Battery-grade lithium carbonate production digital management method and system

Abstract

The invention relates to the technical field of production management, in particular to a battery-grade lithium carbonate production digital management method and system. The method comprises the following steps: acquiring actuator feedback data and actuator control data, and performing control signal-feedback relationship pairing to obtain a control instruction-physical response relationship pair; carrying out standard action response mode identification on the control instruction-physical response relation pair to obtain an actuator standard action response template; real-time actuator electric control behavior data are obtained, physical matching verification is carried out, and an actuator physical action abnormity report is obtained; deducing the physical action exception report of the actuator according to the control instruction-physical response relationship to obtain a process chain segment exception path table; and performing response trend analysis based on the process chain segment abnormal path table to obtain an actuator start-stop control instruction table. The real-time performance and the automation level of production response are improved, and the quality stability and the energy efficiency output performance of the battery-grade lithium carbonate product are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of production management, and in particular to a digital management method and system for battery-grade lithium carbonate production. Background Art

[0002] Battery-grade lithium carbonate, a key raw material for lithium-ion battery cathodes in the new energy industry, is crucial for its purity, particle size distribution, and impurity content. Its physical and chemical properties, including purity, particle size distribution, and impurity content, directly determine the energy density, cycle life, and safety of battery products. Traditional lithium carbonate production processes typically utilize mineral raw materials such as salt lake brine and spodumene, followed by a series of steps including lithium extraction, purification, concentration, lithium precipitation, and drying. However, these processes are limited by high raw material volatility, complex process paths, and a high degree of manual intervention in key parameters, leading to product quality fluctuations and increased energy costs. The current industry's prevalent management approach relies heavily on manual experience. Even some companies have introduced localized information systems, which remain limited to single-point monitoring and data logging, lacking integrated data collection and intelligent decision-making support across the entire process. Traditional systems struggle to achieve real-time response and dynamic adjustment in areas such as multi-source data integration, process anomaly identification, batch traceability management, and energy efficiency assessment and optimization. This results in difficulties ensuring consistency across batches of lithium carbonate products, delayed production decisions, and lagging control of quality deviations, severely restricting the stable supply of high-end battery materials. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a digital management method and system for battery-grade lithium carbonate production to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a digital management method for battery-grade lithium carbonate production comprises the following steps:

[0005] Step S1: acquiring actuator feedback data and actuator control data, and performing control signal-feedback relationship pairing based on the actuator feedback data and the actuator control data to obtain a control command-physical response relationship pair;

[0006] Step S2: performing control periodic sampling on the control instruction-physical response relationship pair to obtain a control period actuator parameter set; performing standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template;

[0007] Step S3: Acquire real-time actuator electronic control behavior data, and perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify response mismatch behavior, and obtain an actuator physical action abnormality report;

[0008] Step S4: deriving the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report to obtain a process chain segment abnormal path table;

[0009] Step S5: Perform actuator action response trend analysis based on the process chain segment abnormal path table, and determine the start-stop control parameter amplitude based on the action response trend analysis results to obtain the actuator start-stop control instruction table; upload the actuator start-stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

[0010] By establishing a one-to-one pairing relationship between control signals and physical responses, the present invention provides a clear causal basis for subsequent analysis. Under conditions where the control-feedback relationship is clear, periodic sampling of the control can effectively filter out sporadic fluctuations, forming a stable and representative set of actuator parameters. This set is used to construct a standard action response template, providing a quantitative reference for determining whether the behavior deviates from the standard. Furthermore, by introducing a physical matching verification mechanism for real-time electronic control behavior and standard templates, response mismatches can be detected in advance during the data acquisition phase, preventing abnormal action delays from causing chain reactions on the production line. Joint analysis of data in abnormality reports and control instructions supports the precise location of the physical process segments affected by abnormalities, thereby improving the granularity of fault diagnosis and shortening fault location time. The establishment of an abnormality path table provides structured input for trend analysis. The amplitude of the start-stop instructions is further controlled in combination with the trend analysis results. When adjusting the amplitude parameters, the dynamic response change rate is used as a judgment benchmark to prevent the control system from causing system oscillation due to excessive amplitude fluctuations, thus achieving flexible adjustment of the actuator while ensuring accuracy. Finally, the start and stop instructions will be uploaded to the management platform to ensure that the management instructions are traceable and can be executed automatically, thereby improving the overall control closed-loop capability and dynamic fault response capability of the lithium carbonate production line.

[0011] Optionally, step S1 specifically includes:

[0012] Step S11: collecting actuator control signals and corresponding feedback signals to obtain actuator feedback data and actuator control data;

[0013] Step S12: performing channel mapping and synchronous pairing on the actuator feedback data and the actuator control data to generate a control feedback channel binding table;

[0014] Step S13: extracting feedback feature segments within a preset action cycle based on the control feedback channel binding table to form a control feedback response sample set;

[0015] Step S14: performing control mode identification on the control feedback response sample set to obtain a controller control mode sample set;

[0016] Step S15: Classify and archive the controller control mode sample set according to actuator type and control mode, and generate control instruction-physical response relationship pairs.

[0017] The present invention clearly collects control signals and feedback signals, maps and synchronizes them, and constructs a unified control feedback channel binding table. This effectively solves the problems of unclear signal attribution and channel mismatch, providing a structural foundation for orderly data management. On this basis, a preset action cycle is set to extract feedback feature segments. On the one hand, non-target behavior interference signals can be eliminated by limiting the cycle, and on the other hand, it is convenient to construct a sample set with time constraints. The cycle is generally set according to the mechanical response delay characteristics of the actuator action (such as 1.5 seconds to 3 seconds) to balance response integrity and real-time performance. Control pattern recognition based on the extracted response sample set helps to distinguish representative control behavior patterns from a large amount of data, improving the model's adaptability and generalization ability to actuator characteristic responses. Further, the identified control patterns are archived according to actuator category and control behavior. The physical response behavior of different types of actuators under different control conditions can be systematically managed, and ultimately a functionally clear control instruction-physical response relationship pair is formed. Subsequent matching analysis, anomaly detection, and path tracing all have a structured and reusable data foundation, fundamentally improving the accuracy, traceability, and response consistency of production line actuator control.

[0018] Optionally, step S12 is specifically as follows:

[0019] Step S121: Retrieve the control output port corresponding to each actuator and the acquisition input channel configuration document of the physical feedback feedback device through the lithium carbonate production line management platform, combine the obtained equipment wiring diagram and on-site wiring labels, establish the actuator-feedback device physical connection relationship, and thus obtain a physical channel index table;

[0020] Step S122: Set the synchronization sliding time window to 0.5s, and use the physical channel index table to perform control signal-feedback signal synchronization window slicing to obtain aligned data segment pairs;

[0021] Step S123: Periodically reorganize the aligned data segment pairs and verify feedback direction consistency, remove channel pairs with feedback value noise exceeding ±10% or missing points, and generate a stable control response data set;

[0022] Step S124: normalize and organize the stable control response data set, and construct a control feedback channel binding table in the format of actuator number, control signal geology, feedback signal geology, feedback delay mean, and response amplitude mean.

[0023] This invention retrieves configuration files for control output ports and feedback input channels from a management platform and constructs a physical connection mapping based on on-site wiring diagrams and wiring labels. This clarifies the physical link between actuators and feedback devices from the source, preventing data channel mismatches caused by incomplete documentation or confusing wiring. A 0.5-second synchronous sliding time window is further set for control-feedback signal slicing. This time window size is based on the typical actuator action response cycle. This ensures full coverage of the feedback segment in most control events while preventing invalid data introduced by excessive window lengths, thereby improving synchronization accuracy and data utilization. Periodic reorganization and directional consistency checks, combined with a ±10% noise threshold and missing point exclusion mechanism, effectively filter out interference caused by signal offset and feedback jitter, improving the reliability and representativeness of subsequent modeling samples. Finally, the output control feedback channel binding table is organized by actuator number and multiple statistical indicators, providing a clear structural description and parameterized definition between control channel pairs. This provides a stable data foundation and physical mapping support for subsequent control mode recognition and action anomaly detection, significantly enhancing the transparency of the control link and the effectiveness of actuator behavior tracking.

[0024] Optionally, step S13 is specifically as follows:

[0025] Step S131: Read each pair of control-feedback channel combinations in the control feedback channel binding table, set the standard action cycle window to 300s, and extract the synchronous control trigger event sequence;

[0026] Step S132: performing boundary regularization and signal interception on the feedback data segments corresponding to each group of control trigger events in the synchronous control trigger event sequence, and setting the response amplitude threshold to 3% to extract the segments with continuous signal changes;

[0027] Step S133: Encode and encapsulate the continuously changing signal segments to generate response curve segment samples;

[0028] Step S134: Summarize the response samples of each channel combination within the standard action cycle window, eliminate the channel combinations and corresponding samples with a response amplitude change standard deviation <0.05 and a response start delay standard deviation <0.3s, and obtain a control feedback response sample set.

[0029] The present invention sets the standard action cycle window to 300 seconds, covering multiple typical control behavior cycles, ensuring that the sampling time is long enough to include various typical responses, while avoiding the introduction of too many redundant states due to excessively long windows, thereby taking into account both timeliness and coverage. In the response segment extraction link, by performing boundary regularization and signal interception on the feedback data segment, and setting a 3% response amplitude threshold, only the continuously changing part with actual dynamic significance is retained, effectively avoiding the noise accumulation caused by intercepting invalid segments in the steady-state stage, and improving the dynamic feature density of the sample. The extracted dynamic segment is then encoded and packaged to construct a structured response curve segment sample, which is convenient for subsequent identification and classification processing. During the sample screening stage, by performing standard deviation analysis on the response samples of each channel combination within the standard action cycle window, and setting the lower limit of the standard deviation of the response amplitude change to 0.05 and the lower limit of the standard deviation of the response start delay to 0.3 seconds, channel samples with highly consistent response behaviors and insufficient discrimination can be filtered out, ensuring that the final constructed control feedback response sample set has sufficient fluctuation representativeness and delay characteristic changes, which helps to accurately identify actuator response links that are interfered with, mismatched or potentially aged, thereby providing high-value behavioral data support for subsequent control mode classification and fault warning.

[0030] Optionally, the control periodic sampling in step S2 is specifically:

[0031] Extract the control signal trigger timestamp of each actuator in the synchronous control trigger event sequence, and group the actuators by actuator number to calculate the average interval time of the control signals of each type of actuator, so as to set the initial control period;

[0032] Analyze the physical response characteristics of the actuator in combination with the control feedback response sample set to set the cycle buffer interval;

[0033] Setting an actuator control cycle window based on an initial control cycle and a cycle buffer;

[0034] Extract the continuous control event sequence of each actuator within the actuator control cycle window according to the control instruction-physical response relationship and generate a periodic event marker set;

[0035] For each action event in the periodic event marker set, the actuator parameter curve segment within the corresponding actuator control period window is extracted to construct a periodic control response sample;

[0036] The periodic control response samples are grouped and counted by actuator number, and the periodic sample segments with response standard deviation < 5% and actuator control times ≥ 30 are screened out to obtain the control period actuator parameter set.

[0037] The present invention can extract the average control interval of each type of actuator by statistically analyzing the control signal trigger timestamps and grouping them by actuator numbers, effectively reflecting the rhythm characteristics of control operations under different process tasks; setting the initial control period in this way not only avoids the incompatibility problem caused by subjective settings, but also improves the scenario adaptability of the model. In combination with the feedback response samples, the response delay, action amplitude and slow start characteristics are further extracted to set the cycle buffer, which can accommodate the response drift caused by physical inertia, load disturbance or communication fluctuation, and enhance the tolerance of the cycle window to the real action of the actuator. The control cycle window constructed on this basis has both dynamic adaptability and the integrity of the sample behavior, which significantly improves the analysis consistency of the data fragments. By extracting the event sequence and slicing the parameter curve within the cycle window, not only can a complete control behavior with a time logical relationship be obtained, but also the stability and continuity of the physical behavior of the actuator in the actual operating environment can be captured. Setting the response standard deviation to less than 5% as the fluctuation threshold and requiring the number of samples to be no less than 30 times can ensure that the screened period samples have significant reproducibility and statistical representativeness, and avoid isolated events or accidental disturbances affecting the subsequent standard template extraction; the role of these parameter settings is to take into account both response consistency and data coverage breadth, so that the final generated control cycle actuator parameter set becomes a high-quality foundation for stable behavior modeling and abnormal response identification.

[0038] Optionally, the standard action response mode recognition in step S2 is specifically:

[0039] The periodic control response samples in the control period actuator parameter set are time-normalized and uniformly mapped to a fixed length of 200 to generate a set of normalized response curves.

[0040] Based on the normalized response curve set, the rising edge slope, peak position, steady-state platform range and fall rate of each response curve are extracted to form a response feature vector set;

[0041] Similarity clustering analysis was performed on the response feature vector set, and the silhouette coefficient threshold was set to 0.65 to screen stable response clusters. The central curve in each cluster was extracted as the standard action response pattern.

[0042] By uniformly mapping all periodic control response samples to a fixed time length of 200, this method achieves time normalization of the response process while maintaining the behavioral profile structure. This allows execution behaviors under different control cycles to be quantified and compared on the same dimension, improving the uniformity of analysis standards and the structural alignment of samples. The length is set to 200 points, which, while balancing response resolution and computational efficiency, can meet the requirement for complete coverage of periodic data for most actuators at the current sampling rate. Key physical features of the response curve, such as the rising slope, peak position, steady-state platform range, and fallback rate, are further extracted. This not only covers the entire process from initial response to stabilization and termination, but also possesses clear physical meaning, accurately characterizing the actuator's responsiveness and control adaptability in various states. By constructing a set of response feature vectors and performing similarity cluster analysis, we can effectively identify response clusters that exhibit consistent behavioral patterns in dynamic environments, thereby uncovering representative stable response patterns. Setting a silhouette coefficient threshold of 0.65 within the clustering process ensures high similarity within each cluster and significant differentiation from other clusters, helping to eliminate marginal samples and atypical behaviors and improving the accuracy and robustness of the standard response pattern. The resulting extracted central curve serves as a standard action response template, providing a highly consistent and representative reference for subsequent real-time behavior matching, anomaly identification, and control optimization.

[0043] Optionally, the physical matching check in step S3 is specifically as follows:

[0044] Read the corresponding standard action response template according to the actuator number of the real-time actuator electronic control behavior data, extract the control signal trigger point and the corresponding physical response delay characteristics in the template, and obtain the control response data to be matched;

[0045] The calibration sliding window length was set to 300 seconds. The real-time actuator electronic control behavior data was segmented into sliding windows. The control signal trigger events and corresponding sensor response curve segments of each actuator within the calibration sliding window were extracted to construct the control-response sample unit to be verified.

[0046] Compare the control-response sample units to be verified with the control response data to be matched one by one, set the response delay deviation tolerance to 0.5s and the response amplitude deviation tolerance to 5%, and calculate the physical response matching degree of each set of samples;

[0047] Based on the physical response matching degree, sample units with a matching degree lower than 90% are selected from the control-response sample units to be verified, and recorded as response mismatch behaviors according to the timestamp and actuator number, and a summary report of the actuator physical action abnormality is generated.

[0048] This invention achieves process-level calibration of real-time behavior by extracting control trigger points and delay characteristics based on standard action response templates. This allows for an explicit dynamic correspondence between control signals and physical responses, enhancing the process-awareness of response monitoring. A sliding window length of 300 seconds helps cover most process execution cycles, avoiding sparse matching samples due to insufficient cycles while ensuring real-time system response. For matching analysis, a delay deviation tolerance of 0.5 seconds effectively identifies subtle response lags, while an amplitude deviation tolerance of 5% accounts for varying device sensitivities and sensor errors, maintaining the ability to identify response offsets while accommodating faults, thereby achieving high matching accuracy. A 90% matching screening threshold balances the risk of false positives and false negatives, ensuring that truly mismatched response samples are accurately labeled as anomalies and avoiding misclassification of fluctuating normal behavior as faults. The resulting anomaly report not only includes the time and serial number of the specific mismatched behavior but also provides a concrete, real-time reference for subsequent control decisions and actuator maintenance, enhancing the system's adaptability to dynamic operating conditions and the accuracy of safety control.

[0049] Optionally, step S4 is specifically:

[0050] Step S41: extracting the actuator number, abnormality type and triggering timestamp corresponding to the abnormal event in the actuator physical action abnormality report, and sorting them in chronological order to generate a time sequence chain table of abnormal events;

[0051] Step S42: establishing an actuator-process node mapping table based on the actuator physical action objects and process position identifiers recorded in the control instruction-physical response relationship pair;

[0052] Step S43: Combine the abnormal event time sequence chain table and the executor-process node mapping table to construct an abnormal impact propagation path diagram, set the maximum propagation depth to 3 layers, find the downstream process node path segment of the abnormal impact propagation path diagram, and record it as a potential abnormal path set;

[0053] Step S44: Filter the path segments in the potential abnormal path set that appear more than 3 times and have an impact window time span of less than 600 seconds, and output them as a process chain segment abnormal path table.

[0054] By extracting and sorting abnormal event numbers, types, and timestamps, this method achieves a structured representation of the dynamic evolution of faults, providing sequential logic support for subsequent impact analysis. Combined with the explicit actuator action locations and process step numbers in the control instruction-physical response relationship, this method accurately establishes a spatial mapping of physical actions, effectively eliminating the ambiguity inherent in traditional logic flow charts that replace actual action locations. By constructing an abnormality propagation path map and setting a maximum propagation depth of three layers, this method captures multi-level chain reactions while ensuring controllable propagation, preventing analysis distortion or invalid path tracing due to excessively long propagation chains. Furthermore, by extracting path segments with a frequency of more than three occurrences and a time span of less than 600 seconds, high-risk propagation patterns can be statistically identified. The frequency parameter ensures that abnormalities are not isolated, sporadic events, while the time window constraint ensures that the identified results are relevant to the actual process and have short-term linkage value. The resulting abnormality path table can serve as a basis for systematic traceability and control strategy adjustment, helping to improve the production system's responsiveness to micro-anomalies and the efficiency of cross-unit collaborative correction.

[0055] Optionally, step S5 is specifically as follows:

[0056] Step S51: Based on the process chain abnormal path table, the actuator number and the time interval of the abnormality corresponding to each abnormal path segment are extracted. The actuator parameter change data within adjacent abnormal intervals are sampled using a sliding window, with a window length of 120 seconds and a step length of 30 seconds, to form a response trend analysis data set.

[0057] Step S52: Perform linear fitting and second-order difference calculation on the continuous parameter change trend of each actuator in the response trend analysis data set to extract the response trend directionality characteristics and response fluctuation intensity characteristics. Set the response directionality threshold to ±5° and the fluctuation intensity threshold to 2.5% to filter out trend deviation events.

[0058] Step S53: grouping trend deviation events by actuator number and calculating the start / stop control priority score;

[0059] Step S54: setting the start-stop control parameter amplitude range according to the start-stop control priority score, wherein the control interval of the actuator with a high priority score is extended by 30 seconds, and the actuator with a low priority score maintains the original setting value, thereby generating an actuator start-stop control instruction table;

[0060] Step S55: perform structural verification on the actuator start-stop control instruction table, upload the verified actuator start-stop control instruction table to the lithium carbonate production line management platform, and complete the issuance of control task instructions.

[0061] By extracting the abnormal time periods of each actuator in the abnormal path and employing a sliding sampling method with a window length of 120 seconds and a step size of 30 seconds, this method effectively captures subtle trend changes in parameter evolution before and after the abnormality, improving the timeliness and data coverage accuracy of trend analysis. Furthermore, linear fitting and second-order difference methods are used to quantify the directionality and fluctuation amplitude of the control curve, respectively. Combined with a set response directionality threshold of ±5° and a fluctuation intensity threshold of 2.5%, this method accurately identifies hidden fluctuation events that exhibit abnormal trends but have not yet triggered physical anomalies, thereby providing proactive awareness of potential risks. Trend deviation events are clustered by actuator number and then assigned a priority rating for start-stop control. This facilitates the rational allocation and adjustment of resources within limited system resources and avoids system oscillation caused by blindly intervening in multiple actuators simultaneously. By extending the control interval of high-priority actuators by 30 seconds to buffer their response frequency, the risk of cumulative trend deviations amplified can be mitigated. Maintaining the original settings for low-priority actuators ensures system stability and continuity of the operating rhythm. The final generated start-stop control instruction table is uploaded to the management platform after structural verification, realizing closed-loop processing from trend identification to response issuance, effectively enhancing the production system's ability to dynamically adjust to trend anomalies and suppress disturbances of local anomalies on the overall rhythm.

[0062] Optionally, this specification also provides a battery-grade lithium carbonate production digital management system for executing the battery-grade lithium carbonate production digital management method as described above, the battery-grade lithium carbonate production digital management system comprising:

[0063] A relationship matching module is used to obtain actuator feedback data and actuator control data, and perform control signal-feedback relationship matching based on the actuator feedback data and actuator control data to obtain a control command-physical response relationship pair;

[0064] The response pattern recognition module is used to perform periodic sampling of the control instruction-physical response relationship to obtain a control period actuator parameter set; perform standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template;

[0065] The mismatch behavior identification module is used to obtain real-time actuator electronic control behavior data, perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify the response mismatch behavior, and obtain an actuator physical action abnormality report;

[0066] The impact path derivation module is used to derive the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report, and obtain the process chain segment abnormal path table;

[0067] The control instruction generation module is used to perform actuator action response trend analysis based on the process chain segment abnormal path table, and determine the start-stop control parameter amplitude based on the action response trend analysis results to obtain the actuator start-stop control instruction table; upload the actuator start-stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

[0068] The digital management system for the production of battery-grade lithium carbonate of the present invention can implement any digital management method for the production of battery-grade lithium carbonate of the present invention, and is used to combine the operation and signal transmission medium between various modules to complete the digital management method for the production of battery-grade lithium carbonate. The internal modules of the system cooperate with each other to improve the real-time and automation level of the production response, thereby ensuring the quality stability and energy efficiency output of high-end battery-grade lithium carbonate products. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0070] Figure 1 This is a schematic flow chart of the steps of the digital management method for battery-grade lithium carbonate production of the present invention;

[0071] Figure 2 Detailed step flow diagram of step S1 in the present invention;

[0072] Figure 3 Detailed flowchart of step S12 in the present invention;

[0073] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0074] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0075] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0076] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0077] To achieve this, please refer to Figures 1 to 3 The present invention provides a digital management method for battery-grade lithium carbonate production, the method comprising the following steps:

[0078] Step S1: acquiring actuator feedback data and actuator control data, and performing control signal-feedback relationship pairing based on the actuator feedback data and the actuator control data to obtain a control command-physical response relationship pair;

[0079] In this embodiment, on the lithium carbonate continuous production line, in order to ensure the monitoring of the control effect of key actuators (such as feed pumps, cooling valves, and heating electric heaters), the actuator control data (such as opening and closing instruction voltage values, current drive signals) and corresponding feedback signals (such as valve position sensor angle values, pressure feedback meter values, and thermistor sampling temperatures) are collected on site and recorded synchronously with a sampling period of 100ms. The control data and feedback signals are structurally paired through the device ID mapping table, and each record is compared and matched by sliding forward and backward on the time axis. The control delay threshold is set to no more than 2s, and the maximum allowable signal offset is set to no more than 3% to determine the actual physical response corresponding to the control behavior, and finally a structured set of control instructions and physical response relationships is formed, which is used as a basis for tracing the response consistency in downstream analysis.

[0080] Step S2: performing control periodic sampling on the control instruction-physical response relationship pair to obtain a control period actuator parameter set; performing standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template;

[0081] In this embodiment, for the obtained control instruction-physical response relationship data set, the control behavior fragments in the stable period under the daily operation state are selected for periodic extraction, and the control cycle window of each actuator is set to 64 seconds. The extracted content includes core indicators such as instruction duration, feedback signal rise / fall edge time, maximum response amplitude, recovery time, etc., to form an action cycle parameter matrix. Multiple working condition sampling methods (such as conventional feeding, temperature start-up, evaporation section exhaust, etc.) are used for feature recognition, and the aggregate statistical method is used to form a periodic response parameter distribution. Taking the median as the response reference benchmark, the fluctuation range is limited to ±15%, and a standard response curve template is constructed to summarize the standard characteristics of typical control behaviors under different process conditions for future matching and verification of real-time control behaviors.

[0082] Step S3: Acquire real-time actuator electronic control behavior data, and perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify response mismatch behavior, and obtain an actuator physical action abnormality report;

[0083] In this embodiment, the real-time electronic control behavior data of the actuator is collected through the industrial field bus, including two types of data: control signal input value and feedback sensor response value, and the data sampling frequency is 20Hz. Combined with the generated standard action response template, a curve fitting comparison is performed on each real-time sampling record. The response deviation threshold is set to a feedback amplitude difference of 10% and a delay difference of more than 1.5 seconds as an abnormality identification condition. When the actual feedback does not reach 80% of the standard response amplitude within the instruction response interval or there is an obvious response lag, the system will mark it as "action response inconsistency". All mismatch records are classified according to the actuator number, and an abnormality report is generated, including information such as the number of mismatches, the average deviation of the mismatch, and typical abnormal behavior segments, which serve as the trigger basis for subsequent process path judgment.

[0084] Step S4: deriving the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report to obtain a process chain segment abnormal path table;

[0085] In this embodiment, based on the control instruction-physical response relationship pair constructed in the previous stage and the actuator abnormal response report, combined with the physical position of the actuator control object in the process flow chart and the corresponding equipment connection relationship, an abnormal impact transmission map is established. After identifying the actuator with an abnormal response, the physical equipment path controlled by it is traversed in the downstream direction from it as the starting point, such as a continuous process chain segment composed of "feeding control valve → mixing tank → metering pump → flow meter". Each path is marked with the direct connection between the actuator and the process unit it controls. If there are more than two consecutive abnormal feedback nodes in the path, the path is classified as a high-risk abnormal segment, forming a structured process abnormal segment table, which is used to evaluate the impact range of the actuator problem on the entire process.

[0086] Step S5: Perform actuator action response trend analysis based on the process chain segment abnormal path table, and determine the start-stop control parameter amplitude based on the action response trend analysis results to obtain the actuator start-stop control instruction table; upload the actuator start-stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

[0087] In this embodiment, based on the abnormal path of the obtained process chain segment, a comprehensive analysis is performed on the historical control behavior and recent response trend of the corresponding actuator. Based on the action response time series within 24 hours, the response amplitude, instruction frequency and recovery stability of each time segment are partitioned and statistically analyzed. If the actuator response behavior shows a continuous decrease in response amplitude or a significant increase in instruction triggering frequency (more than 30% of the daily average) in multiple consecutive cycles, a start-stop control recommendation will be triggered. The control instruction table will clearly state the recommended status of each abnormal actuator (such as temporary deactivation, frequency reduction operation or forced restart), and will be accompanied by parameters such as the recommended reduction ratio (for example, the control frequency drops to 60% of the original set value) and the estimated recovery time (such as monitoring and re-evaluation within 60 minutes). All instructions are uploaded to the host computer system through the field bus and enter the task scheduling link of the management platform as input instructions for automatic control tasks.

[0088] Optionally, step S1 specifically includes:

[0089] Step S11: collecting actuator control signals and corresponding feedback signals to obtain actuator feedback data and actuator control data;

[0090] In this embodiment, a high-frequency sampling controller is deployed in the actuator system supporting the lithium carbonate calcination line and the crystallization section to collect control signal data from devices such as electric valves, frequency converters, and proportional control drivers, while synchronously reading feedback signal source data, including output values ​​of sensors such as angular displacement encoders, thermocouples, and Hall ammeters. The acquisition cycle is set to 50ms to ensure that the dynamic characteristics of the control instruction changes are captured, and the data cache window is set to 10 seconds for stable timing storage. The control signal is indexed by the clock signal of the control system, and the feedback signal is recorded according to the timestamp of the on-site acquisition node. After the acquisition is completed, the two data sets are uniformly formatted into a "device ID-timestamp-signal value" structure as the basic data for subsequent processing.

[0091] Step S12: performing channel mapping and synchronous pairing on the actuator feedback data and the actuator control data to generate a control feedback channel binding table;

[0092] In this embodiment, to accurately correlate actuator control signals with their feedback signals, the device coding rules in the field device configuration document and loop wiring diagram are used to bind and map the collected data by channel number. Control-feedback pairs are matched using the actuator logical ID and sensor physical interface number as unique identifiers. The matching process is configured with a time offset of no more than 500ms to address feedback delays caused by physical wiring variations. After the initial matching is complete, the data continuity and response consistency of each control-feedback channel pair are verified to eliminate isolated, unresponsive control records. A unified control-feedback channel binding table is generated. This table records the control source, feedback source, signal type, and physical location of each channel pair for subsequent periodic data capture and response behavior analysis.

[0093] Step S13: extracting feedback feature segments within a preset action cycle based on the control feedback channel binding table to form a control feedback response sample set;

[0094] In this embodiment, according to the channel binding table, the actuator action cycle is selected as the basic analysis unit, and the feedback response segment is extracted from the synchronous pairing data according to the set cycle window. The default cycle length is set to 8 seconds, and the control signal edge trigger point is selected as the start point in the window, and the maximum response amplitude of the feedback signal is selected as the end reference point, and the data segment in this process is intercepted. The extracted feedback segment must meet the complete start and end response characteristics, including parameters such as initial response delay, maximum response amplitude, and stable maintenance time. The characteristic dimension information such as response delay time, change speed, waveform trend, etc. is recorded for each segment to form a feedback behavior data segment set. Finally, a control feedback response sample set is constructed to further identify the characteristic changes and response mechanisms of the control behavior.

[0095] Step S14: performing control mode identification on the control feedback response sample set to obtain a controller control mode sample set;

[0096] In this embodiment, each feedback segment in the response sample set is grouped and analyzed according to the control channel number and control signal form. By extracting control features such as the duration, intensity level, trigger edge form of the control signal, as well as dynamic indicators such as the corresponding delay, amplitude change and response time window of the feedback segment, the operation style imposed by the controller on the actuator is identified. For example, for a certain electric valve actuator, if its control signal duration is more than 4 seconds and the feedback signal shows a slow linear upward trend, it can be summarized as a "steady-state control" mode. The system also records the frequency of control signal interruption and the proportion of abnormal feedback responses to assist in distinguishing control styles such as "intermittent start and stop" and "reverse drive", and finally forms a controller control mode sample set, which provides a basis for actuator classification and behavior classification.

[0097] Step S15: Classify and archive the controller control mode sample set according to actuator type and control mode, and generate control instruction-physical response relationship pairs.

[0098] In this embodiment, based on the generated controller control mode sample set, each record is labeled with the actuator category (such as solenoid valve, proportional valve, variable frequency pump, etc.), and the sample data is structured and archived using the control mode label (such as fast opening and closing, stable maintenance, and repeated adjustment) as the classification standard. The control responses of the same type of actuators in different working sections are grouped, and key indicators such as their action cycle consistency, control signal amplitude stability, and feedback response standard deviation are counted. Sample pairs that meet the action response consistency of more than 90% and the feedback error of less than 10% are automatically included in the standard control response relationship set and registered as a control instruction-physical response relationship pair. The relationship pair structure includes device ID, control mode type, typical response parameter range, etc., which can be directly used for abnormal diagnosis and action matching analysis.

[0099] Optionally, step S12 is specifically as follows:

[0100] Step S121: Retrieve the control output port corresponding to each actuator and the acquisition input channel configuration document of the physical feedback feedback device through the lithium carbonate production line management platform, combine the obtained equipment wiring diagram and on-site wiring labels, establish the actuator-feedback device physical connection relationship, and thus obtain a physical channel index table;

[0101] In this embodiment, in the actuator network of the lithium carbonate calcination system and the cooling and dust removal section, the on-site configuration data is accessed through the production line management platform, and the configuration entries of the control output port of each electric actuator (such as the angle stroke valve, the speed regulating fan) and its corresponding feedback signal acquisition channel are extracted, and the cable numbers and channel labels recorded in the actual wiring diagram are checked one by one. In order to ensure the accuracy of the mapping, the port of the key actuator (such as the graphite nozzle electronic control unit) is measured by physical inspection to verify the consistency of the platform configuration and wiring. In each pair of physical connection relationships, the control output channel number, signal type (voltage or current), feedback acquisition channel position, actuator model and sampling interface type are clearly recorded, and finally integrated to generate a structured physical channel index table to constrain the subsequent data synchronization operation range.

[0102] Step S122: Set the synchronization sliding time window to 0.5s, and use the physical channel index table to perform control signal-feedback signal synchronization window slicing to obtain aligned data segment pairs;

[0103] In this embodiment, after obtaining the complete physical channel index table, the sliding time window parameter is set to 0.5 seconds according to each group of actuator control-feedback channel pairs, and the continuously recorded data stream is intercepted by sliding window. The control signal serves as the dominant time axis, and the feedback signal is synchronized and matched according to the nearest neighbor principle. In each time window, a group of control-feedback signal pairs are extracted as slice samples. Each group of samples must include the control signal change point, the feedback signal delay response trend and its peak response point to ensure that it can reflect the actual control process. In order to enhance the synchronization accuracy, the segments with a maximum difference of ±50ms in the timestamp are time-calibrated and marked as "timing offset correction segments". After the slicing is completed, all data segments are numbered and archived to form a set of aligned data segment pairs with consistent structure and time synchronization, which serves as the basic data for the next step of stability screening.

[0104] Step S123: Periodically reorganize the aligned data segment pairs and verify feedback direction consistency, remove channel pairs with feedback value noise exceeding ±10% or missing points, and generate a stable control response data set;

[0105] In this embodiment, based on the set of synchronously aligned data segment pairs, the slices are reorganized into a periodic control response sequence according to the action cycle, and the trend consistency of the feedback signal in each cycle is judged. Specifically, the direction of change of the feedback signal in each cycle is calculated and compared with the expected response direction of the control signal. If the direction is inconsistent or the fluctuation is too large, it is marked as an invalid cycle. For valid cycles, fluctuation detection is performed based on the standard deviation of the feedback amplitude. If the noise level exceeds ±10% or there is a null point in the cycle, the channel pair is eliminated. In this way, only channel data pairs with small fluctuations in the feedback curve and accurate direction matching are retained. Finally, a stable control response data set is formed by screening. Each record in the data set retains characteristic parameters such as cycle number, control value, feedback trend curve, and stable duration, which are used for subsequent normalization and standard response modeling.

[0106] Step S124: normalize and organize the stable control response data set, and construct a control feedback channel binding table in the format of actuator number, control signal geology, feedback signal geology, feedback delay mean, and response amplitude mean.

[0107] In this embodiment, based on the stable control response data set, feature normalization processing is performed on each valid data segment. First, the signal units are unified, and the control output and feedback signal are converted into standardized voltage values ​​(0-10V range), and the time data is normalized to millisecond accuracy. Secondly, the actuator equipment is classified according to the actuator device number, and the delay mean of the feedback signal in each category (that is, the time difference from the control change to the start of the feedback response) and the average value of the feedback amplitude (referring to the amplitude difference between the feedback response peak and the baseline) are extracted. The change amplitude of the control signal and the response trend are combined to construct a five-dimensional structured record. Finally, each record is written into the control feedback channel binding table in the form of "actuator number-control signal geology-feedback signal geology-feedback delay mean-feedback amplitude mean", providing a unified data input structure for the subsequent response template matching and fault identification process.

[0108] Optionally, step S13 is specifically as follows:

[0109] Step S131: Read each pair of control-feedback channel combinations in the control feedback channel binding table, set the standard action cycle window to 300s, and extract the synchronous control trigger event sequence;

[0110] In this embodiment, each pair of control-feedback channel combinations in the scheduling control feedback channel binding table is used to read the recorded valid channel index and signal type information. In the specific operation, a group of actuator control channels and feedback channel combinations (such as the electric door control signal and displacement feedback channel for calcining furnace feed execution) are selected, and the standard action cycle window is set to 300 seconds in the recorded continuous data stream. Within this window, each rising edge or falling edge triggered by the control signal is identified as the starting point of the action, and its corresponding timestamp is extracted and synchronously recorded in the control trigger event sequence. During the screening process of synchronous control events, it is necessary to ensure that the time interval between two adjacent events is not less than 30 seconds to exclude misjudgment actions caused by repeated triggering or jitter. After the extraction is completed, each control trigger event record contains the starting time, the direction of change of the control signal amplitude and the corresponding feedback channel index, laying the foundation for subsequent feedback interception.

[0111] Step S132: performing boundary regularization and signal interception on the feedback data segments corresponding to each group of control trigger events in the synchronous control trigger event sequence, and setting the response amplitude threshold to 3% to extract the segments with continuous signal changes;

[0112] In this embodiment, in the extracted control trigger event sequence, the feedback channel data segment corresponding to each event record is processed in turn. Taking the event trigger time as the starting point, the feedback signal data is extracted within 5 consecutive seconds thereafter, and the abnormal lack of sampling points are corrected by linear interpolation. In order to ensure the accuracy of the signal interception boundary, 0.5 seconds are reserved before and after the trigger time point as an observation area to prevent the response curve from being truncation. The criterion for significant change in feedback response is set to a relative change in amplitude of not less than 3%, which is used as the starting standard for identifying the continuous change segment of the signal. After detecting the rising or falling segment of the signal that meets the conditions, its complete fluctuation segment is extracted in turn, and the end point of the interception is defined as the point where the amplitude is stably maintained within the range of ±1% for 0.8 seconds.

[0113] Step S133: Encode and encapsulate the continuously changing signal segments to generate response curve segment samples;

[0114] In this embodiment, after completing the interception of the effective segment of the feedback signal, each continuously changing segment of the signal is structurally encapsulated. The encapsulation content includes: response starting point time, end point time, maximum amplitude, amplitude increase or decrease direction, duration, delay time between the starting point and the maximum value and other parameters. At the same time, the signal segment is normalized into a standard sampling point sequence of length 100 for subsequent comparison and matching. The corresponding channel number and trigger event ID are also embedded in each segment of the code to ensure traceability during the data reconstruction stage. The encoding and encapsulation process uses a unified template format for data filling, and performs a three-point sliding average smoothing process on the signal curve segment with sudden spikes to eliminate the influence of occasional glitches. The response curve segment samples finally generated are saved under an index structure named after the channel group number, which is convenient for batch aggregation and archiving.

[0115] Step S134: Summarize the response samples of each channel combination within the standard action cycle window, eliminate the channel combinations and corresponding samples with a response amplitude change standard deviation <0.05 and a response start delay standard deviation <0.3s, and obtain a control feedback response sample set.

[0116] In this embodiment, after completing the curve segment encapsulation, for each group of control-feedback channel combinations within the standard action cycle window, all the response samples generated by it are summarized, and the standard deviation of the response amplitude change and the standard deviation of the delay from the response starting point to the initial change point in each group of samples are counted. In order to eliminate channel combinations with insufficient response stability or excessive jitter, the lower limit of the amplitude change standard deviation is set to 0.05, and the lower limit of the delay standard deviation is set to 0.3 seconds. If the sample group generated by a channel combination within the period meets one of the above two conditions, it means that its response behavior is not representative or the volatility is too high and needs to be eliminated. The retained channel combination is deemed to have stable response characteristics, and the relevant samples are archived as a control feedback response sample set. This sample set will serve as the core input basis for subsequent standard response template extraction and anomaly identification to ensure high-confidence data support when building the model.

[0117] Optionally, the control periodic sampling in step S2 is specifically:

[0118] Extract the control signal trigger timestamp of each actuator in the synchronous control trigger event sequence, and group the actuators by actuator number to calculate the average interval time of the control signals of each type of actuator, so as to set the initial control period;

[0119] In this embodiment, the trigger timestamp of each actuator control signal is extracted from the actuator trigger record log, and special attention is paid to whether there are jumps, jitters or repeated records in the timestamp change trend. The unique number of the actuator is used as the grouping basis, and the adjacent time differences between the timestamps of each group of control signals are counted separately. The time interval sequence is subjected to a one-time sliding filter process to eliminate short-period outliers with an interval of less than 0.3 seconds. Based on the stable data, the average interval time of the control signal of each type of actuator is calculated separately, such as the average control interval of the feeding screw mechanism is 46.5 seconds, and the average interval of the liquid valve actuator is 18.2 seconds. Based on the statistical results, an initial control cycle is defined for each type of actuator, and the upper and lower limits of the fluctuation range are set to no more than ±10% as the basic basis for cycle evaluation.

[0120] Analyze the physical response characteristics of the actuator in combination with the control feedback response sample set to set the cycle buffer interval;

[0121] In this embodiment, the previously constructed control feedback response sample set is used to analyze the physical response characteristics of various actuators after the control is triggered. The feedback start delay, maximum amplitude response time, and signal stabilization time point of each sample are extracted, and its maximum time span within the sample group is calculated. The sum of the maximum value of the feedback response start delay and the average value of the maximum amplitude lag time is used as the typical delay range of this type of actuator. For example, for an electromagnetically driven actuator, the maximum start delay is 0.7 seconds and the average response time is 1.3 seconds, so the cycle buffer is set to 2.0 seconds. This buffer is used to expand the cycle boundary in subsequent window division to ensure that all response processes are fully covered and to avoid truncation of the response band.

[0122] Setting an actuator control cycle window based on an initial control cycle and a cycle buffer;

[0123] In this embodiment, actuator control cycle windows are established based on the initial control cycle and the set periodic buffer interval. Based on the control signal trigger point, a time window of length (initial control cycle + 2 × buffer interval) is generated for each actuator. For example, for an actuator with a 60-second control cycle and a 2-second buffer interval, the periodic window length is set to 64 seconds, with a 2-second buffer interval at the front and back. Window partitioning is performed using a sliding overlap method, with a window sliding step of no less than half the period length. This ensures overlap between consecutive events, allowing for handling overlapping control command scenarios. This ultimately creates a periodic time partitioning structure for each actuator.

[0124] Extract the continuous control event sequence of each actuator within the actuator control cycle window according to the control instruction-physical response relationship and generate a periodic event marker set;

[0125] In this embodiment, in each constructed actuator control cycle window, the control instruction-physical response relationship is used to filter all control signal records within the time period to extract a complete and continuous sequence of control events. The control instruction number, channel identifier, trigger time and corresponding feedback channel mapping information of each event are recorded. By marking these events in the order of occurrence and organizing them into a periodic event tag set in combination with the actuator number, each event tag is accompanied by the control signal edge type (rising or falling) and the object of action. For example, for the electric valve numbered V-21, its control cycle window contains 5 valid control signals, and each event is uniformly named and archived in the form of tags such as V-21_E1 and V-21_E2 as the basis for locating the response sample.

[0126] For each action event in the periodic event marker set, the actuator parameter curve segment within the corresponding actuator control period window is extracted to construct a periodic control response sample;

[0127] In this embodiment, with the periodic event marker set as a reference, the actuator feedback parameter records in the periodic window where each action event is located are retrieved one by one, the feedback parameter change curve from 1 second before the event occurrence point to 5 seconds after the event occurrence point is extracted, and the signal segment is uniformly sampled to 200 data points. The extracted data includes feedback displacement, pressure, current and other indicators, which are classified according to the actuator type. In this process, the signal is zero-offset corrected to ensure that the starting position of the curve is unified. Each response curve is named with the event number and saved in the periodic sample set to establish a periodic control response sample table with a unified structure. This sample table is subsequently used for normalized response feature evaluation and stability screening, and the total number of samples is maintained at no less than 500 to meet the statistical validity requirements.

[0128] The periodic control response samples are grouped and counted by actuator number, and the periodic sample segments with response standard deviation < 5% and actuator control times ≥ 30 are screened out to obtain the control period actuator parameter set.

[0129] In this embodiment, the periodic control response sample table is classified and summarized, and the standard deviation of the response amplitude change of each type of actuator sample is counted according to the actuator number, and whether the number of controls exceeds 30 times is counted. The selection criteria are set as follows: the standard deviation of the response amplitude change is less than 5%, and the number of samples is not less than 30. In the screening results, if it is found that the pneumatic push rod actuator numbered P-11 has a total of 33 samples that meet this condition, and its amplitude change standard deviation is 3.7%, then this type of sample is included in the control period actuator parameter set. All sample segments that meet the standards are structurally normalized and numbered and archived to provide a benchmark template library for subsequent response template matching and anomaly identification. The final form of the parameter set includes fields such as the standard response average curve, typical response delay, and sample quantity statistical information.

[0130] Optionally, the standard action response mode recognition in step S2 is specifically:

[0131] The periodic control response samples in the control period actuator parameter set are time-normalized and uniformly mapped to a fixed length of 200 to generate a set of normalized response curves.

[0132] In this embodiment, all periodic control response samples are extracted from the control period actuator parameter set, and each response curve is time-normalized. In the processing flow, the original curve segment is first linearly interpolated on the time axis, and the response curves of different lengths are uniformly mapped to a fixed-length format of 200 sampling points. During the interpolation process, the position corresponding to the initial trigger point of the response is retained as the reference starting point to ensure that the rising edges of different curves are aligned. During the normalization process, an endpoint equidistant interpolation method is adopted, while maintaining the continuity and amplitude change trend of the original signal waveform, and prohibiting signal amplitude scaling. All normalized response curves are uniformly archived according to the actuator number, and a normalized response curve set is constructed for subsequent response pattern recognition analysis.

[0133] Based on the normalized response curve set, the rising edge slope, peak position, steady-state platform range and fall rate of each response curve are extracted to form a response feature vector set;

[0134] In this embodiment, based on the set of normalized response curves, four types of key dynamic response indicators are calculated for each normalized curve. First, the slope change rate from the start to the maximum value of the curve is extracted as the rising edge slope, and the threshold is set to the average slope between the starting point and the peak point at 10% of the maximum amplitude; then the sampling point index of the peak position is determined, and its position ratio in the 200-point normalized curve is recorded; then, a sliding window method is used to identify the sections where the median fluctuation of the curve is within the range of ±2% and lasts for more than 15 sampling points, and mark them as the steady-state platform range; finally, the overall descent rate from the peak point to the end point of the curve is calculated, and the normalized rate is estimated according to the time length corresponding to each 10% decrease in amplitude. The above four values ​​are combined into a response feature vector to construct a complete set of response feature vectors. Each feature vector is accompanied by the corresponding actuator number and sample ID information for subsequent pattern division.

[0135] Similarity clustering analysis was performed on the response feature vector set, and the silhouette coefficient threshold was set to 0.65 to screen stable response clusters. The central curve in each cluster was extracted as the standard action response pattern.

[0136] In this embodiment, the constructed response feature vector set is used to cluster and group according to the similarity of the response dynamic characteristics. In the clustering process, the distance metric between each group of response feature vectors is measured using the standardized Euclidean space scale, and the silhouette coefficient threshold is set to 0.65 as the stable cluster screening criterion. If the overall silhouette coefficient of a cluster is lower than the threshold, it is considered to have an unstable internal structure and is eliminated; and the cluster with a silhouette coefficient greater than or equal to 0.65 is retained, and a normalized response curve whose response feature is closest to the mean of all samples is extracted from it as the central curve of the cluster. The central curve is saved as a type of standard action response pattern, and the corresponding actuator category and control event type are marked. Finally, a standard response pattern library is established, and each type of pattern records its parameters such as curve structure, characteristic index mean, sample number and fluctuation range, which are used as a comparison benchmark for subsequent action recognition and response deviation monitoring.

[0137] Optionally, the physical matching check in step S3 is specifically as follows:

[0138] Read the corresponding standard action response template according to the actuator number of the real-time actuator electronic control behavior data, extract the control signal trigger point and the corresponding physical response delay characteristics in the template, and obtain the control response data to be matched;

[0139] In this embodiment, after receiving the real-time actuator electronic control behavior data, the standard response pattern established in the action response template library is first called based on the actuator number in the data. For each actuator number, the control signal trigger time point marked in the corresponding standard template, as well as key dynamic features such as the average response delay, initial response sampling point, and stable segment start and end range recorded in the corresponding physical response curve are extracted. This extraction process ensures the use of a reference template that strictly matches the current actuator model and control instruction type. After the extraction is completed, the key points of these standard action response templates are combined into a unified data unit as a reference feature for subsequent real-time data matching. This data unit contains content such as the relative time of the control signal trigger point, the mean response delay, the amplitude peak position of the response waveform, and the platform segment range.

[0140] The calibration sliding window length was set to 300 seconds. The real-time actuator electronic control behavior data was segmented into sliding windows. The control signal trigger events and corresponding sensor response curve segments of each actuator within the calibration sliding window were extracted to construct the control-response sample unit to be verified.

[0141] In this embodiment, the sliding time window length is set to 300 seconds, and the data stream is segmented. The sliding window advances in steps of 100 seconds, and all control signal trigger events are extracted within each sliding window and classified according to the actuator number. The feedback curve segment corresponding to each trigger event is extracted by reading the synchronous sensor data and setting the interception range to 10 seconds after the trigger, extracting the signal segment and retaining the original sampling point distribution. To ensure the integrity of the data structure, each control-response sample unit includes fields such as the control signal trigger timestamp, the original segment of the response curve, the feedback sensor number, the response start point and end point, and is marked with the sliding window number to which it belongs. All constructed sample units are buffered in the memory and indexed by the actuator number for subsequent one-by-one comparison operations.

[0142] Compare the control-response sample units to be verified with the control response data to be matched one by one, set the response delay deviation tolerance to 0.5s and the response amplitude deviation tolerance to 5%, and calculate the physical response matching degree of each set of samples;

[0143] In this embodiment, for each control-response sample unit in the buffer, the standard action response template that matches the corresponding actuator number is called in turn, and a structured comparison is performed between the two. During the comparison process, the maximum tolerance deviation of the response delay is set to 0.5 seconds, and the maximum tolerance deviation of the response amplitude is set to 5%. For each group of samples, the start-up time of the feedback signal after the control signal trigger point is first compared with the template response delay value to calculate the time deviation; then the maximum response value of the sampling segment is compared with the template peak amplitude to evaluate the amplitude difference. The above two deviation indicators are normalized and calculated separately and then combined to form a physical response matching index. The matching value is expressed in percentage and recorded in each sample unit structure. All calculation operations are completed based on actual numerical operations and do not involve inference structures.

[0144] Based on the physical response matching degree, sample units with a matching degree lower than 90% are selected from the control-response sample units to be verified, and recorded as response mismatch behaviors according to the timestamp and actuator number, and a summary report of the actuator physical action abnormality is generated.

[0145] In this embodiment, based on the generated physical response matching results, all control-response sample units within the sliding window are screened. Samples with a matching degree below 90% are considered abnormal response records, and their original control event timestamp, actuator number, actual response delay, amplitude deviation value, and corresponding standard template number are retained. All abnormal records are sorted chronologically and output as a response mismatch behavior log. This log is further categorized and organized by actuator number to form a final actuator physical action anomaly report. The report includes the time of the anomaly occurrence, a detailed list of abnormal response indicators, the affected actuator number, the matching template number, and a preliminary determination of the abnormal behavior, serving as a basis for further on-site verification and control adjustments by operations and maintenance personnel. If a single actuator experiences more than three mismatches within 24 hours, the system automatically marks it as a key monitoring device and places it in the monitoring priority queue.

[0146] Optionally, step S4 is specifically:

[0147] Step S41: extracting the actuator number, abnormality type and triggering timestamp corresponding to the abnormal event in the actuator physical action abnormality report, and sorting them in chronological order to generate a time sequence chain table of abnormal events;

[0148] In this embodiment, in the generated actuator physical action anomaly report, each record includes the actuator number, the anomaly category (such as response delay exceeding the limit, amplitude mismatch), and the timestamp of the anomaly occurrence. The system calls the exception log parsing component to read all exception entries in sequence, and uses the timestamp field as the main sorting basis to arrange the records in ascending order to build a complete time chain table structure. During the generation process, the system sets the time format to "yyyy-MM-dd HH:mm:ss" to ensure accuracy to the second level, and merges and marks the actuator numbers of repeated records to retain the latest anomaly type and its time point. The final generated abnormal event time series chain table is cached in the analysis engine memory in the form of a JSON structure. Each entry includes three fields: number, anomaly category, and timestamp, for subsequent linkage with process node mapping.

[0149] Step S42: establishing an actuator-process node mapping table based on the actuator physical action objects and process position identifiers recorded in the control instruction-physical response relationship pair;

[0150] In this embodiment, a control instruction-physical response relationship table is read. This table details the physical location of each actuator, the equipment unit to which it belongs, and the process node number to which it belongs. By extracting the actuator number field and the process node number field from this table, a mapping relationship set is constructed. To ensure the accuracy of the mapping results, the system confirms the physical connection location of each actuator by comparing it with the process equipment topology diagram and establishes a unified naming convention for the process node numbers, such as using "CLEAN-01" and "DRY-04" to identify each process node. In the process of generating the actuator-process node mapping table, actuators with one-to-many mappings are recorded with multiple node numbers, and a "physical connection priority" field is added to reflect the primary and secondary active nodes. This mapping table is stored in a tabular structure, which supports reverse query of the process node by actuator number.

[0151] Step S43: Combine the abnormal event time sequence chain table and the executor-process node mapping table to construct an abnormal impact propagation path diagram, set the maximum propagation depth to 3 layers, find the downstream process node path segment of the abnormal impact propagation path diagram, and record it as a potential abnormal path set;

[0152] In this embodiment, the abnormal event sequence list and the executor-process node mapping table are called. For each abnormal event, the corresponding process node is first searched according to the executor number. Then, a directed path graph is constructed according to the process flow topology graph configured in the workshop control system, with the process nodes as node units in the graph and the ordered connection relationships between nodes as edges. Starting from the process node where the abnormality occurs, the node connection path of up to 3 levels is searched in the downstream path in sequence, and all node numbers in the path segment and the corresponding propagation path sequence are recorded. Each path segment must contain the triggering abnormal node and its subsequent 1 to 3 directly connected process nodes. The system records the path content in a nested list structure and assigns a unique identification number to each path. The constructed propagation path graph forms a set of potential abnormal paths, which is used to determine the possible transmission range of the abnormal impact.

[0153] Step S44: Filter the path segments in the potential abnormal path set that appear more than 3 times and have an impact window time span of less than 600 seconds, and output them as a process chain segment abnormal path table.

[0154] In this embodiment, for the generated set of potential abnormal paths, the statistical window is set to 24 hours, and the frequency of occurrence of each path segment is counted. If a path segment is triggered by an abnormal event more than 3 times within 24 hours, and the corresponding abnormal event time span is within 600 seconds, the path segment is marked as a high-frequency abnormal channel. To this end, the start time and the end time in each path segment are differenced to ensure that the time span is accurately determined. At the same time, in order to exclude the virtual high frequency of the path segment caused by repeated abnormal records, the timestamp sequence of the path segment needs to be filtered for non-repeating identification. Finally, the path segments that meet the dual conditions of frequency of occurrence and time window are output as a process chain segment abnormal path table, and are presented in the format of "abnormal path number-path content-abnormal frequency-time window span" for reference by maintenance personnel and for investigation and processing.

[0155] It is particularly important that the downstream process node path segment of the abnormal impact propagation path diagram in step S43 is specifically:

[0156] Extract the execution units with sharp changes in controller feedback signals and significant start-stop response lags in the abnormal impact propagation path diagram as abnormal root nodes, and mark the abnormal root nodes as propagation entry points;

[0157] In this embodiment, when processing the abnormal impact propagation path diagram, the feedback signal curve of each execution unit in the path diagram is first extracted, and the maximum change rate of the feedback signal within the abnormal time window is calculated. If the feedback signal changes by more than 10% within 1 second, and the corresponding start-stop action response time exceeds twice the expected control cycle of 200ms, the execution unit is determined to be the root node of the abnormal behavior. This type of node must meet both the "violent feedback fluctuation" and "obvious action lag" conditions before it will be marked as a propagation entry point. For each marked entry point, its number, process node, feedback characteristic value and other information are recorded, and it is used as the starting node of the downstream path scan and saved in the abnormal propagation entry index table.

[0158] Using the abnormal impact propagation path graph, traverse the downstream process nodes connected to the propagation entry point, and record the path segment characteristics from the propagation entry point to the downstream process nodes to obtain the downstream reachable node path set;

[0159] In this embodiment, starting from the marked propagation entry point, the process control topology diagram is called to traverse all downstream process nodes directly or indirectly connected to the entry point step by step. Whenever a complete path segment from the entry point to a downstream node is identified, the node numbers, path levels, and connection characteristics between nodes in the path segment are recorded, such as the control transfer direction, execution trigger type, etc. In order to improve the recognition efficiency, the maximum path depth is set to 5 layers, and the minimum length of the path segment is set to no less than 2 nodes. All qualified path segments are summarized into the downstream reachable node path set, and a unique index number is assigned to each path segment for subsequent path characteristic analysis.

[0160] The path strength of each path segment in the set of downstream reachable node paths is evaluated, and the lower limit threshold of the path strength is set to 0.6 to screen the path segments and obtain the candidate table of abnormal propagation key paths;

[0161] In this embodiment, for the obtained set of downstream reachable node paths, a path strength value is calculated for each path segment. The strength value is obtained by weighting indicators such as the controller connection density in the path segment, the physical influence weight between actuators, and the trigger frequency. During the calculation process, the lower limit threshold of the path strength is set to 0.6. If the comprehensive strength of a path segment is lower than this value, the path will be eliminated from the candidate set. The strength assessment needs to be combined with the trigger data frequency and the physical connection model within the past 3 days to ensure that the path segment assessment is both real-time and stable. After the screening is completed, a candidate table of abnormal propagation key paths is generated, in which each path segment is accompanied by a strength value and a connection link description for priority sorting.

[0162] The delay information of each edge in the propagation path in the abnormal propagation critical path candidate table is summed up, and the total propagation delay of the actuator-process node mapping table is used as the main sorting indicator to perform priority sorting, and high-priority paths are screened as potential abnormal paths, thereby obtaining a potential abnormal path set.

[0163] In this embodiment, in the abnormal propagation critical path candidate table, the control instruction propagation delay information between each node in each path segment is extracted. The delay data of each hop is provided by the executor-process node mapping table, and the delays of all edges in the same path segment are accumulated item by item to obtain the total propagation delay of the path. Subsequently, all candidate paths are prioritized from high to low according to the total propagation delay, giving priority to abnormal channels with fast impact and fast feedback. Finally, the path segments with the top 20% of the total propagation delay ranking are set as high-priority paths, and a set of potential abnormal paths is screened and generated. Each high-priority path is accompanied by a propagation entry number, a path node sequence, a total propagation delay, and a propagation strength value, forming a complete data structure submitted to the abnormal processing link.

[0164] Optionally, step S5 is specifically as follows:

[0165] Step S51: Based on the process chain abnormal path table, the actuator number and the time interval of the abnormality corresponding to each abnormal path segment are extracted. The actuator parameter change data within adjacent abnormal intervals are sampled using a sliding window, with a window length of 120 seconds and a step length of 30 seconds, to form a response trend analysis data set.

[0166] In this embodiment, after receiving the abnormal path table for each process chain segment, the system retrieves the actuator's historical work logs and sensor monitoring data based on the actuator number and annotated start and end time of the abnormality for each abnormal path segment. For each actuator, a 120-second sliding analysis window is set, extending 60 seconds forward and backward from the abnormal interval, with a step size of 30 seconds. Control signals and feedback parameter data are sampled, including current, voltage, position deviation, and execution status bits. The sampled data is stored in JSON format and archived by actuator number to form a response trend analysis dataset for subsequent execution status trend assessment.

[0167] Step S52: Perform linear fitting and second-order difference calculation on the continuous parameter change trend of each actuator in the response trend analysis data set to extract the response trend directionality characteristics and response fluctuation intensity characteristics. Set the response directionality threshold to ±5° and the fluctuation intensity threshold to 2.5% to filter out trend deviation events.

[0168] In this embodiment, a trend fitting process is performed on the continuous parameter sequence collected within each sliding window in the response trend analysis dataset. The least squares method is used to fit the slope of the actuator feedback parameter changes along the time axis, and the fluctuation intensity index is calculated by combining the second-order differences of adjacent data points. To identify response excursion trends, a directional excursion threshold of ±5 degrees is set, combined with a fluctuation intensity threshold of 2.5% to identify anomalous trend segments. Once the trend slope exceeds the directional threshold and the fluctuation intensity index exceeds the set threshold, the window is recorded as a trend excursion event, and the occurrence time, actuator number, and excursion characteristic value are marked.

[0169] Step S53: grouping trend deviation events by actuator number and calculating the start / stop control priority score;

[0170] In this example, all identified trend deviation events are grouped by actuator number. A start-stop control priority score is calculated based on each actuator's average control trigger frequency, response deviation amplitude, and process node location priority. This score is weighted by three factors: control frequency (0.4), response deviation (0.3), and process priority (0.3). Each actuator's final score ranges from 0 to 1. The system normalizes the scores and stores them in a start-stop score table, which serves as an important reference for subsequent adjustments to the control cadence.

[0171] Step S54: setting the start-stop control parameter amplitude range according to the start-stop control priority score, wherein the control interval of the actuator with a high priority score is extended by 30 seconds, and the actuator with a low priority score maintains the original setting value, thereby generating an actuator start-stop control instruction table;

[0172] In this embodiment, control parameter adjustment strategies are set based on the start-stop control priority score table. Specifically, for actuators with a priority score higher than 0.7, their start-stop intervals are extended by 30 seconds based on the original control instructions, and this adjustment strategy is locked for the next 15 minutes. Actuators with scores between 0.4 and 0.7 maintain their original control intervals. Actuators with scores below 0.4 are marked as "low response risk" and do not require control parameter adjustment. All control adjustment strategies are uniformly coded and organized to generate an actuator start-stop control instruction table, which contains each device's number, the control interval adjustment range, and the effective duration.

[0173] Step S55: perform structural verification on the actuator start-stop control instruction table, upload the verified actuator start-stop control instruction table to the lithium carbonate production line management platform, and complete the issuance of control task instructions.

[0174] In the present embodiment, after generating the actuator start-stop control instruction table, its structural integrity is checked item by item to ensure that each control record contains the fields such as actuator unique number, control type (start / stop), adjustment value and instruction effective time to prevent missing items or conflicts. After completion of the verification, the control instruction table is pushed to the task scheduling port of the lithium carbonate production line management platform via the MQTT protocol, and the reception confirmation mechanism of the scheduling system is triggered. After the task is successfully received, the platform will feedback ACK signal to confirm that it is successfully issued, and record the control instruction execution timestamp as the triggering basis for the synchronous response of the field equipment.

[0175] It is particularly important to calculate the start-stop control priority score as follows:

[0176] Based on the executor number, abnormal path segment identifier, and time interval recorded in the trend deviation event after number grouping, the execution parameter change data of each executor within 300 seconds before and after the abnormality is extracted;

[0177] In this embodiment, based on numbered and grouped trend deviation events, the actuator number, associated abnormal path segment identifier, and specific abnormal time interval are extracted from the records. For each actuator, operating parameter change data is retrieved from the field data collection platform within a 300-second time interval before and after the abnormality occurred. Parameters include control current, voltage, displacement feedback, and operating status codes. All data is organized at second-level resolution and arranged chronologically to form a structured table of actuator parameter changes for subsequent segmentation and trend quantification. The data is then archived in CSV format and stored in the local analysis cache directory.

[0178] Set the execution window length to 60 seconds and the step length to 15 seconds, divide the execution parameter change data into window segments, and calculate the first-order slope, second-order difference variance, peak-to-valley difference, and fluctuation frequency for each window to generate a set of trend feature vectors.

[0179] In this embodiment, the extracted execution parameter change data is windowed with each 60-second time segment, and the step size is set to 15 seconds to ensure that both the analysis coverage and the data granularity are taken into account. Within each window segment, the displacement feedback quantity is linearly fitted with the time series to extract the first-order slope, and the second-order difference variance is calculated for the difference sequence of consecutive time points to evaluate the degree of response fluctuation. At the same time, the difference between the maximum and minimum values ​​in the data segment is extracted as the peak-to-valley amplitude indicator, and the fluctuation frequency is calculated based on the number of positive and negative fluctuation cycles. The four types of eigenvalues ​​extracted from all window segments are superimposed in chronological order to generate a set of trend feature vectors, which are uniformly attributed to the corresponding actuator number and abnormal path segment identifier.

[0180] The influence factor of the path segment is calculated based on the occurrence frequency, abnormal duration and path depth of each path segment in the process chain segment abnormal path table, and the influence factor of the path segment to which the trend feature vector set belongs is combined with the corresponding response feature vector set to form a weighted feature input set;

[0181] In this embodiment, the abnormal path table of the process chain segment is read, and each path segment is assigned a value according to the number of occurrences, the duration of the abnormality, and the path level to which it belongs, and then weighted and combined. The weights are set to 0.4 for the frequency of occurrence, 0.4 for the duration, and 0.2 for the path depth, to calculate the path segment impact factor. The calculation formula of the impact factor can be: The frequency unit is times, the duration unit is seconds, and the depth penalty is set to the square of the path depth. The path depth is based on the first layer, and the influence coefficient decreases by 0.3 with each layer. Subsequently, using the path segment identifier as an index, the trend feature vector set is bound to the influence factor of each path segment. The influence factor is multiplied by each dimension of the trend feature vector to form a weighted feature value, which is finally combined into a weighted feature input set at the actuator dimension.

[0182] Based on the trend feature vector set of each actuator, the response directionality scoring index is set as the absolute value of the linear fitting slope, and the fluctuation intensity scoring index is set as the standard deviation of the second-order difference to perform actuator response fluctuation scoring. The absolute value of the slope exceeding 0.1 is mapped to 10 points, and less than 0.01 is mapped to 0 points. The standard deviation of the fluctuation intensity exceeding 2% is mapped to 5 points, and less than 0.2% is mapped to 0 points. The absolute value of the slope and the standard deviation of the fluctuation intensity are weighted averaged to obtain the fluctuation score value of the feature vector, and the maximum value is taken according to the actuator number as the fluctuation score of the actuator.

[0183] In this embodiment, a characteristic score is assigned to each actuator's response behavior based on its weighted trend feature set. Specifically, the absolute value of the first-order slope in each set of trend vectors is extracted as the basis for scoring the response directionality. If the absolute value of the slope is greater than 0.1, the mapping score is set to 10 points; if it is less than 0.01, the mapping score is set to 0 points. Simultaneously, the standard deviation of the second-order difference is extracted as the basis for scoring the intensity of the response fluctuation. Mappings with a standard deviation greater than 2% are scored as 5 points, and mappings with a standard deviation less than 0.2% are scored as 0 points. These two scoring indicators are weighted 0.6 and 0.4, respectively, and combined to obtain a fluctuation score for each set of trend feature vectors. Ultimately, the maximum value of all scores for each actuator is used as the representative score for the actuator's fluctuation behavior.

[0184] The weighted feature input set is integrated with the volatility score, and the priority score formula is set as follows:

[0185]

[0186] In this example, the weighted feature input set is fused with the actuator's fluctuation score. The final priority score is calculated by combining the weighted feature's average amplitude value with the fluctuation score, weighted at a ratio of 0.7:0.3. During this process, the priority scoring formula is used to take the median of each actuator's weighted feature vector within the mean range as the representative feature. This is then combined with its maximum fluctuation score to create a table of actuator start / stop control priority scores. Each row in this table records fields such as the actuator number, trend feature mean, fluctuation score, and final priority score, providing data support for downstream start / stop control decisions.

[0187] The priority score results of all actuators are summarized to obtain the start-stop control priority score.

[0188] Optionally, this specification also provides a battery-grade lithium carbonate production digital management system for executing the battery-grade lithium carbonate production digital management method as described above, the battery-grade lithium carbonate production digital management system comprising:

[0189] A relationship matching module is used to obtain actuator feedback data and actuator control data, and perform control signal-feedback relationship matching based on the actuator feedback data and actuator control data to obtain a control command-physical response relationship pair;

[0190] The response pattern recognition module is used to perform periodic sampling of the control instruction-physical response relationship to obtain a control period actuator parameter set; perform standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template;

[0191] The mismatch behavior identification module is used to obtain real-time actuator electronic control behavior data, perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify the response mismatch behavior, and obtain an actuator physical action abnormality report;

[0192] The impact path derivation module is used to derive the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report, and obtain the process chain segment abnormal path table;

[0193] The control instruction generation module is used to perform actuator action response trend analysis based on the process chain segment abnormal path table, and determine the start-stop control parameter amplitude based on the action response trend analysis results to obtain the actuator start-stop control instruction table; upload the actuator start-stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

[0194] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0195] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A digital management method for battery-grade lithium carbonate production, characterized in that: The following steps are involved: Step S1: acquiring actuator feedback data and actuator control data, and performing control signal-feedback relationship pairing based on the actuator feedback data and the actuator control data to obtain a control command-physical response relationship pair; Step S2: performing control periodic sampling on the control instruction-physical response relationship pair to obtain a control period actuator parameter set; performing standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template; Step S3: Acquire real-time actuator electronic control behavior data, and perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify response mismatch behavior, and obtain an actuator physical action abnormality report; Step S4: deriving the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report to obtain a process chain segment abnormal path table; Step S5: performing actuator action response trend analysis based on the process chain segment abnormal path table, and determining the start / stop control parameter amplitude based on the action response trend analysis result to obtain an actuator start / stop control instruction table; Upload the actuator start and stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

2. The digital management method for battery-grade lithium carbonate production according to claim 1, wherein: Step S1 is specifically as follows: Step S11: collecting actuator control signals and corresponding feedback signals to obtain actuator feedback data and actuator control data; Step S12: performing channel mapping and synchronous pairing on the actuator feedback data and the actuator control data to generate a control feedback channel binding table; Step S13: extracting feedback feature segments within a preset action cycle based on the control feedback channel binding table to form a control feedback response sample set; Step S14: performing control mode identification on the control feedback response sample set to obtain a controller control mode sample set; Step S15: Classify and archive the controller control mode sample set according to actuator category and control mode, and generate control instruction-physical response relationship pairs.

3. The digital management method for battery-grade lithium carbonate production according to claim 2, wherein: Step S12 is specifically as follows: Step S121: Retrieve the control output port corresponding to each actuator and the acquisition input channel configuration document of the physical feedback feedback device through the lithium carbonate production line management platform, combine the obtained equipment wiring diagram and on-site wiring labels, establish the actuator-feedback device physical connection relationship, and thus obtain a physical channel index table; Step S122: Set the synchronization sliding time window to 0.5s, and use the physical channel index table to perform control signal-feedback signal synchronization window slicing to obtain aligned data segment pairs; Step S123: Periodically reorganize the aligned data segment pairs and verify feedback direction consistency, remove channel pairs with feedback value noise exceeding ±10% or missing points, and generate a stable control response data set; Step S124: normalize and organize the stable control response data set, and construct a control feedback channel binding table in the format of actuator number, control signal geology, feedback signal geology, feedback delay mean, and response amplitude mean.

4. The digital management method for battery-grade lithium carbonate production according to claim 2, wherein: Step S13 is specifically as follows: Step S131: Read each pair of control-feedback channel combinations in the control feedback channel binding table, set the standard action cycle window to 300s, and extract the synchronous control trigger event sequence; Step S132: performing boundary regularization and signal interception on the feedback data segments corresponding to each group of control trigger events in the synchronous control trigger event sequence, and setting the response amplitude threshold to 3% to extract the segments with continuous signal changes; Step S133: Encode and encapsulate the continuously changing signal segments to generate response curve segment samples; Step S134: Summarize the response samples of each channel combination within the standard action cycle window, eliminate the channel combinations and corresponding samples with a response amplitude change standard deviation <0.05 and a response start delay standard deviation <0.3s, and obtain a control feedback response sample set.

5. The digital management method for battery-grade lithium carbonate production according to claim 1, wherein: The control periodic sampling in step S2 is specifically as follows: Extract the control signal trigger timestamp of each actuator in the synchronous control trigger event sequence, and group the actuators by actuator number to calculate the average interval time of the control signals of each type of actuator, so as to set the initial control period; Analyze the physical response characteristics of the actuator in combination with the control feedback response sample set to set the cycle buffer interval; Setting an actuator control cycle window based on an initial control cycle and a cycle buffer; Extract the continuous control event sequence of each actuator within the actuator control cycle window according to the control instruction-physical response relationship and generate a periodic event marker set; For each action event in the periodic event marker set, the actuator parameter curve segment within the corresponding actuator control period window is extracted to construct a periodic control response sample; The periodic control response samples are grouped and counted by actuator number, and the periodic sample segments with response standard deviation < 5% and actuator control times ≥ 30 are screened out to obtain the control period actuator parameter set.

6. The digital management method for battery-grade lithium carbonate production according to claim 1, characterized in that: The standard action response pattern recognition in step S2 is specifically as follows: The periodic control response samples in the control period actuator parameter set are time-normalized and uniformly mapped to a fixed length of 200 to generate a set of normalized response curves. Based on the normalized response curve set, the rising edge slope, peak position, steady-state platform range and fall rate of each response curve are extracted to form a response feature vector set; Similarity clustering analysis was performed on the response feature vector set, and the silhouette coefficient threshold was set to 0.65 to screen stable response clusters. The central curve in each cluster was extracted as the standard action response pattern.

7. The digital management method for battery-grade lithium carbonate production according to claim 1, characterized in that: The physical matching check in step S3 is specifically as follows: Read the corresponding standard action response template according to the actuator number of the real-time actuator electronic control behavior data, extract the control signal trigger point and the corresponding physical response delay characteristics in the template, and obtain the control response data to be matched; The calibration sliding window length was set to 300 seconds. The real-time actuator electronic control behavior data was segmented into sliding windows. The control signal trigger events and corresponding sensor response curve segments of each actuator within the calibration sliding window were extracted to construct the control-response sample unit to be verified. Compare the control-response sample units to be verified with the control response data to be matched one by one, set the response delay deviation tolerance to 0.5s and the response amplitude deviation tolerance to 5%, and calculate the physical response matching degree of each set of samples; Based on the physical response matching degree, sample units with a matching degree lower than 90% are selected from the control-response sample units to be verified, and recorded as response mismatch behaviors according to the timestamp and actuator number, and a summary report of the actuator physical action abnormality is generated.

8. The digital management method for battery-grade lithium carbonate production according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41: extracting the actuator number, abnormality type and triggering timestamp corresponding to the abnormal event in the actuator physical action abnormality report, and sorting them in chronological order to generate a time sequence chain table of abnormal events; Step S42: establishing an actuator-process node mapping table based on the actuator physical action objects and process position identifiers recorded in the control instruction-physical response relationship pair; Step S43: Combine the abnormal event time sequence chain table and the executor-process node mapping table to construct an abnormal impact propagation path diagram, set the maximum propagation depth to 3 layers, find the downstream process node path segment of the abnormal impact propagation path diagram, and record it as a potential abnormal path set; Step S44: Filter the path segments in the potential abnormal path set that appear more than 3 times and have an impact window time span of less than 600 seconds, and output them as a process chain segment abnormal path table.

9. The digital management method for battery-grade lithium carbonate production according to claim 1, characterized in that: Step S5 is specifically as follows: Step S51: Based on the process chain abnormal path table, the actuator number and the time interval of the abnormality corresponding to each abnormal path segment are extracted. The actuator parameter change data within adjacent abnormal intervals are sampled using a sliding window, with a window length of 120 seconds and a step length of 30 seconds, to form a response trend analysis data set. Step S52: Perform linear fitting and second-order difference calculation on the continuous parameter change trend of each actuator in the response trend analysis data set to extract the response trend directionality characteristics and response fluctuation intensity characteristics. Set the response directionality threshold to ±5° and the fluctuation intensity threshold to 2.5% to filter out trend deviation events. Step S53: grouping trend deviation events by actuator number and calculating the start / stop control priority score; Step S54: setting the start-stop control parameter amplitude range according to the start-stop control priority score, wherein the control interval of the actuator with a high priority score is extended by 30 seconds, and the actuator with a low priority score maintains the original setting value, thereby generating an actuator start-stop control instruction table; Step S55: perform structural verification on the actuator start-stop control instruction table, upload the verified actuator start-stop control instruction table to the lithium carbonate production line management platform, and complete the issuance of control task instructions.

10. A digital management system for battery-grade lithium carbonate production, characterized in that: For executing the digital management method for battery-grade lithium carbonate production according to claim 1, the digital management system for battery-grade lithium carbonate production comprises: A relationship matching module is used to obtain actuator feedback data and actuator control data, and perform control signal-feedback relationship matching based on the actuator feedback data and actuator control data to obtain a control command-physical response relationship pair; The response pattern recognition module is used to perform periodic sampling of the control instruction-physical response relationship to obtain a control period actuator parameter set; perform standard action response pattern recognition based on the control period actuator parameter set to obtain an actuator standard action response template; The mismatch behavior identification module is used to obtain real-time actuator electronic control behavior data, perform physical matching verification on the real-time actuator electronic control behavior data according to the actuator standard action response template, identify the response mismatch behavior, and obtain an actuator physical action abnormality report; The impact path derivation module is used to derive the physical path segments that affect the process based on the control instruction-physical response relationship and the actuator physical action abnormality report, and obtain the process chain segment abnormal path table; The control instruction generation module is used to perform actuator action response trend analysis based on the process chain segment abnormal path table, and determine the start-stop control parameter amplitude based on the action response trend analysis results to obtain the actuator start-stop control instruction table; upload the actuator start-stop control instruction table to the lithium carbonate production line management platform to execute the actuator management task.

Images