clustering analysis and statistical distribution of a large amount of historical failure data, for example, the health index is divided into four clusters through K-Means clustering, combined with the experience judgment of field experts on the actual equipment conditions represented by different health index intervals, such as wear degree, pollution level and risk tolerance, and finally determined; for example, a health index lower than 25 usually corresponds to the end of component life or a precursor of a serious failure, and a maintenance personnel needs to be dispatched immediately to maintain the equipment.

[0061] The method combines quantified dynamic correlation features and health assessment results, generates a dynamic state evolution decision model based on a preset knowledge rule base, and comprises the following specific steps:

[0062] The preset knowledge rule base comprises a failure correlation rule, a risk assessment knowledge base, and a database of replacement knowledge rules; the state evolution decision model activates a failure rule matching mechanism according to the lubricating medium pollution degree and other conditions in the knowledge rule base, realizes risk source identification and fault conduction path prediction; by combining the rules in the knowledge rule base and real-time data, the state evolution decision model generates oil replacement prediction suggestions, a component risk priority list, and a detailed maintenance strategy report, realizes intelligent lubrication management, and improves maintenance efficiency.

[0063] Further, the model is determined to be a directed graph model. The nodes of the model represent equipment components such as bearings, gearboxes, and pumps. The node state of the model is quantified through fuzzy logic reasoning, which includes the following: the fuzzy logic reasoning adopts a triangular membership function; the fuzzy rules are based on the IF-THEN form; and the defuzzification method adopts the barycenter method. The edges represent the transition paths between states, and are accompanied by quantitative transition probabilities or influence strengths, such as the probability value of transitioning from moderate pollution to early failure. The multi-level health and abnormal states of the lubrication system are normal operation, slight wear, moderate pollution, overheating warning, early failure, and severe failure. The threshold values can be adaptively adjusted through historical data analysis and expert experience, including using a moving average method or an adaptive algorithm based on model prediction error for dynamic calibration, to ensure that the initial threshold values are reasonable and effective.

[0064] After determining the model framework, new dynamic correlation features are identified through the multi-dimensional correlation analysis module of the model. At the same time, these features are accurately matched with the preset failure correlation rules through the knowledge rule base; the update of the model weights and activation conditions adopts an iterative optimization algorithm based on Bayesian update; and the loss function is defined as the mean square error between the predicted state and the actual state after maintenance. The model calculates the gradient of the loss function with respect to the model weights and activation conditions, and iteratively updates the parameters in the gradient direction to more accurately reflect the actual causal relationship. The failure correlation rules are mined through the Apriori algorithm, wherein the minimum support is set to 0.1, and the minimum confidence is set to 0.7. The Bayesian update algorithm updates the transition probability by calculating the posterior probability. Reinforcement learning can adopt the Q-learning algorithm, with the state defined as the equipment health degree and the lubricating medium state, the action defined as the weight adjustment amplitude, the reward function defined as the negative value of the prediction error, the learning rate defined as 0.01, and the discount factor defined as 0.9. The iterative adjustment algorithm based on historical data statistical frequency calculates the occurrence frequency of each rule through an exponential weighted moving average, and adjusts the weights accordingly.

[0065] The initial structure of the model is built based on expert knowledge, equipment fault tree analysis and historical failure modes. The initial weights or probabilities are set according to expert experience or historical failure frequency. The equipment physical connection topology refers to the real mechanical, hydraulic, lubrication pipeline connection relationship between the internal components of the equipment and their functional dependencies. Through the above operations, it is ensured that the initial state evolution decision model structure is real and effective.

[0066] The failure correlation rules stored in the knowledge rule base are, for example, that an abnormal lubrication pump causes the distributor to be blocked, which in turn causes the data acquisition point to be insufficiently supplied with oil, and finally causes the component to wear. These rules are constructed by combining the experience of domain experts, historical failure data analysis and the equipment physical connection topology model, using association rule mining algorithms or decision tree algorithms; the storage form of the rules can use production rules or semantic nets.

[0067] The failure feature condition is defined as a combination of threshold values or time series patterns of a series of indicators, such as the oil pollution degree being higher than a certain threshold value and the vibration root mean square value being higher than another threshold value, or the temperature continuously rising for a certain number of days. When the multi-dimensional correlation analysis module identifies new and stronger dynamic correlation features, the model will match these features with the failure correlation rules stored in the knowledge rule base, such as discovering that there is a new strong correlation between the continuous abnormal output pressure of the lubrication pump and the continuous 0 reading of the flow sensor of a certain downstream distributor. After data verification and expert confirmation, the weights and activation conditions of the corresponding paths in the model will be updated to make them more close to the current actual operating state of the equipment. The weights are quantified as probability values between 0 and 1, representing the possibility of state transition. The activation condition is defined as a combination of threshold values of a set of indicators. The updating of the weights and activation conditions uses Bayesian updating, reinforcement learning and iterative adjustment algorithms based on historical data statistics frequency.

[0068] The method applies a robust model updating trigger mechanism. When the oil pollution degree breaks through the preset warning threshold, a maintenance action is prompted, such as when the particle count in the lubricating medium exceeds a certain level of the ISO4406 standard, such as 17 / 15 / 12, and the water content exceeds 200ppm, the system will immediately prompt a maintenance action and simultaneously start the knowledge rule base calibration process of the model. The system considers that the abnormality of key lubricating medium parameters in the equipment is not only an alarm signal, but also a strong model feedback learning signal. This signal prompts the evaluation and targeted adjustment of the related rules in the knowledge rule base to optimize the real-time performance and prediction accuracy of the model; at the same time, the knowledge rule base calibration includes fine-tuning of the rule weights and re-evaluation of the trigger conditions to avoid over-learning and instability of the model.

[0069] The method calls a preset risk assessment knowledge base to generate a risk report including oil replacement prediction suggestions, component risk priority lists, and maintenance strategy reports based on a dynamically updated state evolution decision model. During the execution of the risk assessment knowledge base, first, risk sources are identified according to nodes in the model that meet preset failure feature conditions. These conditions are defined according to historical data and domain knowledge to identify which components and data acquisition points have entered a high-risk state.

[0070] The risk assessment knowledge base retrieves a conduction path template from an associated knowledge rule base according to a device physical connection topology. The path template describes in detail how a fault propagates from a risk source to other related components through physical connection relationships. Based on these templates, the model can predict the potential impact range of risks and their conduction paths. Among them, the conduction path template adopts the form of a weighted directed graph; the nodes in the graph represent components, the edges represent physical connections and fault propagation paths, and the weights of the edges represent propagation probabilities or speeds.

[0071] The conduction path template adopts the form of a weighted directed graph; the nodes in the graph represent components, the edges represent physical connections and fault propagation paths, and the weights of the edges represent propagation probabilities or speeds. After obtaining the template in the risk assessment knowledge base, the model loads path-related weight factors, node influence factors, and environment adaptation correction factors to perform multi-dimensional risk superposition evaluation, ensuring the accuracy and sufficiency of the risk assessment results.

[0072] Further, the path-related weight factors directly adopt the probability values of state transitions in the state evolution model, such as the transition probability from the wear state to the early failure state; the node influence factors are determined by the analytic hierarchy process combined with expert scoring, and the evaluation criteria include downtime loss, maintenance cost, safety risk, and production impact, which are obtained based on equipment criticality evaluation and expert scoring; and the environment adaptation correction factor is dynamically adjusted using a Sigmoid nonlinear function. The environment adaptation correction factor reflects the acceleration or delay effect of environmental factors on fault occurrence and development and is determined by fitting historical data. The comprehensive risk assessment algorithm adopts a weighted summation method. The evaluation results are finally mapped to discrete risk levels, and the final risk level mapping rules are as follows: a total risk value of 0 to 0.2 is low risk; 0.2 to 0.5 is medium risk; 0.5 to 0.8 is high risk; and 0.8 to 1.0 is extremely high risk.

[0073] The model calls a preset replacement knowledge rule base to predict the oil replacement period based on the chemical degradation parameters of the lubricating medium such as acid value, viscosity change rate and real-time contamination data. The replacement knowledge rule base includes but is not limited to the rule that when the acid value of the lubricating medium exceeds 0.4 mgKOH / g and the running time exceeds 800 hours, and at the same time, the equipment is in medium or high load working condition, it is recommended to replace within the next 30 days. The system dynamically matches the rules according to the load working condition of the equipment, predicts the remaining effective life of the lubricating medium, and gives an optimized replacement time suggestion.

[0074] The model generates a maintenance strategy report based on the evaluation. When the risk evaluation result identifies that a specific component such as the main bearing has an extremely high risk and the failure type is wear, the model will retrieve and recommend the corresponding lubricating medium model such as ISO VG68 synthetic lubricating medium from the predefined maintenance knowledge base; while providing detailed replacement and maintenance steps. For example, step 1 empty old oil, step 2 flush the system, step 3 add new oil to the specified liquid level, with accompanying illustrations. In addition, the maintenance strategy report also provides preventive measures such as increasing the frequency of oil analysis once a month, checking the wear of the sealing parts. The above suggestions come from expert experience, equipment manufacturer's manual and historical maintenance records, and are sorted and stored, and the model makes recommendations through semantic matching algorithm.

[0075] Based on the fault propagation path obtained by analyzing the device topology, the risk impact range can be predicted and determined, and the risk level of the sudden failure can be quantified by superimposing multiple factors

[0076] The risk level is set according to the risk level, and the system generates a visual report to intuitively display the key hidden dangers of the equipment, and important failures are excluded first, greatly shortening the time of executing decisions.

[0077] The state evolution decision model generates and outputs a detailed risk report. The report includes a list of risk source distribution including all risk components and data collection points; high-risk path identification clearly indicates the key path through which the failure may propagate; and a maintenance priority list sorted by risk level and impact degree of maintenance tasks. According to the risk report, a maintenance strategy report is further generated.

[0078] The component risk priority list in the risk report clearly marks the paths with risk levels exceeding the preset threshold and high-risk identifiers such as color coding and flashing icons to highlight potential failure propagation paths and the most dangerous components. This provides a visual basis for maintenance personnel to make maintenance decisions.

[0079] The maintenance strategy report in the risk report includes specific maintenance recommendations. The recommendations cover recommended lubricating medium models, detailed replacement and maintenance steps, and other necessary preventive measures. After maintenance according to the risk report, the system will feedback the maintenance results to the operation data chain.

[0080] At the same time, the model optimization is triggered by the calibration and iterative optimization mechanism of the knowledge base based on error feedback to drive the decision model update and strategy optimization. When the prediction error of the lubricating medium replacement cycle and the component failure exceeds the allowed deviation, i.e., the absolute error between the predicted occurrence time and the actual occurrence time of the system is less than or equal to 7 days or the relative error is less than or equal to 10%, the parameter revision process of the knowledge rule base is triggered. Specifically, the update of the state evolution decision model adopts an iterative optimization algorithm based on error feedback, including a policy gradient-based time series model algorithm. Based on the algorithm, the associated factor weight is the strategy parameter in the model, and the goal is to maximize the long-term reward, i.e., to minimize the prediction error and improve the maintenance efficiency. The state space definition is the device health, the lubricating medium state and the prediction error, and the action space is defined as the adjustment amplitude of the specific associated factor weight in the knowledge rule base. The algorithm uses the Monte Carlo policy gradient algorithm, the policy network is a neural network containing two fully connected layers, and the output layer is the probability distribution of the adjustment amplitude; the loss function is defined as the negative logarithmic policy gradient multiplied by the reward value; the optimizer is Adam, the learning rate is 0.0001; and the policy parameter is each associated factor weight. The adjustment amplitude in the action space can be defined as a floating point number in [-0.1, 0.1], and the specific associated factor weight in the knowledge rule base is modified by multiplying this floating point number by the current weight and adding it.

[0081] Through the interaction between the model prediction results and the actual maintenance results, the model realizes reinforcement learning and iteratively updates the maintenance strategy. This makes the model more accurately reflect the actual state evolution, while realizing the closed-loop optimization of intelligent lubrication management. In addition to adjusting the weights and thresholds, the calibration and revision process of the knowledge rule base can also assist in optimizing the model structure. Through continuous verification and iteration, the prediction and data management method of the system can continuously improve the data traceability accuracy and the effectiveness of the maintenance strategy, and ultimately realize the intelligentization of the entire lubrication management system.

[0082] The system is fully verified in a controlled environment combined with historical data, and automatically re-evaluates, calibrates and even adjusts part of the prediction model and the parameters of the knowledge rule base under the confirmation of experts, including adjusting the parameters of the multi-dimensional correlation analysis module, updating the threshold of the failure correlation rule and the model parameters.

[0083] The application constructs an end-to-end and intelligent solution from multi-source data collection of lubricating medium to closed-loop optimization of maintenance strategy. The system overcomes the shortcomings of data dispersion, insufficient real-time monitoring, low prediction accuracy and maintenance waste in traditional lubrication management, and solves the problems of weak data correlation, inaccurate fault propagation prediction and neglect of environmental factors in preliminary data-driven management. The system realizes real-time monitoring and accurate prediction of the health status of the lubrication system by making the data chain traceable, the characteristic dynamic correlation quantifiable, the decision model updatable and the risk assessment multi-dimensionally superimposed. Finally, the system provides data traceability for intelligent prediction and safe operation of the reliable operation state of industrial equipment.

[0084] Embodiment 2

[0085] This embodiment will explain in detail how to intelligently predict the lubrication system of the filling machine on the beer production line by the method proposed in the application. The method aims to solve the problems of prediction blind area, response lag and low maintenance efficiency in traditional lubrication management.

[0086] Take the filling machine on a certain beer production line as an example. The distribution of its 26 data collection points includes one three-way vibration sensor and one temperature sensor for each of the main bearing areas, a total of 4 bearings, a total of 16 collection points; one vibration sensor and one temperature sensor are configured for each input and output shaft of the gearbox, a total of 4 collection points; one flow sensor and one pressure sensor are provided for each key lubrication point of the gland machine, a total of 4 collection points; the hydraulic pump station includes one vibration sensor, one temperature sensor and one pressure sensor, a total of 2 collection points. The distribution of the 26 data collection points ensures coverage of the key stress, friction and transmission parts of the filling machine.

[0087] Firstly, this embodiment explains how the model traces the demand data according to the traceable operation data chain, and details how to obtain multi-source operation data of the lubricating medium throughout its life cycle; at the same time, based on the unique batch identification and space-time metadata, cross-stage data association is established to form a traceable operation data chain such as the bottle washing machine, labeling machine and bottle flushing machine in the packaging equipment. These equipment have many data collection points and often work in special working conditions such as humid, high temperature and limited space. At the same time, these packaging equipment also generally use automatic point-to-point lubrication systems such as Beckart, a German company, which integrates a flow sensor in the point electric pump to monitor the oil passage and low liquid level alarm function.

[0088] The system obtains the unique batch number of each batch of lubricating medium by scanning the unique two-dimensional code of each batch of lubricating medium at the input stage; and extracts multi-source operation data strongly associated with the batch number. For example, a batch of lubricating grease for lubricating the bearing of a bottle washing machine, the batch number is 20250101-001, and the manufacturer is X brand; at the same time, the initial performance data of the batch of lubricating grease is recorded, such as drop point 280 degrees Celsius, cone penetration 265; the injection amount and time of each injection to the bottle washing machine, labeling machine and bottle filling machine specific data acquisition point.

[0089] The system continuously collects multi-dimensional data of the bottle washing machine, labeling machine and bottle filling machine in actual production, including equipment operation efficiency indicators. For example, the temperature real-time data of the left bearing component IDXPJ-AXC-L005 of the bottle washing machine; the noise decibel value of the gear meshing component IDTBJ-GRL-002 of the labeling machine. In terms of lubricating medium state indicators, online sensors are deployed in the oil tank and pipeline, including pipeline flow sensors for detecting blockage, real-time monitoring of lubricating medium or grease particle count such as 17 / 15 / 12 level in accordance with ISO4406 standard, and moisture content such as 800ppm of moisture content in the oil return of the bottle washing machine, which is higher than that of the bottle filling machine. In terms of environmental condition indicators, environmental sensors are deployed beside the equipment to collect real-time environmental temperature, humidity and even corrosive gas concentration of the area where the equipment is located. All these multi-source operation data are stamped with accurate time stamp and equipment, component ID, and logically associated with the batch number of the lubricating medium currently in use.

[0090] At the disposal stage, when the lubricating medium reaches the replacement cycle and the equipment components are replaced due to lubrication problems, the system records the analysis report of the waste oil, such as the acid value of the waste oil of the bottle washing machine soaring to three milligrams of potassium hydroxide per gram, and the wear metal element iron content reaching 150ppm through offline spectral analysis. At the same time, the detailed information of the failed components is recorded, such as the serious pitting corrosion of the gear ring of the bottle filling machine. These scrap data are also associated with the batch number of the lubricating medium and the failed component ID.

[0091] Through the above mechanism, all the collected data are closely linked to form a traceable operation data chain from the production, storage, filling of the lubricating medium, to the state change in the equipment operation, to the retirement of the lubricating medium, and to the failure of the component. For example, by tracing the batch number 20250101-001 of the lubricating grease, the system can clearly see that the batch of lubricating grease was filled into the bearing of the bottle washing machine XPJ-AXC-L005 on January 10, 2015, and in the following six months, the running environment humidity of the bearing was always higher than 85% relative humidity, resulting in continuous increase of the moisture content of the lubricating grease. Finally, in early July, the bearing temperature abnormally increased and a slight abnormal sound was produced. The system confirmed that the lubricating grease was failed due to hydrolysis and caused early wear of the bearing according to the waste oil analysis. Through the end-to-end tracing capability of the data chain, the understanding of the complex causal relationship among the performance of the lubricating medium, the health status of the equipment, and the environmental impact is greatly enhanced.

[0092] This embodiment mainly explains in detail how to generate oil replacement prediction suggestions, component risk priority lists, and maintenance strategy reports through data analysis and model construction around the intelligent prediction and data tracing functions.

[0093] Firstly, the system extracts the equipment operation efficiency indicators and lubricating medium state indicators of the root mean square vibration value of the main bearing and the bearing temperature from the multi-source operation data chain of the filling machine in real time. At the same time, the system extracts the online particle count of the gear oil through the sensor, which needs to meet the specific grade requirements of the ISO4406 standard. In addition, the moisture content, dielectric constant, and other lubricating medium state indicators and environmental condition indicators are also extracted through the sensor. These data are combined with the oil chemical degradation parameters such as acid value, viscosity change in the historical waste oil analysis report, and real-time contamination data.

[0094] Before the construction of the state evolution decision model, the dynamic correlation characteristics among these indicators are quantified through the multi-dimensional correlation analysis module in the model. For example, when the root mean square vibration value of the main bearing continuously exceeds the warning line and the particle count of the gear oil increases sharply, the model can identify the strong correlation between bearing wear and oil contamination. Then, the model fuses these dynamic correlation characteristics and the health assessment results of each component of the filling machine, generates a dynamic state evolution decision model based on the preset knowledge rule base. For example, the dynamic correlation characteristic output result can be that the lubricating grease degradation accelerates to cause the abnormal increase of the bearing temperature of the capping machine, which in turn causes the aggravation of the bearing wear, and finally leads to the increase of the risk of equipment downtime. The model will mark the key indicator threshold and decision point on this path. When it is monitored that the temperature of a certain bearing of the capping machine continuously exceeds the safety upper limit, for example, seventy degrees Celsius, and the oxidation degree index of the corresponding lubricating grease significantly rises in a short period of time through the dielectric constant change monitoring, the system will match the rule of lubricating grease accelerated oxidation under high temperature and high load leading to bearing failure, and accordingly adjust the weight of this path in the model, so that it is more sensitive in subsequent prediction.

[0095] The system calls the preset risk assessment knowledge base based on this dynamically updated state evolution decision model. For example, when it is detected that the vibration index of the main bearing component No. GZJ-AXC-001 of the filling machine has been in a high-level early warning state, and the contamination degree of its lubricating medium has reached the threshold value that needs to be replaced immediately, the model will identify the bearing as a high-risk source. Subsequently, the model loads the preset conduction path template, such as the failure of the main bearing may be transmitted to the motor through the gear, and the multi-factor calculation model, including the path correlation weight factor, the node influence factor and the environment adaptation correction factor, carries out risk superposition assessment. Finally, the model generates a detailed risk report, which clearly points out that the main bearing of the filling machine is the highest risk, and reasons that it may occur serious failure within one hundred hours in the future under the current working condition.

[0096] The system further generates a specific maintenance strategy report based on the risk assessment results output by the model and in combination with the risk assessment report and actual management resources. The report will first give oil replacement prediction suggestions, such as predicting that the effective life of the lubricating medium in the filling machine gear box has been shortened to 60% of the original period, i.e. from two thousand hours to one thousand two hundred hours, according to the real-time deterioration data of the lubricating medium and the current load working condition model, and suggests to replace it within the next fifteen days; at the same time, the report will generate a component risk priority list and a risk distribution map; the report lists the main bearing of the filling machine and the related gear box as the highest maintenance priority and traces and locates; finally, the report gives specific maintenance suggestions including recommended lubricating medium model, replacement method, and bearing inspection and replacement scheme report. The report suggests to perform emergency inspection on the main bearing and replace it with a specific model of high-performance bearing to avoid potential downtime risk.

[0097] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent prediction and data traceability management for lubrication system, characterized in that, The method comprises the following steps: Obtaining multi-source running data of lubricating medium in the whole life cycle, establishing cross-stage data association based on unique batch identification and space-time metadata to form a traceable running data chain; Based on the running data chain, extracting equipment running efficiency indicators, lubricating medium state indicators and environmental condition indicators, combining oil chemical degradation parameters and real-time pollution data, and quantifying the dynamic correlation characteristics between them through a multi-dimensional correlation analysis module; combining the dynamic correlation characteristics and system health assessment results, generating a state evolution decision model based on a pre-set knowledge rule base; Based on the state evolution decision model, calling a pre-set risk assessment knowledge base to generate oil replacement prediction suggestions, component risk priority list and maintenance strategy report; executing the operation according to the maintenance strategy report, feeding back the maintenance results to the running data chain, and driving the decision model update and strategy optimization through the knowledge base calibration and iterative optimization mechanism based on error feedback.

2. The lubrication system oriented intelligent prediction and data traceability management method of claim 1, wherein, The multi-source running data in the whole life cycle comprises: In the input stage, the unique batch identification of the lubricating medium, the process parameters at the actual filling time and the initial performance indicators of the lubricating medium are obtained; in the running stage, the multi-type sensor network deployed in the key components of the equipment and the data acquisition points are used to collect the equipment running efficiency indicators, the lubricating medium state indicators and the environmental condition indicators in quasi-real time; in the disposal stage, the key degradation parameters of the waste oil and the component wear state data are recorded.

3. The lubrication system oriented intelligent predictive and data traceability management method of claim 1, wherein, The cross-stage data association relationship for forming the traceable running data chain is established, and the specific steps comprise: Taking the unique batch identification of the lubricating medium as the core traceability key, and based on the traceability key, logically associating the running performance of the batch lubricating medium at different devices and different time periods, all related abnormal events and the final scrap analysis and the component wear data caused thereby to ensure the longitudinal continuity of the data flow; Through the space-time metadata associated with all multi-source running data, the correlation index between the environmental influence intensity data and the oil degradation characteristic data is established, so as to analyze the specific influence of environmental factors on the lubricating medium degradation rate in the time period and the space region, and identify the influence mechanism of the external environment on the health of the lubricating system.

4. The lubrication system oriented intelligent predictive and data traceability management method of claim 1, wherein, The construction of the state evolution decision model comprises the following steps: A directed graph model is constructed, in which the nodes represent various health and abnormal states of the equipment components and the lubricating system, and the edges represent the transition paths between the states; The update rule of the decision model is generated, which comprises matching the failure correlation rules stored in the pre-set knowledge rule base based on the dynamic correlation characteristics identified by the multi-dimensional correlation analysis module through the multi-dimensional correlation analysis module, and activating and updating the weight and activation condition of the corresponding path in the model; The update trigger of the decision model comprises immediately prompting the maintenance action and simultaneously starting the knowledge rule base calibration process when the oil pollution degree breaks through the pre-set warning threshold, the calibration process involves fine-tuning the rule weight and re-evaluating the trigger condition, and the data verification and manual confirmation are performed.

5. The lubrication system oriented intelligent predictive and data traceability management method of claim 1, wherein, The risk assessment knowledge base execution comprises the following steps: According to the nodes in the state evolution decision model that meet the pre-set failure feature conditions, the risk sources are identified; According to the device physical connection topology, a conduction path template is retrieved from the associated knowledge rule base, which describes in detail how the fault is transmitted from a risk source to related components through physical connection relationship and predicts its potential impact range; Load path association weight factor, node impact factor and environmental adaptation correction factor; Based on the factor superposition value, a risk level assessment is generated by a comprehensive risk assessment algorithm, and a detailed risk report is output, which includes risk source distribution, high-risk path identification and maintenance priority list sorted according to risk level and impact degree.

6. The lubrication system oriented intelligent predictive and data traceability management method of claim 1, wherein, The maintenance strategy report generation includes: The oil replacement prediction suggestion is based on the chemical degradation parameters of the lubricating medium and the real-time pollution data, and calls the preset replacement knowledge rule base to match the equipment load working condition, so as to combine the actual degradation degree of the lubricating medium and the actual operation load of the equipment, predict the remaining effective life of the lubricating medium and give the optimized replacement time suggestion; The component risk priority list clearly marks the paths and high-risk identification whose risk level exceeds the preset threshold through visual form; The maintenance strategy report also includes specific maintenance suggestions, including recommended lubricating medium model, detailed replacement and maintenance steps and preventive measures.

7. The lubrication system oriented intelligent predictive and data traceability management method of claim 1, wherein, The decision model update and strategy optimization includes: Obtain the equipment state recovery data within a preset period after maintenance, including whether the data collection point and component performance index are restored to the health baseline; Compare the data feature differences before and after maintenance, identify the actual influence of maintenance operation on equipment state and lubricating medium performance, and dynamically adjust the association factor weight in the knowledge rule base, and adjust through the iteration optimization algorithm based on error feedback, so as to more accurately reflect the actual cause and effect relationship; When the prediction error of lubricating medium replacement period and component failure exceeds the allowed deviation, trigger the knowledge rule base parameter revision process, which includes adjusting the parameters of the multi-dimensional association analysis module, updating the threshold of the failure association rule and reconstructing the model parameters, realizing the closed-loop optimization of the whole intelligent lubrication management system.

Images