Cutting fluid multi-parameter adaptive monitoring and early warning method and system based on dynamic threshold

Abstract

The present invention relates to the field of industrial information and data processing technology, specifically to a multi-parameter adaptive monitoring and early warning method and system for cutting fluid based on dynamic thresholds. The present invention acquires and standardizes the multi-modal data of the cutting fluid in real time, and constructs a coupled attenuation prediction model in combination with the processing condition parameters to predict the joint attenuation trend under specific working conditions. The cutting fluid performance time series is modeled by a Gaussian process, the confidence interval boundary is dynamically adjusted, and an adaptive single parameter dynamic threshold is calculated. The Copula function is then used to learn the nonlinear dependency structure between multiple parameters, identify the joint over-limit risk, and dynamically adjust the multi-parameter combination threshold according to preset rules. The system monitors the real-time data and triggers an early warning when an anomaly is found, uses a digital computer for intelligent decision support processing, and combines the maintenance knowledge base and process optimization rules to generate accurate maintenance recommendations. In addition, the edge intelligent unit ensures the system's independent operation capability in the event of a communication interruption.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial information and data processing, and in particular to a cutting fluid multi-parameter adaptive monitoring and early warning method and system based on dynamic thresholds. Background Art

[0002] The core challenge facing cutting fluid industry monitoring and early warning lies in the inadequacy of digital data processing technology. Specifically, existing technologies suffer from inefficient data preprocessing when processing massive amounts of multi-source, heterogeneous data. This makes it difficult to effectively clean, denoise, and format real-time, highly concurrent sensor data, directly impacting the accuracy of subsequent analysis.

[0003] Furthermore, after initial data processing, feature extraction and data fusion capabilities are severely inadequate. Traditional methods struggle to automatically identify and extract effective features reflecting the key performance characteristics of cutting fluids from complex data streams, resulting in the omission of deep correlations between the data. Fixed thresholds prevent accurate monitoring and prediction of cutting fluid status, significantly limiting the ability to deeply explore the complex nonlinear relationships between the macroscopic state of cutting fluids and their microscopic performance degradation, making it difficult to construct a digital model that accurately reflects their underlying mechanisms. Summary of the Invention

[0004] The present invention provides a multi-parameter adaptive monitoring and early warning method and system for cutting fluids based on dynamic thresholds. By deeply integrating multimodal data from industrial information and data processing with processing parameters, a cutting fluid coupling attenuation prediction model is constructed. A Gaussian process is used to dynamically generate a single parameter adaptive threshold. Copula functions are used to identify multi-parameter nonlinear dependencies and joint over-limit risks, allowing for dynamic adjustment of multi-parameter combination thresholds. The system monitors the cutting fluid status in real time, triggering an early warning if the threshold is exceeded, and using intelligent decision support based on industrial information and data processing to provide precise maintenance and process adjustment recommendations. This significantly improves the accuracy, foresight, and robustness of early warnings, providing strong support for intelligent management of cutting fluids and production optimization.

[0005] The cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic thresholds includes:

[0006] Collecting multimodal data of the cutting fluid in real time, and using a Gaussian process to perform data interpolation and time alignment on data with sampling frequency differences in the multimodal data to generate standardized multimodal data;

[0007] Combining the standardized multimodal data, historical maintenance data and current working condition parameters, a cutting fluid coupling attenuation prediction model is constructed to obtain the future state of the cutting fluid;

[0008] According to the future state of the cutting fluid, the cutting fluid performance time series is modeled by a Gaussian process and the confidence interval boundary is dynamically adjusted to calculate the adaptive single parameter dynamic threshold of the key performance indicator of the cutting fluid;

[0009] Copula functions are used to learn the nonlinear dependency structure between multimodal data, identify the risk of multi-parameter joint over-limit, and dynamically adjust the multi-parameter combination threshold according to the preset propagation path based on the interaction rules between parameters.

[0010] According to the multimodal data monitored in real time, trigger warning conditions are set, and a digital computer is used for intelligent decision support processing to provide users with intelligent maintenance suggestions including adjustment, replenishment and replacement of cutting fluid and adjust the process.

[0011] Preferably, the cutting fluid coupling attenuation prediction model includes:

[0012] Taking the current and predicted machining parameters as input, the deduction can characterize the mutual influence relationship between the various performance indicators of the cutting fluid and the joint attenuation trend under a specific combination of working conditions, and predict the evolution trajectory of each performance indicator.

[0013] Preferably, the calculation of the single parameter dynamic threshold includes:

[0014] Based on the output results of the cutting fluid coupling attenuation prediction model, the time series data of cutting fluid performance is modeled using a Gaussian process non-parametric probability model through industrial information and data processing;

[0015] A confidence interval is given according to the predicted mean and variance of the Gaussian process, and a single parameter dynamic threshold of a key performance indicator of the cutting fluid is calculated by dynamically adjusting the boundary of the confidence interval.

[0016] Preferably, the multi-parameter combination threshold includes:

[0017] Calculating a single parameter dynamic threshold based on the historical statistical distribution of the standardized multimodal data and incorporating the output of the cutting fluid coupling attenuation prediction model;

[0018] The nonlinear dependency structure between the standardized multimodal data is learned using a Copula function to identify the risk of multi-parameter joint exceeding the limit, a nonlinear combination calculation is performed on the key performance indicators of the cutting fluid, and the multi-parameter combination threshold is dynamically adjusted according to a preset propagation path.

[0019] Preferably, the intelligent decision support process includes:

[0020] Using transfer learning methods, model parameters pre-trained on corresponding equipment, working conditions and cutting fluid types are used for initialization and continuous optimization of the cutting fluid coupling attenuation prediction model;

[0021] Based on the prediction model of the cutting fluid's multi-performance coupling attenuation and the machining quality prediction analysis results, specific suggestions for adjusting the cutting fluid parameters are generated and a replenishment operation of the cutting fluid additive is triggered.

[0022] Preferably, the intelligent decision support process further includes:

[0023] When the system triggers an early warning, it uses the type and degree of degradation parameters and the remaining service life predicted by the attenuation prediction model as query conditions to match maintenance plans in the maintenance knowledge base. Combined with historical maintenance records, it comprehensively evaluates and generates specific maintenance operation recommendations, including precise replenishment dosage, adjustment parameter range, and recommended replacement time window.

[0024] Generate suggestions for adjusting machining parameters based on the preset process optimization rule library.

[0025] The cutting fluid multi-parameter adaptive monitoring and early warning system based on dynamic thresholds includes:

[0026] The cutting fluid parameter acquisition and processing module is used to collect multimodal data of cutting fluid in real time. For data with different sampling frequencies in the multimodal data, Gaussian process is used for data interpolation and time alignment to generate standardized multimodal data.

[0027] The cutting fluid attenuation prediction module is used to combine standardized multimodal data, historical maintenance data, and current operating parameters to build a cutting fluid attenuation prediction model. Based on the future state of the cutting fluid, the cutting fluid performance time series is modeled through a Gaussian process and the confidence interval boundaries are dynamically adjusted to calculate the adaptive single-parameter dynamic threshold of the cutting fluid's key performance indicators. The dynamic threshold discrimination module is used to use the Copula function to learn the nonlinear dependency structure between multimodal data, identify the risk of multi-parameter joint over-limit, and dynamically adjust the multi-parameter combination threshold according to the preset propagation path based on the interaction rules between parameters.

[0028] The early warning and decision-making module is used to set triggering early warning conditions based on the multimodal data monitored in real time, and use a digital computer to perform intelligent decision support processing to provide users with intelligent maintenance suggestions including adjusting, replenishing and replacing cutting fluids and adjusting processes. Compared with the existing technology, the beneficial effects of the present invention are:

[0029] 1. This cutting fluid early warning method establishes a single parameter adaptive dynamic threshold through a Gaussian process and utilizes a Copula function to identify multi-parameter nonlinear dependencies and joint over-limit risks. This significantly improves the accuracy and timeliness of early warnings, effectively avoiding the false positives and missed alerts associated with traditional fixed thresholds. By comprehensively understanding the multi-parameter coupling relationships, this method optimizes the information perception capabilities of the Industrial Internet of Things (IIoT), extends the service life of cutting fluids in industry, reduces production risks and costs, and provides an intelligent and efficient maintenance solution.

[0030] 2. By deeply integrating machining parameters into the cutting fluid coupled attenuation prediction model, this method can deduce the mutual influence between various performance indicators and their combined attenuation trends under specific operating conditions, thereby achieving a more comprehensive and in-depth understanding of the cutting fluid's health. This multi-performance coupled evolution prediction capability, based on adaptive operating conditions, provides strong technical support for more reliable early warning, more refined lifespan assessment, and more intelligent maintenance decision-making for Industrial Internet of Things information services.

[0031] 3. Traditional monitoring systems typically only provide basic alarms and lack targeted maintenance recommendations. This invention combines a maintenance knowledge base with a process optimization rule base, combined with cutting fluid performance degradation predictions, to generate precise maintenance strategies, such as optimizing refill plans and adjusting machining parameters. This system not only extends the life of cutting fluids and tools, but also reduces machining defects caused by degradation, maintains industrial data integration services, and thus improves overall production quality and efficiency. This capability is particularly important in high-end manufacturing, significantly reducing the risk of unplanned downtime and improving the stability of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of a cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic thresholds proposed in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the structure of a cutting fluid multi-parameter adaptive monitoring and early warning system based on dynamic thresholds proposed in an embodiment of the present invention;

[0034] Figure 3 This is an interactive process diagram of the cutting fluid multi-parameter adaptive monitoring and early warning method and system based on dynamic thresholds proposed in the embodiment of the application of the present invention. DETAILED DESCRIPTION

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0036] Example 1:

[0037] See also Figures 1 to 3 The present invention provides a cutting fluid multi-parameter adaptive monitoring and early warning method and system based on dynamic thresholds. The technical solution is as follows:

[0038] The cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic thresholds includes:

[0039] S10. Real-time collection of multimodal data of the cutting fluid. For data with sampling frequency differences in the multimodal data, data interpolation and time alignment are performed using a Gaussian process to generate standardized multimodal data.

[0040] S20. Combining the standardized multimodal data, historical maintenance data and current operating parameters, constructing a cutting fluid attenuation prediction model;

[0041] S30. Based on the output of the cutting fluid coupled attenuation prediction model, model the cutting fluid performance time series using a Gaussian process and dynamically adjust the confidence interval boundaries to calculate the adaptive single parameter dynamic threshold of the cutting fluid key performance indicator;

[0042] S40. Use the Copula function to learn the nonlinear dependency structure between multiple parameters of the cutting fluid, identify the risk of multiple parameters exceeding the limit, combine the interaction rules between parameters, and dynamically adjust the multi-parameter combination threshold according to the preset propagation path;

[0043] S50. Compare the cutting fluid status parameters monitored in real time with the multi-parameter combination threshold value. When the detection exceeds the threshold value, a warning condition is triggered, and a digital computer is used for intelligent decision support processing to provide users with intelligent maintenance suggestions including adjustment, replenishment and replacement of cutting fluid and adjust the process.

[0044] As an embodiment of the present invention, refer to Figure 1 ,Flowchart of the cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic threshold.

[0045] Furthermore, the multimodal data acquisition method for industrial databases involves integrating three independent online sensor technologies at key nodes in the cutting fluid circulation system, including the supply line, return line, and fluid tank, to collect multimodal data. Microscopic sensors monitor the cutting fluid's physical and chemical parameters, including pH, conductivity, turbidity, refractometer, absorbance, and temperature. Biosensors detect volatile products produced by microbial metabolism in real time to assess biocontamination. Furthermore, electrochemical ORP sensors indirectly reflect microbial activity through changes in redox potential. Combined with the principles of impedance microbiology, this method provides comprehensive and dynamic monitoring of the cutting fluid's health. After multimodal data acquisition is complete, the physical units of the sensor data are uniformly converted, and the multi-source data is mapped to a unified dimension through normalization. Given the current cutting fluid state parameters as initial conditions, numerical integration methods and current and predicted operating parameters are used as input to predict the changing trends of these key indicators over the next 8 to 24 hours.

[0046] Furthermore, the specific steps of deploying and executing the online parameter estimation algorithm are to select the EKF parameter estimation algorithm, recursively update it, use the hybrid model to predict the state variables at the current moment based on the model parameters and state estimated at the previous moment, and determine the unified alignment time point t_align.

[0047] Generate target alignment time points periodically, and perform alignment and interpolation for each t_align:

[0048] Synchronous high-frequency data (pH, temperature, RPM): The latest data point at or closest to time t_align is directly extracted from the respective buffers.

[0049] Asynchronous data (turbidity, ORP): Extract all historical timestamps and corresponding value data points around t_align from the sliding window buffer corresponding to the asynchronous sensor. Using the extracted timestamps as input and the corresponding values ​​as output, dynamically construct a Gaussian process regression model for the sensor to model smooth trends and noise. Use the trained Gaussian regression model to predict the target time point t_align to obtain the interpolated value for the sensor.

[0050] All aligned numerical data are normalized to a unified dimension to generate standardized multimodal data.

[0051] Furthermore, the current spindle speed, table and toolholder movement speed, and cutting depth are directly read through the machine tool's CNC system. The production plan for the future period is obtained from the Advanced Planning and Scheduling System (APS), and the workpiece material, tool type, and parameters in the preset machining program corresponding to the planned machining task are parsed. Furthermore, the workpiece material and tool type are input as attributes associated with the specific machining task. The workpiece material and tool type parameters are numerically encoded, and the continuous parameters of speed, feed, and cutting depth are standardized to make their numerical range comparable with the cutting fluid's own parameters.

[0052] The processed working condition parameters, together with the cutting fluid's own multimodal parameters and historical data, are used as input features of the Transformer coupled attenuation prediction model.

[0053] Key performance indicators, concentrations of key chemical components, and operating parameters of cutting fluids are used as nodes. Edges represent the direct influence between these indicators. There are edges between the lubricity node and the cooling node, between the micro-indicator node and the pH node and the rust prevention node, and between the operating parameter node and all performance nodes. Edge weights are preset.

[0054] Furthermore, the input sequence for the sequence modeling process is set to include the multi-dimensional performance states of the cutting fluid at historical moments, the corresponding operating parameters, and timestamp information. The attention mechanism is set to learn which state and operating condition combinations in the historical sequence should be focused on when predicting future performance, as well as the interactions between different performance indicators.

[0055] The Transformer model is trained for supervised learning using historical data containing multi-parameter cutting fluid monitoring data, corresponding operating condition data, and known performance degradation results. The loss function is set as the weighted sum of the prediction errors of multiple performance indicators. The trained model performs joint degradation trend prediction. After receiving the current cutting fluid state and current and future predicted operating condition parameters, it simultaneously outputs the evolution trajectory of the cutting fluid's key performance indicators over the next 8 to 24 hours, i.e., the predicted value sequence.

[0056] By deeply integrating the processing condition parameters into the coupled attenuation model, and using advanced algorithms to deduce the working condition parameters deeply into the cutting fluid coupled attenuation prediction model, the mutual influence between various performance indicators and the joint attenuation trend under specific working conditions are deduced, thereby achieving a more comprehensive and in-depth understanding of the health status of the cutting fluid. It can also make a leap based on the adaptive multi-performance coupling evolution prediction of working conditions, providing strong technical support for more reliable early warning, more refined life assessment and more intelligent maintenance decision-making.

[0057] Furthermore, historical time series data of specific parameters in healthy states are collected. The mean and standard deviation of each parameter are calculated and their probability distribution is fitted to set a baseline threshold range: ,

[0058] in, is the parameter mean, is the standard deviation, and It is a coefficient selected based on risk tolerance and parameter characteristics.

[0059] The single parameter dynamic threshold is the result of the combined effect of the baseline threshold, historical statistical distribution characteristics, and future short-term forecast trends. No. The prediction value obtained from the sampling period is , dynamic lower limit It can be expressed as:

[0060] ;

[0061] in, is the weight factor, Indicates the maximum value, is the current value of the parameter, This is a set absolute safety lower limit. With this calculation method, when the predicted rate drops rapidly, the dynamic threshold will be triggered earlier than the static threshold.

[0062] Historical data points for specific performance indicators are collected chronologically from cutting fluid sensors or laboratory analysis data. Based on the characteristics of the cutting fluid performance time series, radial basis functions are used to capture key trends, supplemented by a white noise kernel to simulate measurement errors. Subsequently, the kernel function's hyperparameters are continuously learned and adjusted using maximum likelihood estimation optimization to ensure the model best fits the historical data.

[0063] Furthermore, the preprocessed cutting fluid performance time series data is used to train the Gaussian process model. The training process includes calculating the covariance matrix and the hyperparameters of the learning model. The trained Gaussian process model is used to predict the future short-term cutting fluid performance indicators. The Gaussian process can give the predicted mean and predicted variance, based on which the confidence interval is calculated. For performance indicators that fluctuate within a certain range, the dynamic threshold can simultaneously include the upper and lower limits of the adjusted confidence interval to form a normal interval. The initial confidence level is set to ,set up is the mean, is the variance, and the confidence interval is:

[0064] ;

[0065] in, is the Z value corresponding to the confidence level in the standard normal distribution.

[0066] Based on the prediction results of cutting fluid performance attenuation, the system dynamically adjusts the asymmetric confidence interval: when an accelerated attenuation trend is detected, the warning lower limit is adjusted closer to the predicted mean to achieve early warning; at the same time, the boundary width is automatically adjusted according to the fluctuation range of historical data, tightening when stable and loosening when fluctuations are large, and continuously optimizing through false alarm feedback to form an adaptive dynamic threshold warning mechanism.

[0067] This cutting fluid monitoring and early warning method uses a Gaussian process nonparametric probability model to model performance degradation. It then dynamically adjusts the confidence interval of a single parameter based on future short-term trend predictions to generate adaptive thresholds. Furthermore, using Copula functions to deeply learn and identify the nonlinear dependencies between multiple cutting fluid parameters, it accurately identifies the risk of combined over-limit conditions and dynamically adjusts thresholds. Furthermore, through comprehensive insights into the multi-parameter coupling relationships, it enables a more comprehensive assessment of the overall health of the cutting fluid, optimizes cutting fluid management, extends its service life, and provides a more intelligent and efficient cutting fluid maintenance solution.

[0068] Copula function can connect multiple one-dimensional marginal distribution functions to construct their joint distribution function, that is, any one dimensional joint distribution function Can be done through a Dimensional Copula Function and its marginal distribution functions Expressed as ,in It is standardized multimodal data;

[0069] By constructing a joint distribution, the dependency structure of multiple variables is separated from their respective marginal distributions for modeling. By taking the partial derivative of the joint cumulative distribution function, the joint probability of the standardized multimodal data being simultaneously in an unfavorable state is obtained. The multi-parameter combination threshold is determined, the over-limit area is defined, and the risk of multiple parameters jointly exceeding the limit is identified. In this embodiment, the joint probability of entering this pre-defined "over-limit area" in the next three sampling periods is 0.3647, and the risk threshold is set to 0.10. Because , so a multi-parameter combination warning is triggered.

[0070] Table 1: Rule-based decision logic construction

[0071]

[0072] To more accurately capture the nonlinear dependency structure of cutting fluid multi-parameter data that changes dynamically over time and enhance the robustness of multi-parameter joint over-limit risk identification, this method deeply optimizes the selection of Copula functions, parameter learning, and data processing. Cutting fluid performance degradation is a dynamic process, and the dependency structure between its parameters may also evolve over time. The mutual influence between parameters may differ during the initial wear phase and the later exhaustion phase. To capture this dynamic nature, the parameter learning strategy for the Copula function adopts the following strategy: at each preset update interval, only the data within the current rolling window is used. The Copula parameters are gradually updated based on stochastic gradient descent to obtain the self-updating Copula family parameters and their maximum likelihood estimates. This ensures that the Copula model always reflects the latest changes in the dependency structure and allows for timely adjustments to the assessment of multi-parameter joint risk.

[0073] By incorporating future short-term trend forecasts into the calculation of multi-parameter combined thresholds, warnings are no longer based solely on current values; instead, they can "foresee" the risk of cutting fluid exceeding limits. By predicting the decline in the lubricity index (LI) and the increase in surface roughness (Ra), the system can issue early warnings, buying valuable reaction time for operators. This suggests that decreased lubricity leads to increased surface roughness, providing deeper insights for fault diagnosis and the development of refined predictive maintenance strategies. This provides deeper insights for fault diagnosis and maintenance decision-making. Accurately assessing joint risks and understanding the path of degradation propagation enable the development of more refined predictive maintenance strategies, effectively avoiding equipment downtime and production losses.

[0074] Furthermore, it is necessary to collect historical monitoring data and performance degradation data from multiple similar devices connected during industrial network operation. This data includes machine tools of different models but sharing the same processing principles, as well as data generated under similar operating conditions (processing the same material with slightly different parameters, or using cutting fluids from the same series but different batches / brands). This historical monitoring and performance degradation data comprehensively characterizes the dynamic changes of cutting fluids during actual industrial network operation, providing critical support for building and validating cutting fluid degradation models.

[0075] This data is aggregated to construct a relatively large dataset of universal cutting fluid performance degradation. The Transformer model is applied to this universal dataset to learn the general patterns of cutting fluid performance degradation and the universal dependencies between parameters. While protecting data privacy, the fine-tuned model parameters across multiple edge devices are aggregated to update the global pre-trained model. This updated model is then distributed to enable continuous collaborative model evolution.

[0076] Transfer learning enables the model to start learning from a better initial state on the target task, especially when the target task data is scarce and difficult to obtain. The universal knowledge learned by the pre-trained model from diverse source data helps improve the model's performance when faced with unseen and slightly different data in the target task, enhancing the model's generalization ability and robustness as well as its adaptability to unexpected situations.

[0077] Furthermore, the maintenance knowledge base construction process includes storing standardized maintenance plans for different cutting fluid degradation parameters and degrees of degradation, including mild, moderate, and severe, as well as varying remaining useful life (RUL) intervals. Each plan details recommended operating procedures, required materials, and precautions. Knowledge is then represented using conditional judgment rules, frameworks, and semantic networks.

[0078] The degradation parameter type and degree are directly obtained from the early warning information, and the RUL prediction is obtained from the attenuation prediction model.

[0079] An algorithm that uses keyword matching, fuzzy matching, and vector space models to calculate the similarity between query conditions and knowledge base cases is used to match the most suitable maintenance plan, ensuring that the pushed plan is highly relevant. Comprehensive evaluation and adjustment are carried out in combination with historical maintenance records from the Industrial Internet of Things technology system service. The service stores past maintenance operations through its integrated historical record database, including time, type, specific measures, operators, and feedback on maintenance effects.

[0080] The failure of cutting fluids is not caused by extreme anomalies in a single parameter, but by the synergistic effect of the "sub-healthy" state of multiple parameters. Considering each parameter as a node, and the pairwise dependencies and third-order and higher interaction rules between parameters as edges or hyperedges, a parameter interaction network is constructed, which serves as the core basis for subsequent identification of joint over-limit risks and dynamic adjustment of combined thresholds. When the initial parameters fluctuate abnormally, the network can dynamically track their impact path - the abnormal signal propagates in the network through edges and hyperedges, ultimately leading to a chain reaction of key performance indicators of the cutting fluid. When a synergistic effect is formed in a specific way, the system can promptly warn of potential failure risks and provide a more refined decision-making basis for preventive maintenance.

[0081] By integrating the maintenance experience and best practices of industry experts, the system can automatically match and recommend the most optimized maintenance plan based on real-time operating conditions, thereby achieving intelligent and standardized maintenance decisions, greatly reducing reliance on individual experience, and ensuring the scientific nature and consistency of decision-making. On this basis, the system combines historical maintenance records for deep learning and dynamic adjustment, so that each maintenance recommendation, whether it is supplementary dosage or replacement cycle, is more in line with the actual needs of specific equipment and specific working conditions, achieving precision and personalization of maintenance operations, effectively avoiding "one-size-fits-all" extensive management, and significantly improving the pertinence and actual effectiveness of maintenance. When a warning event occurs, the system can respond quickly and provide specific and actionable maintenance guidance, greatly improving emergency response capabilities and problem-solving efficiency, and significantly shortening fault diagnosis and processing time.

[0082] Table 2: Specific maintenance operation suggestions based on the maintenance knowledge base

[0083]

[0084] To achieve reliable monitoring and early warning in edge computing environments, the system employs a multi-level data validation mechanism, including range checks, null value and outlier detection, and rate-of-change checks. Invalid data is then marked, removed, and interpolated for repair. At the feature extraction level, the system calculates time-domain and frequency-domain features in real time, including spectral energy, dominant frequency, and derived features for specific parameters. A lightweight early warning model is deployed using a three-tiered architecture: simple threshold alarms, rule-based if-then alarms, and decision tree optimization neural networks. As a key component of the Industrial Internet of Things (IIoT) architecture, the edge intelligent unit possesses complete autonomous operation capabilities. Local data storage and processing, as well as the local deployment of core monitoring logic, enable continuous offline operation even in complex IIoT environments. If the IIoT communication link is interrupted, the edge intelligent unit automatically switches to offline mode, triggering alerts via indicators and buzzers based on locally deployed lightweight models and rules. It also caches all generated monitoring data and synchronizes it to the cloud when IIoT network communication is restored, ensuring data integrity and enabling subsequent advanced analysis.

[0085] This design enables the system to maintain core monitoring and basic early warning functions even when communication is limited or components fail, ensuring critical data integrity and business continuity. Initial processing and early warning assessment at the edge improves the system's real-time response speed and efficiency, which is particularly critical for abnormal events requiring rapid response, avoiding processing delays in the cloud. Furthermore, given the high processing pressure on the cloud due to the massive redundancy and high transmission costs of raw data, edge filtering and feature extraction intelligently screen and refine information at the data source, not only improving system response speed and decision-making efficiency, but also maximizing data value at a lower cost. Autonomous capabilities further enhance system robustness and availability, ensuring independent execution of core monitoring and basic early warning during network outages, ensuring essential security monitoring. Furthermore, source data verification improves data quality, providing more reliable input for complex cloud-based model analysis, thereby enhancing analytical accuracy. Ultimately, the rapid response at the edge and the in-depth analysis in the cloud effectively complement each other, forming a layered intelligent monitoring and early warning system.

[0086] Furthermore, the integrated monitoring method is used to obtain a dynamic threshold-based cutting fluid multi-parameter adaptive monitoring and early warning system, which includes a cutting fluid parameter acquisition and processing module, a cutting fluid attenuation prediction module, a dynamic threshold discrimination module, and an early warning and decision module. Figure 2 This is a schematic diagram of the structure of a cutting fluid multi-parameter adaptive monitoring and early warning system based on dynamic thresholds proposed in an embodiment of the present invention, corresponding to:

[0087] The cutting fluid parameter acquisition and processing module is used to collect multimodal data of cutting fluid in real time. For data with different sampling frequencies in the multimodal data, Gaussian process is used for data interpolation and time alignment to generate standardized multimodal data.

[0088] The cutting fluid attenuation prediction module is used to combine standardized multimodal data, historical maintenance data, and current operating parameters to build a cutting fluid attenuation prediction model. Based on the future state of the cutting fluid, the cutting fluid performance time series is modeled through a Gaussian process and the confidence interval boundaries are dynamically adjusted to calculate the adaptive single-parameter dynamic threshold of the cutting fluid's key performance indicators. The dynamic threshold discrimination module is used to use the Copula function to learn the nonlinear dependency structure between multimodal data, identify the risk of multi-parameter joint over-limit, and dynamically adjust the multi-parameter combination threshold according to the preset propagation path based on the interaction rules between parameters.

[0089] The early warning and decision-making module is used to set triggering early warning conditions based on the multimodal data monitored in real time, use a digital computer to perform intelligent decision support processing, and provide users with intelligent maintenance suggestions including adjustment, replenishment and replacement of cutting fluid and adjust the process.

[0090] This system achieves comprehensive perception of the cutting fluid status through multimodal data fusion technology; constructs a dynamic threshold adaptive mechanism based on the Transformer prediction model and Copula function, which can accurately identify the risk of joint degradation of multiple parameters; adopts an edge-cloud collaborative architecture, ensures real-time response capabilities through lightweight edge computing, and relies on cloud big data to optimize long-term prediction performance; ultimately forms a complete intelligent decision-making closed loop from status monitoring, intelligent early warning to maintenance recommendation generation, and realizes data-driven automated maintenance decision-making throughout the entire process.

[0091] The present invention establishes a single parameter adaptive dynamic threshold through Gaussian process, and uses Copula function to identify multi-parameter nonlinear dependence and joint over-limit risks, which significantly improves the accuracy and timeliness of cutting fluid warning, and effectively avoids the false alarms and omissions of traditional fixed thresholds. Deep integration of processing condition parameters achieves a comprehensive insight into the health status of the cutting fluid and the prediction of coupling attenuation under specific working conditions. In conjunction with the intelligent decision support system to provide precise maintenance and process adjustment suggestions, and the edge intelligent unit to ensure system robustness, the present invention provides powerful and intelligent technical support for lean management, life extension and production optimization of cutting fluids, significantly reducing operating costs and production risks.

[0092] At the same time, this method provides highly customized and operational intelligent maintenance decision support, providing users with specific and effective data, thereby extending the effective life of the cutting fluid, reducing the operating personnel's professional system adaptability and multi-performance coupling analysis capabilities and maintenance costs, and improving production efficiency.

[0093] Example 2:

[0094] This example uses high-precision titanium alloy (such as Ti-6Al-4V) for milling. This difficult-to-machine material places extremely high demands on the cooling, lubrication, and chemical stability of the cutting fluid, and titanium alloys are prone to chemical reactions with the cutting fluid. This material is commonly used in aerospace parts, requiring extremely high machining precision.

[0095] It is necessary to pay attention to lubricity, cooling efficiency, and chemical stability at the same time, including specific corrosion inhibitor concentration, titanium ion dissolution concentration, and oxidation resistance. The coupling effect between these properties is more prominent in the processing of high-temperature and highly chemically active titanium alloys.

[0096] Introducing dynamic assessment of the remaining useful life (RUL) of cutting fluids under current and future production schedules. Maintenance decisions are not simply based on refilling. Instead, based on RUL and performance predictions, a phased maintenance strategy is adopted, including minor adjustments, intensive refilling, and planned replacements. This may also trigger coordinated adjustments to process parameters to complete urgent tasks without immediate cutting fluid replacement.

[0097] Maintenance knowledge bases and rules are more sophisticated and require more considerations. The knowledge base needs to include specific additives for titanium alloy processing, including high-temperature lubricants, titanium corrosion inhibitors, and antioxidants. The dose-response relationships can be more complex and influenced by other parameters. Decisions must consider maintenance costs, downtime, and the impact of future production plans on cutting fluid performance.

[0098] Therefore, in titanium alloy machining, the multi-performance coupling of cutting fluids requires coordinated monitoring, and the RUL (Run-Under-Limit) can be dynamically predicted by combining the attenuation characteristics under high-temperature, high-activity conditions. Based on the RUL and performance coupling effects, a phased maintenance strategy is developed, integrating a dose-response knowledge base for titanium alloy-specific additives. This allows for a comprehensive trade-off between maintenance costs, downtime, and production schedules, enabling dynamic optimization decisions.

[0099] Furthermore, based on RUL and performance predictions, a phased maintenance strategy is implemented, including minor adjustments, intensive replenishment, and planned replacements. This triggers coordinated adjustments to process parameters to complete urgent tasks without immediate cutting fluid replacement. The entire maintenance process relies on a continuously updated cutting fluid performance database. The Transformer model continuously optimizes the judgment thresholds and response strategies at each stage, achieving continuous improvement in maintenance efficiency.

[0100] By systematically integrating key parameter indicators of cutting fluids, their corresponding threshold standards, current values ​​monitored in real time, and model-based future trend predictions, the intelligent system can clearly present targeted maintenance measures and optional process adjustment suggestions, thereby transforming the complex monitoring and decision-making process into an intuitive and easy-to-understand action guide, realizing intelligent cutting fluid management.

[0101] Table 3: Intelligent monitoring and maintenance decisions

[0102]

[0103] "Intelligent Monitoring and Maintenance Decision-Making" intuitively presents the system's real-time monitoring results of key cutting fluid parameters and the corresponding maintenance strategies. The core technical foundation supporting these intelligent decisions is reflected in the predictive model and rule base. This forms a complete closed loop of intelligent management of industrial information and data processing: advanced algorithmic models provide a scientific basis for decision-making, and execution feedback continuously optimizes the model accuracy in the rule base. This synergistic mechanism of data-driven, model-supported, and decision-feedback is the core advantage of this system in achieving predictive maintenance.

[0104] Reference Figure 3 , which is an interactive process diagram of the cutting fluid multi-parameter adaptive monitoring and early warning method and system based on dynamic thresholds proposed in the embodiment of the application of the present invention.

[0105] Table 4: Prediction model and rule base summary

[0106]

[0107] This example clearly illustrates how the present invention provides industrial data integration services under demanding and complex working conditions. It demonstrates how, through multi-parameter, multi-model, and multi-stage intelligent decision-making, it achieves refined management of cutting fluids, predictive maintenance, and effective collaboration with production processes. This improves working condition adaptability, deepens performance insights, optimizes maintenance strategies, proactively mitigates risks, and achieves collaborative and economical production and maintenance. This fully demonstrates the significant advantages and advancements of the present invention compared to traditional methods and industrial Internet scenarios.

[0108] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic thresholds, characterized in that: The following steps are involved: Collecting multimodal data of the cutting fluid in real time, and using a Gaussian process to perform data interpolation and time alignment on data with sampling frequency differences in the multimodal data to generate standardized multimodal data; Combining the standardized multimodal data, historical maintenance data and current working condition parameters, a cutting fluid coupling attenuation prediction model is constructed to obtain the future state of the cutting fluid; The cutting fluid coupled attenuation prediction model includes: taking current and predicted machining parameters as input, deducing the mutual influence relationship between various performance indicators of the cutting fluid and the joint attenuation trend under a preset working condition combination, and predicting the evolution trajectory of each performance indicator; Based on the future state of the cutting fluid, a time series model of the cutting fluid performance is constructed using a Gaussian process and the confidence interval boundaries are dynamically adjusted to calculate an adaptive single parameter dynamic threshold value of a key performance indicator of the cutting fluid. The calculation of the single parameter dynamic threshold value includes: based on the output result of the cutting fluid coupling attenuation prediction model, a Gaussian process non-parametric probability model is used to model the time series data of the cutting fluid performance through industrial information and data processing to obtain a Gaussian process prediction mean and variance; a confidence interval is given based on the Gaussian process prediction mean and variance, and the single parameter dynamic threshold value of the key performance indicator of the cutting fluid is calculated by dynamically adjusting the confidence interval boundaries; Copula functions are used to learn the nonlinear dependency structure between multimodal data, identify the risk of multi-parameter joint over-limit, and dynamically adjust the multi-parameter combination threshold according to a preset propagation path in combination with the interaction rules between the parameters. The multi-parameter combination threshold includes: being based on the historical statistical distribution of the standardized multimodal data and incorporating the output results of the cutting fluid coupling attenuation prediction model; using Copula functions to learn the nonlinear dependency structure between the standardized multimodal data to identify the risk of multi-parameter joint over-limit, performing nonlinear combination calculations on the key performance indicators of the cutting fluid, and dynamically adjusting the multi-parameter combination threshold according to the preset propagation path; According to the multimodal data monitored in real time, trigger warning conditions are set, and a digital computer is used for intelligent decision support processing to provide users with intelligent maintenance suggestions including adjustment, replenishment and replacement of cutting fluid and adjust the process.

2. The cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic threshold according to claim 1 is characterized in that: The intelligent decision support process includes: Using transfer learning methods, model parameters pre-trained on corresponding equipment, working conditions and cutting fluid types are used for initialization and continuous optimization of the cutting fluid coupling attenuation prediction model; Based on the prediction model of the cutting fluid multi-performance coupling attenuation and the processing quality prediction analysis results, specific suggestions for adjusting the cutting fluid multi-modal data are generated and the replenishment operation of the cutting fluid additive is triggered.

3. The cutting fluid multi-parameter adaptive monitoring and early warning method based on dynamic threshold according to claim 2 is characterized in that: The intelligent decision support process further includes: When the system triggers an early warning, it uses the type and degree of degradation parameters and the remaining service life predicted by the attenuation prediction model as query conditions to match maintenance plans in the maintenance knowledge base. Combined with historical maintenance records, it comprehensively evaluates and generates specific maintenance operation recommendations, including precise replenishment dosage, adjustment parameter range, and recommended replacement time window. Generate suggestions for adjusting machining parameters based on the preset process optimization rule library.

4. The cutting fluid multi-parameter adaptive monitoring and early warning system based on dynamic threshold is characterized by: include: The cutting fluid parameter acquisition and processing module is used to collect multimodal data of cutting fluid in real time. For data with different sampling frequencies in the multimodal data, Gaussian process is used for data interpolation and time alignment to generate standardized multimodal data. The cutting fluid attenuation prediction module is used to combine standardized multimodal data, historical maintenance data and current working condition parameters to build a cutting fluid attenuation prediction model to obtain the future state of the cutting fluid; The cutting fluid coupling attenuation prediction model includes: taking the current and predicted processing parameters as input, deducing the mutual influence relationship between various performance indicators of the cutting fluid and the joint attenuation trend under the preset working condition combination, and predicting the evolution trajectory of each performance indicator; according to the future state of the cutting fluid, modeling the cutting fluid performance time series through Gaussian process and dynamically adjusting the confidence interval boundary, and calculating the adaptive single parameter dynamic threshold of the key performance indicator of the cutting fluid; the calculation of the single parameter dynamic threshold includes: based on the output result of the cutting fluid coupling attenuation prediction model, through industrial information and data processing, using the Gaussian process non-parametric probability model to model the time series data of the cutting fluid performance to obtain the Gaussian process prediction mean and variance; giving a confidence interval based on the Gaussian process prediction mean and variance, and calculating the single parameter dynamic threshold of the key performance indicator of the cutting fluid by dynamically adjusting the confidence interval boundary; A dynamic threshold discrimination module is used to use Copula functions to learn the nonlinear dependency structure between multimodal data, identify the risk of multi-parameter joint over-limit, and dynamically adjust the multi-parameter combination threshold according to a preset propagation path in combination with the interaction rules between parameters. The multi-parameter combination threshold includes: based on the historical statistical distribution of the standardized multimodal data and incorporating the output results of the cutting fluid coupling attenuation prediction model; using Copula functions to learn the nonlinear dependency structure between the standardized multimodal data to identify the risk of multi-parameter joint over-limit, performing nonlinear combination calculations on the key performance indicators of the cutting fluid, and dynamically adjusting the multi-parameter combination threshold according to the preset propagation path; The early warning and decision-making module is used to set triggering early warning conditions based on the multimodal data monitored in real time, use a digital computer to perform intelligent decision support processing, and provide users with intelligent maintenance suggestions including adjustment, replenishment and replacement of cutting fluid and adjust the process.

Images