Spray humidification control method based on moisture ratio online monitoring in wet mixing process of grinding wheel forming material

Abstract

The present application relates to a spray humidification control method based on moisture ratio online monitoring of wet mixing process of grinding wheel forming material, and the core scheme comprises the following steps: obtaining real-time moisture ratio signal through high-frequency sampling and filtering, extracting standardized local features through sliding interception, normalization and outlier rejection; constructing trend three-dimensional feature vector by using difference, polarity turning and curvature algorithm, realizing identification of six basic phase states of dynamic working condition, and forming discretized trend phase space through space mapping and gridding; constructing weighted directed graph based on historical evolution path, and extracting high-confidence parameter response template; combining real-time state triple instruction and physical boundary, dynamically setting spray control parameters, driving the actuator, and forming closed-loop update in feedback. The method improves the identification accuracy of the system to the dynamic evolution of the wet mixing process and the intelligence, stability and response speed of the spray regulation.

Description

Technical Field

[0001] This invention relates to the field of automated control technology for wet mixing processes, and in particular to a spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel molding materials. Background Technology

[0002] Currently, automated spray humidification control systems for wet mixing processes of grinding wheel forming materials generally adopt technical approaches such as single-point moisture ratio adjustment, feedback error driving, or spray control command generation based on predictive models. Mainstream industry solutions largely rely on online moisture ratio monitoring devices, such as near-infrared moisture sensors. These sensors acquire real-time material moisture data and compare it with preset target values ​​to form a feedback loop. Dynamic adjustment of moisture content is achieved using methods such as PID control, quantitative flow / pressure control, and periodic spray compensation. Furthermore, some advanced systems attempt to introduce predictive algorithms, such as ARIMA and LSTM time-series models, to predict short-term moisture ratio trends and adjust spray parameters in advance, thereby improving the control's foresight and response speed. With the increasing complexity of wet mixing processes and the growing influence of batch fluctuations and environmental disturbances, the industry is increasingly focusing on the control system's ability to capture dynamic trends in moisture ratio and the need for adaptive adjustment under multiple operating conditions.

[0003] Representative technical solutions typically use single-point deviation or prediction error as the core decision variable. The main processes include online moisture ratio monitoring, single-point target matching, error calculation, flow / pressure regulation, and feedback correction. These technologies are suitable for standard process scenarios with small moisture ratio fluctuations and high material moisture content uniformity, meeting the needs of routine automated humidification control. However, when facing complex scenarios such as batch material moisture fluctuations and changes in ambient temperature and humidity, this method, relying on absolute values ​​or single-step prediction, is prone to problems such as slow response, unstable regulation, and insufficient closed-loop robustness.

[0004] Existing technologies suffer from the following significant shortcomings. First, traditional control strategies ignore the trend characteristics of moisture ratio over time and cannot dynamically adjust spray control parameters based on trend evolution. Their adjustment targets are primarily current values ​​or predicted future points, lacking the identification and response to local trend states (such as accelerated increases, decelerated decreases, inflection point transitions, etc.), resulting in limited system adaptability. Second, control parameter adjustments lack dynamic tuning; spray flow rate, pressure, and cycle are configured solely based on fixed rules or manual experience, making it difficult to map trend change paths under complex operating conditions in real time. Third, while some predictive models attempt forward adjustment, they suffer from problems such as accumulated prediction errors, insufficient model generalization ability, overfitting risks, and strong dependence on external environmental data, failing to guarantee control robustness and accuracy in actual production scenarios. Furthermore, current technical approaches generally lack embedded evaluation of temporal characteristics such as trend state dwell time and abrupt change reliability, failing to effectively distinguish between noise disturbances and true trend changes, easily leading to control command failures or frequent switching.

[0005] Therefore, current automated spray humidification control technology for wet mixing processes of grinding wheel forming materials urgently needs a method for dynamic parameter self-tuning based on the evolution of moisture ratio trends. Specific requirements include: achieving accurate identification and encoding of the moisture ratio time series trend phase; establishing nonlinear control semantics through trend phase mapping; enabling the system to adaptively adjust spray parameters according to the trend change path, improving its perception and response capabilities to complex working conditions; and ensuring closed-loop robust control compatible with material batch fluctuations and environmental disturbances, guaranteeing the safety and effectiveness of spray control parameter adjustments. This technology will bring a new trend-driven intelligent tuning mechanism to the field of automated control of wet mixing processes, significantly overcoming the limitations of traditional single-point deviation-driven and pure predictive control methods, and providing strong support for achieving high-precision, high-robust closed-loop control of wet mixing humidification processes. Summary of the Invention

[0006] This application provides a spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel molding material, which aims to solve one of the problems or issues of the prior art mentioned in the background art.

[0007] The spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming material provided in this application specifically includes: S1: Obtain the original moisture ratio signal sequence within a continuous time window during the wet mixing process, and perform sliding truncation processing on the original moisture ratio signal sequence to generate local moisture ratio data segments for trend feature extraction.

[0008] S2: Calculate the first-order differential symbol sequence, the second-order differential polarity inflection point, and the local curvature change rate based on the local moisture ratio data fragment, so as to identify and generate the basic phase state code that characterizes the dynamic characteristics of the current working condition.

[0009] S3: The basic phase state encoding is mapped to a two-dimensional coordinate system consisting of the rate of change dominant horizontal axis and the acceleration dominant vertical axis to construct a discretized trend phase space coordinate point set.

[0010] S4: Based on the frequency statistics of the transitions between the trend phase spatial coordinate points under historical conditions, construct a directed graph structure with weighted attributes to generate a trend phase transition map that records high-frequency evolution paths.

[0011] S5: Based on the typical evolutionary chain in the trend phase transition spectrum where the cumulative occurrence exceeds a preset threshold, define a dedicated spray control parameter response template library that includes parameter adjustment directional strategies and step-by-step incremental rules.

[0012] S6: Real-time monitoring of the residence time of the basic phase state code in the trend phase space at the current moment, and combined with the comparison results of the historical average residence time and the multi-scale fluctuation energy ratio, to generate a triplet control command containing the phase code, residence level and evolution chain confidence.

[0013] S7: Based on the triplet control command, retrieve matching items from the dedicated spray control parameter response template library, and combine the nozzle response delay upper limit and the minimum adjustable pressure step physical constraint boundary to perform parameter safety mapping to generate a dynamically tuned spray control parameter set.

[0014] S8: The spray actuator is driven to move using the dynamically tuned spray control parameter set, and the new moisture ratio feedback signal after the movement is re-inputted to the sliding interception link to complete the iterative update of the closed-loop control process.

[0015] The spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming material provided in this application has the following beneficial effects: (1) The spray control method driven by the trend dynamics of moisture ratio in a wet mixing process provided in this application significantly improves the dynamic adaptability of the control system under complex working conditions by constructing a trend phase modeling and evolution chain response mechanism. Traditional control strategies mostly rely on the deviation between the measured value of moisture ratio and the set target for feedback adjustment, or use a predictive model to generate future values ​​to achieve feedforward compensation. However, in actual production scenarios with large batch differences of materials and frequent environmental disturbances, such methods are prone to problems such as response lag, severe overshoot, or even oscillation instability. This scheme abandons the direct dependence on absolute values ​​and instead explores the inherent trend dynamics of the moisture ratio sequence. It uses a sliding time window to extract continuous segments and identifies six basic trend phases based on the first-order difference sign, the second-order difference polarity transition, and the local curvature change rate: stable maintenance, accelerated rise, decelerated rise, accelerated fall, decelerated fall, and inflection point transition. Each phase is represented by a unique code and located in a two-dimensional trend space, realizing a refined semantic abstraction of the dynamic rhythm of the process. This design enables the system to move beyond simply judging the "current value" and instead understand the "stage of the change process," thus providing a more forward-looking contextual basis for control decisions and effectively overcoming the technical shortcomings of traditional single-point deviation control in nonlinear and time-varying environments where it is slow to respond.

[0016] (2) Furthermore, by establishing an online-updable trend phase transition map and extracting typical evolution chains, combined with a dwell time adaptive mechanism and parameter response template matching strategy, this scheme realizes the accumulation of experience and dynamic optimization of control logic, and significantly improves the robustness and intelligence level of the closed-loop system. The phase transition paths that frequently occur under historical conditions are constructed into a weighted directed graph structure and continuously updated through an online learning mechanism. When the cumulative frequency of a certain path reaches the standard, it is marked as a "typical evolution chain", reflecting the common trend evolution law in wet mixing. A directional control strategy template is configured for each typical chain. For example, in the "accelerated rise → inflection point transition" chain, the spray cycle is shortened first and the pressure benchmark value is increased stepwise to make the adjustment action consistent with the trend evolution rhythm. At the same time, a phase dwell time monitoring mechanism is introduced. If the current phase continues to exceed the historical average, the template's enhanced execution level is triggered to enhance the adjustment force to prevent excessive trend inertia from causing loss of control. If an abnormal jump is detected (such as skipping the intermediate phase), the multi-scale fluctuation energy ratio verification process is started to determine whether it is a real trend jump and avoid misjudgment that causes violent disturbances. The above mechanisms together construct a control system with self-learning, state perception and risk warning capabilities, which enables proactive adaptation and precise intervention to the inherent evolutionary laws of the wet mixing process without the need for external prediction models, fuzzy rules or reinforcement learning frameworks.

[0017] (3) In addition, the spray control module receives a triplet instruction consisting of "phase encoding + residence level + evolution chain confidence" and combines it with the physical constraints of the equipment to complete the safety mapping output, ensuring that the control strategy is both forward-looking and meets engineering feasibility, significantly improving the practicality and deployment stability of the system. Unlike traditional controllers that rely on precise numerical settings or complex algorithm reasoning, this solution transforms trend information into an operable semantic carrier and quickly retrieves matching strategies through a local response template library, greatly reducing computational overhead and implementation complexity; at the same time, it fully considers the actual limitations of the equipment, such as the minimum adjustment step size and response delay, to ensure that the output parameters are within the safe and feasible domain, avoiding the failure of the actuator or increased wear due to idealized instructions. The overall architecture does not rely on knowledge graphs, reinforcement learning or fuzzy control structures, has good interpretability and scalability, and is suitable for the migration and deployment of mixed equipment of different models. This method fundamentally changes the paradigm shift of wet mixing control from "passive correction" to "active adaptation to trends." In particular, it can maintain the stability and consistency of moisture control when facing common disturbances such as fluctuations in raw material moisture content and changes in climate temperature and humidity. It is significantly better than existing methods based on static thresholds or fixed gain adjustment.

[0018] In summary, this solution constructs a complete closed-loop response chain from phase recognition and evolutionary analysis to strategy matching by taking the trend of moisture ratio change itself as the core control semantics. This enables deep perception and intelligent control of the dynamic characteristics of the wet mixing process. It not only effectively avoids the strong dependence of traditional methods on absolute values ​​and prediction models, but also achieves substantial breakthroughs in response sensitivity, control robustness, and engineering applicability, providing a brand-new technical path for high-precision, adaptive industrial process control. Attached Figure Description

[0019] Figure 1 This is the main flow chart of a spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming materials.

[0020] Figure 2 This is a sub-flowchart of a spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming materials.

[0021] Figure 3 This is another sub-flowchart of the spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming material. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0023] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] like Figure 1 As shown, this application provides a spray humidification control method based on online monitoring of the moisture ratio during the wet mixing process of grinding wheel forming material, specifically including: S1: Obtain the original moisture ratio signal sequence within a continuous time window during the wet mixing process, and perform sliding truncation processing on the original moisture ratio signal sequence to generate local moisture ratio data segments for trend feature extraction.

[0025] S2: Calculate the first-order differential symbol sequence, the second-order differential polarity inflection point, and the local curvature change rate based on the local moisture ratio data fragment, so as to identify and generate the basic phase state code that characterizes the dynamic characteristics of the current working condition.

[0026] S3: The basic phase state encoding is mapped to a two-dimensional coordinate system consisting of the rate of change dominant horizontal axis and the acceleration dominant vertical axis to construct a discretized trend phase space coordinate point set.

[0027] S4: Based on the frequency statistics of the transitions between the trend phase spatial coordinate points under historical conditions, construct a directed graph structure with weighted attributes to generate a trend phase transition map that records high-frequency evolution paths.

[0028] S5: Based on the typical evolutionary chain in the trend phase transition spectrum where the cumulative occurrence exceeds a preset threshold, define a dedicated spray control parameter response template library that includes parameter adjustment directional strategies and step-by-step incremental rules.

[0029] S6: Real-time monitoring of the residence time of the basic phase state code in the trend phase space at the current moment, and combined with the comparison results of the historical average residence time and the multi-scale fluctuation energy ratio, to generate a triplet control command containing the phase code, residence level and evolution chain confidence.

[0030] S7: Based on the triplet control command, retrieve matching items from the dedicated spray control parameter response template library, and combine the nozzle response delay upper limit and the minimum adjustable pressure step physical constraint boundary to perform parameter safety mapping to generate a dynamically tuned spray control parameter set.

[0031] S8: The spray actuator is driven to move using the dynamically tuned spray control parameter set, and the new moisture ratio feedback signal after the movement is re-inputted to the sliding interception link to complete the iterative update of the closed-loop control process.

[0032] Step S1: Obtain the original moisture ratio signal sequence within a continuous time window during the wet mixing process, and perform sliding truncation processing on the original moisture ratio signal sequence to generate local moisture ratio data segments for trend feature extraction. Specifically, this includes: S1.1: The analog voltage signal output by the online near-infrared moisture sensor installed at the outlet of the wet mixer is subjected to high-frequency sampling and digital filtering to eliminate environmental electromagnetic interference and generate a discrete original moisture ratio signal sequence with a unified timestamp.

[0033] For the analog voltage signal of the online near-infrared moisture sensor installed at the outlet of the wet mixer, a high-frequency sampling action is performed before entering the digital processing stage. The sampling frequency is selected to be a multiple higher than the reciprocal of the sensor response time to avoid aliasing effect, and the time interval between sampling points is kept constant to take into account the time resolution required for trend analysis.

[0034] The original discrete voltage sequence obtained by sampling is input into a bandpass digital filter. The passband boundary of the filter is set according to the operating band of the near-infrared sensor and the spectral characteristics of the signal components under the actual process environment, so as to remove low-frequency temperature and humidity drift components and high-frequency electromagnetic interference noise.

[0035] Amplitude correction is performed on the bandpass filtered voltage sequence. The voltage value is mapped to the corresponding moisture percentage value using the sensor calibration curve. A precise and uniform time identifier is added to each sampling point according to the timestamp generation function of the process control clock synchronization module.

[0036] A timestamp resampling mechanism is used to ensure consistency of timestamps for data from different batches or channels, so that the sliding capture process can extract continuous data segments under the same time reference.

[0037] Data integrity verification algorithms are used to detect and remove abnormal data points caused by momentary sensor failure or communication errors. For example, the upper and lower limits of the physical reasonable range of moisture ratio are set for rapid boundary determination.

[0038] By using high-frequency sampling, digital filtering, amplitude correction, and timestamp unification, the analog voltage signal is transformed into a discrete moisture ratio original signal sequence with a unified timestamp, thus realizing a stable and interference-free input dataset that can be directly used for trend feature extraction.

[0039] For example, a near-infrared moisture sensor, model NIR-850, is installed at the outlet of the wet mixer. Its maximum response time is 0.5ms, the sampling frequency is set to 2000Hz, and the time interval is 0.5ms. The sampled signal is processed by an FIR bandpass filter with a passband range of 0.1Hz to 30Hz and a filter order of 128 to eliminate low-frequency drift and high-frequency interference from the environment. Amplitude correction uses the factory calibration curve, and the calibration relationship is as follows: Where W represents the moisture content percentage, and V represents the filtered voltage value (in volts). Timestamp standardization uses a 1ms resolution global time identifier output from a synchronous clock module to ensure consistent time base across different channels. Abnormal data detection sets the moisture content range to 0-40%, and all sampling points outside this range are discarded. For stable operating conditions, the original moisture content signal sequence output in this step shows no missing or abnormal values ​​within a continuous 1000-point sampling period, the signal noise standard deviation is significantly reduced, and the amplitude changes smoothly, making it suitable for subsequent trend feature extraction.

[0040] S1.2: Construct a first-in-first-out circular buffer based on the original discrete moisture ratio signal sequence, and set the sliding step size parameter according to the preset trend analysis time window length to form a dynamic data buffer queue that supports continuous streaming processing.

[0041] Based on the discretized moisture ratio original signal sequence output by S1.1, a circular buffer is constructed using a memory buffer mechanism that supports first-in-first-out structure characteristics. This ensures that the earliest stored data is automatically overwritten when data is enqueued, thereby achieving continuous signal cyclic storage within a limited capacity.

[0042] When constructing the circular buffer, the buffer capacity is precisely set to the number of sampling points corresponding to the preset trend analysis time window length, and this capacity threshold is synchronously stored in the buffer logic control unit to ensure that the buffer queue can meet the extraction of complete trend segments.

[0043] The sliding step size is generated by the step size parameter calculation unit based on the time resolution and window overlap rate required for trend analysis. The formula for calculating the sliding step size is as follows: in, This represents the number of sampling points corresponding to the sliding step size. The length of the trend analysis time window is in seconds. The sampling frequency is Hz. This represents the window overlap time in seconds.

[0044] The index control logic that writes the sliding step size parameter into the cache queue, together with the read and write pointers of the circular buffer, realizes dynamic streaming data advancement, so that the latest signal fragment within the window length range can be obtained every time the interception is triggered.

[0045] After the cache queue is built, a data synchronization identification mechanism is configured to bind the time field of all data points in the circular buffer through a unified timestamp, ensuring that the data segments in the subsequent sliding truncation stage have strict temporal continuity.

[0046] By using a circular buffer and sliding step control method, the original discrete moisture ratio signal sequence from the previous step is transformed into a dynamic data buffer queue that supports continuous streaming trend feature extraction, thereby enabling uninterrupted trend analysis data input.

[0047] For example, the sampling frequency of the discrete moisture ratio original signal sequence obtained by high-frequency sampling and digital filtering of the output signal of the near-infrared moisture sensor installed at the outlet of the wet mixer is set to 200Hz, the trend analysis time window length is set to 5 seconds, the window overlap time is set to 1 second, and the step size parameter calculation unit calculates the sliding step size sampling points as 5×200-1×200=800 points according to the formula. The capacity of the ring buffer is set to 1000 points to ensure coverage of the window length and leave safety redundancy. When the signal is continuously input, the read and write pointer of the buffer queue advances in 800-point steps. Each time the index control logic advances, it extracts a continuous data segment containing the latest 1000 points from the buffer and binds it to a unified timestamp field for use by the S1.3 sliding interception unit. Under the above configuration, the dynamic data buffer queue runs continuously on site, ensuring that the wet mixing process trend analysis module can receive stable segments at a millisecond-level signal update frequency, significantly improving the real-time performance and accuracy of trend analysis and subsequent spray control parameter tuning.

[0048] S1.3: Utilize the dynamic data cache queue to perform a periodic sliding truncation operation, separating the continuous fixed-duration data segments before the current moment from the discretized original moisture ratio signal sequence to generate local moisture ratio data segments for characterizing instantaneous operating conditions.

[0049] S1.4: Perform zero-mean normalization and outlier removal on the generated local moisture ratio data fragments to eliminate the influence of inter-batch baseline drift and output standardized local moisture ratio data fragments as direct input objects for subsequent calculation of first-order difference symbol sequences.

[0050] Step S2: Based on the local moisture ratio data fragment, calculate the first-order difference symbol sequence, the second-order difference polarity inflection point, and the local curvature change rate to identify and generate a basic phase state code characterizing the dynamic characteristics of the current operating condition. Specifically, this includes: S2.1: Obtain a local moisture ratio data segment, perform point-by-point first-order difference operation on the local moisture ratio data segment to generate an instantaneous rate of change sequence, and perform symbolic mapping processing on the instantaneous rate of change sequence based on a zero threshold to output a first-order difference symbolic sequence representing the direction of increase or decrease in moisture ratio.

[0051] A point-by-point index structure is established for standardized local moisture ratio data segments to ensure that data points are available for the differential calculation module to call in chronological order.

[0052] The first-order difference calculation unit is invoked to perform rate of change calculations on adjacent data points in the index structure, using the difference formula. Among the changes The instantaneous rate of change This represents the moisture content at the current moment. This represents the moisture ratio value at the previous moment.

[0053] Call the zero threshold determination module to determine the preset threshold. The rate of change series is compared point by point using the sign mapping formula. Performs symbolization, outputting the sign bit as a positive, negative, or zero value.

[0054] The symbolized output is assembled into a first-order difference symbol sequence according to the original sequence position, and a timestamp index is attached to support subsequent inflection point identification.

[0055] Through the above chain processing method, standardized local moisture ratio data fragments are transformed into first-order difference symbol sequences that characterize the direction of increase or decrease in moisture ratio, thereby achieving a quantitative expression of trend direction characteristics.

[0056] For example, the standardized local moisture ratio data segment collected by the near-infrared moisture sensor at the outlet of the wet mixer is 100 points long, with a sampling period of 1 second, and the change rate calculation unit sets the threshold θ to 0.002. The difference calculation module executes the formula for each pair of adjacent data points. For example, at point 50, the moisture ratio is 0.215, and at point 49, it is 0.214, with a change rate Δr = 0.001. Because the absolute value is less than the threshold of 0.002, the symbol mapping module classifies it as zero. At point 51, the moisture ratio is 0.219, with a change rate Δr = 0.004, which is greater than the threshold of 0.002 and is classified as positive. The final output first-order difference symbol sequence contains three types of symbols: positive, negative, and zero. The time position corresponding to each symbol is consistent with the direction of moisture ratio change. Verification shows that this significantly improves the accuracy of trend direction determination and the stability of subsequent phase recognition.

[0057] S2.2: Based on the first-order differential symbol sequence, perform a logical XOR comparison on adjacent symbol bits to identify the symbol flip position, and mark the symbol flip position as the second-order polarity inflection point to generate a set of key nodes that characterize the reversal of the moisture ratio change trend.

[0058] The first-order difference symbol sequence output from the preceding sub-step S2.1 is used as the operation object, and a logical XOR operation is performed on adjacent symbol bits as input pairs. For each pair of adjacent symbol bits, binary matching is performed on the symbol bit values, mapping the "increase" and "decrease" symbol combinations to numerical pairs, and the logical difference is determined by the XOR operation result. The positions with a value of 1 in the XOR operation result vector are defined as candidate nodes for symbol reversal, and the position index of the candidate node is recorded in conjunction with the time series index information. A continuity check is performed on the candidate node position indexes to eliminate false reversal events caused by zero-value noise, and a threshold filtering algorithm is used to retain nodes whose phase direction has truly reversed. The set of node positions after continuity check and noise removal is marked as a set of second-order polarity inversion points, and this set is structured and stored as key index data for subsequent curvature calculation. Through the above processing method, the first-order difference symbol sequence of the previous step is transformed into a set of key nodes containing only direction reversal information, realizing the standardized extraction of the polarity reversal feature of the moisture ratio change trend.

[0059] S2.3: Based on the local moisture ratio data segment and the second-order polarity inflection point, the local radius of curvature of each data point in the segment is calculated using the three-point circle center fitting algorithm, and the reciprocal of the local radius of curvature is taken to generate a local curvature change rate sequence that characterizes the degree of trend curvature.

[0060] The three-point circle fitting algorithm is applied to the set of second-order polarity inflection points output in step S2.2 and the local moisture ratio data fragments after standardization in step S1.4. Triples containing the target data point and its adjacent data points are selected as the fitting input. A Euclidean distance calculation matrix is ​​established between each pair of points using the coordinates of each triplet. Based on this, the corresponding chord length and the perpendicular bisector of the chord midpoint are constructed, and the intersection of the three sides is determined as the center coordinates of the fitted circle. The local radius of curvature is obtained by calculating the Euclidean distance between the center coordinates and the coordinates of any input point. During the calculation, triples containing polarity inflection points are given priority weights to improve the ability to capture curvature abrupt changes. The obtained radius of curvature is then divided by its reciprocal to obtain the rate of change of curvature, calculated using the following formula: Where k is the rate of change of curvature, and R is the local radius of curvature. To maintain numerical stability, amplitude limiting compensation is performed on results with a radius of curvature lower than a preset minimum observable radius threshold, replacing them with the corresponding rate of change of curvature to prevent numerical distortion caused by extremely small radii. The rate of change of curvature of all data points is arranged into a local rate of change of curvature sequence according to time series, and the sequence is filtered to remove isolated outliers, so that the trend of the rate of change of curvature is consistent with the actual moisture ratio dynamics. By using three-point circle center fitting and the reciprocal calculation of the radius of curvature, the polarity inflection points identified in the previous step and the geometric characteristics of the original data are transformed into a quantifiable rate of change of curvature sequence, achieving a precise numerical representation of the degree of trend curvature.

[0061] For example, the sampling frequency for moisture ratio data at the outlet of the wet mixer is set to 200Hz. Data points from 5ms before and after the sampling point are selected to form a triplet window. Euclidean distance is calculated for the coordinates of adjacent points, yielding chord lengths of 1.2, 1.1, and 1.3 (unit: standardized scale value). Based on these chord lengths, a triangle and its perpendicular bisector are constructed, resulting in the center coordinates (0.15, 0.22). The distance R from the center to the target point is 0.85. Substituting R into the formula... The curvature change rate was found to be 1.176. For the detected polarity inflection point triplet, the curvature change rate was increased to 1.25 to enhance the mutation response. A curvature change rate sequence of length 50 was continuously calculated. After outlier removal, the waveform was smooth and consistent with the trend of the rising moisture ratio inflection point. This sequence was ultimately used for subsequent three-dimensional feature vector construction, showing significantly improved trend recognition accuracy and mutation capture capability.

[0062] S2.4: Based on the first-order difference symbol sequence, the second-order polarity inflection point, and the local curvature change rate sequence, a three-dimensional feature vector containing the characteristics of change direction, inflection existence, and curvature intensity is constructed. The three-dimensional feature vector is then input into a predefined six-state classification rule base for matching and judgment to output the initial phase category label.

[0063] Based on the first-order difference symbol sequence, the second-order polarity inflection point, and the local curvature change rate sequence, the three types of input data are loaded into the independent channels of the feature construction unit to ensure that the correspondence between the direction of change, trend reversal, and curvature intensity in the time series is completely preserved.

[0064] Vector encoding is performed on the first-order difference symbol sequence, mapping the symbol "+" to the value 1, the symbol "-" to the value -1, and the symbol "0" to the value 0. This numerical representation enables the quantitative expression of the direction-of-change component.

[0065] A sparse matrix is ​​constructed for the set of second-order polarity inflection points. The inflection state at each time index position is marked as 1, and the non-inflection state is marked as 0, which serves as the binarized input for the inflection existence component. This binary matrix ensures that the phase reversal feature is aligned in the feature vector.

[0066] Normalization is performed on the local curvature change rate sequence, scaling the curvature change rate of each data point according to the extreme value range within a sliding window, so that the curvature intensity component is limited to the range [0,1], to unify the numerical scale with other components; the normalization formula is as follows: Where C is the rate of change of the original curvature, C min and C max These represent the minimum and maximum rates of curvature change within the local window, respectively.

[0067] The numerical change direction component, the binary inflection existence component, and the normalized curvature intensity component are horizontally concatenated according to the time index to form a three-dimensional feature vector matrix, with each row corresponding to a data point and each column corresponding to a feature component.

[0068] The three-dimensional feature vector matrix is ​​input into a predefined six-state classification rule base. The rule base contains matching conditions based on feature component thresholds and combinational logic. The initial phase category label of each data point is determined by row-by-row logic. This determination process depends on the combinational relationship between feature components. For example, when the direction of change is 1, the inflection existence is 0, and the curvature intensity is less than 0.2, it is matched as a stable rising category, ensuring that the identified phase conforms to the actual performance of trend dynamics.

[0069] Through the above processing method, the differential and curvature feature results of the previous step are transformed into a three-dimensional feature vector that can be directly matched by classification rules, thereby achieving accurate output of the initial phase category label.

[0070] For example, in a wet-mixing process, the length of a data segment representing the local moisture ratio of a batch of materials is set to 20 seconds. After quantifying the first-order difference symbol sequence change direction component, a data array of length 200 is obtained. After binarization, the set of second-order polarity inflection points has approximately 15 inflection markers at corresponding index positions. The local curvature change rate, after normalization, has a maximum value of 1.0, a minimum value of 0.0, and an average value of 0.35. The three components are concatenated into a 200×3 feature matrix according to the time index, and a six-state classification rule base is applied for matching. For example, the rule base defines that when the change direction is -1, the inflection existence is 1, and the curvature intensity is greater than 0.5, it is judged as an accelerating descent category. In this batch of data, the number of points matching this condition is approximately 30. Another rule is that when the change direction is 1, the inflection existence is 0, and the curvature intensity is between 0.1 and 0.3, it is judged as a stable ascending category. The number of points matching this condition is approximately 50. The initial phase category label output by logical determination can be stably mapped to a digital code after subsequent S2.5 smoothing and denoising processing, effectively improving the accuracy of trend analysis and spray control strategy linkage.

[0071] S2.5: Based on the initial phase category label, perform state smoothing and denoising processing to eliminate high-frequency phase jitter caused by sensor noise, and map the processed stable state to a unique digital code to generate the final basic phase state code that characterizes the dynamic characteristics of the current working condition.

[0072] like Figure 2 As shown, step S3: Map the basic phase state code to a two-dimensional coordinate system composed of the rate of change-dominant horizontal axis and the acceleration-dominant vertical axis to construct a discretized trend phase space coordinate point set. Specifically, this includes: S3.1: Obtain the basic phase state code generated by the previous steps and its corresponding first-order difference symbol sequence and second-order difference polarity inflection point data. Perform normalization processing based on the local curvature change rate values ​​within the sliding time window to generate standardized trend rate dominant component values ​​and trend acceleration dominant component values.

[0073] The basic phase state code generated in the preceding steps, along with its associated first-order difference symbol sequence and second-order difference polarity inflection point data, are obtained as the raw inputs for calculating the trend rate dominance and trend acceleration dominance components. A numerical scan is performed on the local curvature change rate sequence within the sliding time window to extract the maximum, minimum, and mean values, which are used as the basis for dynamically setting the normalization parameters. During the normalization process, the absolute value of the instantaneous rate of change is divided by the maximum absolute value of the instantaneous rate of change within the window to generate standardized trend rate dominance component values, achieved through the following formula: Among them, v i Let max(|v) be the first-order difference value of the i-th sampling point.j |) represents the maximum absolute value of all instantaneous rates of change within the window. Normalization of the dominant acceleration component values ​​involves dividing the absolute value of the local rate of change of curvature by the maximum absolute value of the local rate of change of curvature within the window to generate standardized trend acceleration dominant component values, achieved through the following formula: Among them, a i Let max(|a|) be the rate of change of local curvature at the i-th sampling point. j |) represents the maximum absolute value of all curvature change rates within the window. The normalized trend rate and trend acceleration dominant components are output as numerical matrices, which are then provided to the subsequent two-dimensional vector synthesis processing unit. By introducing dynamic maximum value calculation and absolute value operation during the normalization process, the preceding multidimensional feature results are transformed into standardized component data that possesses both comparability and numerical stability, achieving scale uniformity of the trend feature vector under different operating conditions.

[0074] S3.2: Based on the standardized trend rate dominant component value and trend acceleration dominant component value, a two-dimensional vector synthesis process is performed using an orthogonal coordinate mapping algorithm to generate an original trend phase space vector characterizing the dynamic characteristics of the current working condition.

[0075] Based on the standardized trend rate and trend acceleration dominant component values ​​obtained through S3.1 processing, a two-dimensional orthogonal mapping input matrix is ​​established as the execution object of this sub-step. The trend rate dominant component value is used as the horizontal axis input component, and after numerical reading, it is assigned to the first coordinate position of the two-dimensional vector. The trend acceleration dominant component value is used as the vertical axis input component, and assigned to the second coordinate position of the two-dimensional vector, forming an input pair with orthogonal component characteristics. An orthogonal coordinate mapping algorithm is used to define the unit basis vector corresponding to each coordinate component, ensuring that the horizontal and vertical axis vectors maintain an orthogonal relationship with a zero inner product in numerical space. Scalar multiplication is performed on the horizontal and vertical basis vectors, mapping the trend rate and trend acceleration dominant component values ​​proportionally to the basis vector directions, forming component vectors in two directions. Vector addition is performed on the two component vectors, and they are superimposed and synthesized in two-dimensional space to obtain the original trend phase space vector that fully represents the dynamic characteristics of the current working condition. The two-dimensional vector synthesis is completed using the following formula: in This is the product of the dominant trend rate component value and the horizontal axis unit vector. This is the product of the dominant component value of the trend acceleration and the unit vector of the vertical axis. Through the above orthogonal vector synthesis process, the standardized component values ​​from the previous step are transformed into the original trend phase space vector that can be directly used for spatial positioning and subsequent grid quantization, thus mapping the trend features to a unified spatial scale while maintaining the independence and orthogonality between different feature components.

[0076] S3.3: Perform discrete grid quantization processing on the original trend phase space vector, and perform coordinate rounding operation according to the preset change rate dominant horizontal axis resolution and acceleration dominant vertical axis resolution to generate discrete trend phase space grid coordinates with unique index identifier.

[0077] The original trend phase space vector output by S3.2 is used as the input object to read the preset parameters of the horizontal axis resolution dominated by the rate of change and the vertical axis resolution dominated by the acceleration.

[0078] The resolution ratio scaling operation is performed on the horizontal and vertical components of the original trend phase space vector to form a scaled two-dimensional component group.

[0079] Input the scaling value of the horizontal axis component into the rounding function, and perform down-rounding according to the preset resolution to generate the discrete coordinate index of the horizontal axis.

[0080] Input the scaling value of the vertical axis component into the rounding function, and perform downward rounding according to the preset resolution to generate the discrete coordinate index of the vertical axis.

[0081] Combine the discrete coordinate indices of the horizontal axis and the discrete coordinate indices of the vertical axis into a unique coordinate pair, and use this coordinate pair as the index key to establish a spatial grid coordinate identifier.

[0082] A hash encoding algorithm is used to generate a unique index for the spatial grid coordinate identifiers mentioned above, forming a standard index number for the spatial discretization of trend phase.

[0083] By performing discrete grid quantization, the trend space vector orthogonally synthesized in the previous step is mapped to discrete trend phase space grid coordinates with unique index identifiers, thereby realizing the fixed coordinate system required for subsequent spatial aggregation and topology verification.

[0084] For example, the horizontal component of the original trend phase space vector is set to 2.37, the vertical component to 4.86, the preset horizontal resolution for rate of change dominance is 0.5, and the vertical resolution for acceleration dominance is 0.25. The formula for calculating the horizontal scaling value is: in This is the scaling value for the horizontal axis. For the original horizontal axis components, This is the resolution on the horizontal axis. Substituting the parameters yields... The calculated result is 4.74. The formula for calculating the vertical axis scaling value is: in This is the scaling value on the vertical axis. The original vertical axis component, Let be the resolution of the vertical axis. Substituting the parameters, we get... The calculated result is 19.44. Rounding down the horizontal axis scaling value of 4.74 yields a horizontal axis discrete index of 4; rounding down the vertical axis scaling value of 19.44 yields a vertical axis discrete index of 19. These are combined into a coordinate pair (4,19), and hash-encoded to generate the index identifier H419. In actual testing, this operation ensures that the quantization accuracy of the trend phase space meets the preset resolution requirements and significantly improves the efficiency of subsequent spatial aggregation and topology consistency verification.

[0085] S3.4: Based on the correlation between the discretized trend phase space grid coordinates and the basic phase state code, perform a space coordinate point set aggregation operation to construct a discretized trend phase space coordinate point set containing phase attribute labels.

[0086] Based on the correlation between the discretized trend phase space grid coordinates and the basic phase state code, the correlation mapping table is called to read the state code corresponding to each grid coordinate as a phase attribute label, forming the attribute association data of the initial spatial coordinate point set.

[0087] The initial set of spatial coordinate points is grouped by phase attribute label. The label consistency judgment algorithm is used to aggregate grid coordinates with the same state code into the same set to ensure that the spatial points of the same phase form a coherent data cluster.

[0088] Within the group set, based on the numerical difference between the trend rate dominance horizontal axis and the trend acceleration dominance vertical axis, the Euclidean distance from each point to the centroid of the group is calculated. Coordinate points whose distance exceeds the preset difference threshold are selected and marked as boundary points for subsequent optimization of the aggregation pattern.

[0089] For the identified boundary points, perform neighborhood expansion aggregation to merge them with the coordinates of points with the same label in the neighborhood. Use a recursive search method to ensure that the set of boundary points maintains connectivity with the main cluster in the spatial structure, avoiding cluster splitting caused by local fluctuations.

[0090] The centroid location and member coordinate index of each group cluster are encapsulated into spatial aggregation entries with attribute descriptions to construct a discretized set of trend phase spatial coordinate points containing phase attribute labels.

[0091] Through the above aggregation process, the discretized grid coordinates and state encoding results from the previous step are transformed into a set of coordinate points with structured phase attributes, realizing the clustered expression of the trend phase space in the attribute dimension, and preparing an ordered data structure for subsequent topology consistency verification.

[0092] S3.5: Perform spatial topology consistency verification using the discretized trend phase spatial coordinate point set, remove isolated abnormal coordinate points caused by noise interference and correct the continuity of neighborhood coordinates, so as to output the final discretized trend phase spatial coordinate point set for subsequent map construction.

[0093] The discretized trend phase spatial coordinate point set input data is initialized by performing a neighborhood search, and a neighborhood set for each coordinate point is established according to the preset spatial radius parameter.

[0094] The number of neighboring points for each coordinate point is counted based on the neighborhood set. When the number of neighboring points is less than the noise judgment threshold, the coordinate point is marked as an isolated outlier.

[0095] Perform a culling operation on the coordinates of the points marked as isolated outliers, removing them from the original set of discretized trend phase space coordinates.

[0096] For the remaining coordinate points after elimination, calculate their spatial topological relationships and identify broken neighborhood connection chains.

[0097] The identified broken connection chains are corrected for continuity by generating completion points through interpolation algorithms and adding the completion points to the coordinate point set to restore topological continuity.

[0098] Re-execute the topology consistency check on the completed coordinate point set to confirm that the connection status of all neighborhood chains satisfies the continuity constraint.

[0099] Through the above spatial topology consistency verification and correction process, the discretized trend phase spatial coordinate point set of the previous step is transformed into the final coordinate point set that eliminates noise interference and maintains neighborhood continuity, thus providing a stable and reliable spatial data foundation for the subsequent construction of trend phase transition maps.

[0100] For example, in the wet mixing process monitoring scenario, the discretized trend phase spatial coordinate point set contains 200 two-dimensional coordinate points. The resolution of the horizontal axis dominated by the rate of change is set to 0.1, the resolution of the vertical axis dominated by acceleration is set to 0.1, the neighborhood radius parameter is set to 0.15, and the noise judgment threshold is set to 3. When performing neighborhood search, the K-nearest neighbor method is used to count the number of neighbors for each point. When the number of neighbors for a point is 2, it is marked as an isolated outlier and removed. After removing isolated outliers, it was found that 5 neighborhood connection chains were broken. Bilinear interpolation was used to generate complete points. After interpolation completion, the continuity of all chains was restored. The final output discretized trend phase spatial coordinate point set has 194 points. The spatial topology remains complete and stable, which significantly improves the accuracy and robustness of path identification in the subsequent trend phase transition map construction process.

[0101] like Figure 3 As shown, step S4: Based on the frequency statistics of transitions between the trend phase spatial coordinate points under historical operating conditions, a directed graph structure with weighted attributes is constructed to generate a trend phase transition map that records high-frequency evolution paths. Specifically, this includes: S4.1: Obtain the set of trend phase spatial coordinate points in the continuous time series, and perform adjacency pair extraction processing on the set of trend phase spatial coordinate points in chronological order to generate an original phase transition event sequence containing start phase code and end phase code.

[0102] The input conditions include the final discretized trend phase space coordinate point set output by step S3.5, which contains timestamps, phase attribute labels, and two-dimensional spatial index information. Based on this coordinate point set, it is first sorted in ascending order according to the timestamp field, so that the coordinate points in the same batch form a strict time series structure and maintain the integrity of the original acquisition order. Based on the sorting result, for any two adjacent records, the corresponding phase state code is retrieved by calling the index mapping table, forming a phase code pair consisting of a start code and a stop code. Adjacency pair extraction is performed on each pair of phase codes, and the phase code pair along with the time span information is encapsulated into a transition event record. This record is saved in the form of a four-tuple of start phase code, stop phase code, start time, and stop time. During the generation of transition event records, a window constraint mechanism is used to limit the phase jump detection range. When the time difference of the phase jump is less than the preset minimum change period threshold, the transition event is marked as suspicious and an "low confidence" label is attached. All transition event records are batch stored in the original event sequence cache queue to ensure that subsequent frequency statistics processing can directly traverse the event set in chronological order. By extracting adjacency pairs and encapsulating time spans, the results of the previous step are transformed into a standardized sequence of raw phase transition events, enabling the extraction of ordered transition paths of trend phase space coordinate points under historical operating conditions. For example, in a historical dataset of a wet-mixing production line, the trend phase space coordinate point set contains 5000 nodes, each with a millisecond-level timestamp and a basic phase state code. After ascending sorting, the average time span between adjacent nodes is 120ms. The system extracts start and end codes for each pair of adjacent nodes; for example, the start code is "accelerated rise," and the end code is "inflection point transition," and encapsulates transition event records with a start time of 1620000000000ms and an end time of 1620000000120ms. A window constraint mechanism is used, setting a minimum change period threshold of 100ms; events with a time difference less than this threshold are marked as low confidence. In this embodiment, the final original phase transition event sequence contains 4,700 high-confidence records and 300 low-confidence records. These data ensure the accuracy and reliability of path statistics in subsequent frequency statistics.

[0103] S4.2: Based on the original phase transfer event sequence, perform similar path aggregation statistical processing, and use the frequency statistics algorithm to calculate the cumulative number of occurrences of each unique phase transfer path under historical operating conditions, so as to generate a phase transfer frequency statistics table with frequency counting attributes.

[0104] The unique phase transition path refers to a non-repeating abstract transition pattern extracted from massive amounts of original phase transition events through aggregation and statistical analysis. This pattern is composed of a defined starting phase code and an ending phase code. The generation process is as follows: the system extracts and pairs phase spatial coordinate points from a continuous time series, generating a large number of specific instantaneous transition records "from phase A to phase B"; then, through path aggregation processing, all specific events are categorized according to their starting and ending phase codes, and each combination of starting and ending codes defines a unique transition path.

[0105] Obtain the original phase-transfer event sequence as input, specifying that each item in the event sequence contains a two-element combination of a start phase code and a stop phase code.

[0106] The event sequence is processed to ensure path uniqueness, mapping the combination of two elements to a standardized path identifier key value, ensuring that paths of the same type have consistent index references during the statistical process.

[0107] Based on the unique path identifier key value, a counting mapping table structure is established, with the key value as the main index and the corresponding cumulative occurrence count as the main value. The count is incremented by one for each key value by traversing the event sequence.

[0108] A frequency statistics algorithm is used to summarize the counting mapping table, and the cumulative occurrence of the same path identifier is aggregated to form a complete set of frequency information.

[0109] The frequency information set is converted into a two-dimensional data table structure with a path identifier field and a cumulative occurrence field, namely the phase transition frequency statistics table, which serves as the direct input for subsequent weight normalization calculations.

[0110] By aggregating statistical data from similar paths, the raw transfer event data from the previous step is transformed into a structured statistical table with frequency count attributes, thereby achieving a quantitative expression of the intensity of trend phase paths.

[0111] For example, under a historical operating condition of 120 minutes of continuous wet mixing process, the original phase transition event sequence contains 1480 records, with the starting phase code ranging from 1 to 6 and the ending phase code ranging from 1 to 6. After performing path uniqueness processing, 30 different path identifier key values ​​are formed, for example, the path identifier from code 2 to code 3 is "2-3". After establishing a counting mapping table, each event is traversed and counted, resulting in a cumulative occurrence count of 215 for path "2-3" and 178 for path "3-4". Using a frequency statistics algorithm, the mapping table is transformed into a two-dimensional data table, where the path identifier "2-3" corresponds to a cumulative occurrence count of 215, the path identifier "3-4" corresponds to a cumulative occurrence count of 178, and so on for the remaining paths. The generated phase transition frequency statistics table can be used to construct transition probability weight values ​​in subsequent steps. For example, when calculating the frequency probability value of path "2-3", assuming the total number of path occurrences is 1480, the probability weight value of approximately 0.145 is obtained by calculating 215 / 1480 using the following formula, which significantly improves the accuracy of high-frequency path identification and ensures the targeted configuration of spray control parameter response templates.

[0112] S4.3: Perform weight normalization calculation based on the cumulative occurrence count in the phase transfer frequency statistics table, and use a frequency probability mapping mechanism to convert the cumulative occurrence count into a transfer probability weight value that represents the possibility of path occurrence, so as to generate a phase transfer weighted dataset with probability weight attributes.

[0113] The input execution object is the phase transition frequency statistics table obtained by S4.2, which contains the unique phase transition path and its cumulative occurrence count.

[0114] The total frequency of occurrence of each phase transfer path in the statistics table is summed to obtain the total number of occurrences N of all paths under historical operating conditions.

[0115] Based on the cumulative number of occurrences for each path Total number of occurrences Construct the frequency probability mapping formula: in Let be the transition probability weight value for path i.

[0116] Perform normalization calculations on the results of the above formulas, and then normalize all... Adjust to meet The probability distribution is constrained, and the normalized weight values ​​are kept to a specified number of decimal places to control the calculation precision.

[0117] The starting phase code, ending phase code, and normalized probability weight value of each path are encapsulated into transition record entries with probability weight attributes, and then summarized to form a phase transition weighted dataset.

[0118] By using frequency probability mapping and normalization, the counting results of S4.2 are transformed into transition probability weight data that characterize the likelihood of path occurrence, thus realizing the quantitative weight definition before constructing the trend phase transition map.

[0119] For example, in the case of wet-mixed grinding wheel forming material, the statistics table contains 12 unique phase transfer paths, with a total occurrence count N of 2400. For path 3, its cumulative occurrence count C i The value is 300, so substitute it into the formula. The initial probability is 0.125. After normalizing all path probabilities, the weight value of path 3 remains 0.125. This weight, along with the starting phase code "02" and the ending phase code "04", forms a transition record entry and is stored in the dataset. In the validation process, the transition-weighted dataset supports the directed graph generation algorithm to quickly locate high-probability paths, significantly improving the efficiency and stability of trend phase transition map construction.

[0120] S4.4: Construct topological connections using the phase transition weighted dataset, and map phase codes to nodes and transition probability weights to directed edge weights using a directed graph generation algorithm to generate an initial trend phase transition map that records the full evolution path and its intensity.

[0121] The phase-transition weighted dataset containing the start phase code and the end phase code is obtained as input. The data parsing unit of the directed graph generation algorithm is called to convert the phase code values ​​into node instances, and a unique index is established in the node instance to facilitate subsequent edge connection retrieval.

[0122] Based on the transition probability weight values ​​in the phase transition weighted dataset, a mapping table is constructed, and the weight values ​​corresponding to each pair of start and end nodes are stored as attribute data of directed edges in the edge list of the graph structure to be generated.

[0123] During the construction of the edge list, a topological connection relationship check is performed to ensure that every node has at least one incoming or outgoing edge, and isolated nodes that do not meet the conditions are either added or removed to maintain topological closure.

[0124] The weighted directed graph generation routine is called, and the node index table and directed edge weight table are used as inputs. The graph generation algorithm is used to construct an initial trend phase transition graph containing the full trend evolution path. The graph uses the set of nodes to represent the trend phase state and the set of edges and their weight attributes to represent the intensity of the phase transition.

[0125] During the generation process, the transition probability weights are normalized to the [0,1] interval using a mathematical mapping method. The mapping formula is as follows: Where w is the normalized weight value, p is the original transition probability, and p min With p max These are the minimum and maximum transition probabilities in the current dataset, respectively, to ensure the numerical consistency of the weights in graph visualization and path calculation.

[0126] Through the above topological connection and weight mapping processing, the phase transition weighted dataset from the previous step is transformed into an initial trend phase transition graph containing all node attributes and edge weights, thus realizing a complete mapping of the trend evolution path in the structured graph model.

[0127] For example, in the historical dataset of wet mixing processes, there are basic phase state codes A1 to A6. The phase transition weighted dataset records a transition probability of 0.75 for A2→A3, 0.62 for A3→A4, and 0.81 for A4→A5. After performing the node mapping operation, node instances N2, N3, N4, and N5 are generated, and edges E(2,3), E(3,4), and E(4,5) and their weight values ​​are stored in the edge list. A topology connection check is called to confirm that all the above nodes are closed connections. Calculations are performed according to the normalized formula, with p... max =0.81, P min Taking =0.62 as an example, the normalized weights corresponding to A2→A3 are calculated as follows: The result is 0.684, which is mapped to the edge weight attribute E(2,3) in the graph. The generated initial trend phase transition graph contains all high-frequency paths and weight distributions, which can significantly improve the accuracy and stability of typical evolutionary chain extraction in subsequent high-frequency path screening and structure optimization.

[0128] S4.5: Perform high-frequency path filtering and structural optimization processing on the initial trend phase transition map, filter low-probability transition edges and retain high-confidence connection paths according to a preset threshold, so as to generate a final trend phase transition map that only records high-frequency evolution paths and has dynamic update capability.

[0129] Obtain all directed edge data and their corresponding transition probability weights from the initial trend phase transition map, construct a weight threshold judgment matrix for screening, and use this matrix as the input condition for high-frequency path screening.

[0130] Based on the weight threshold judgment matrix, a threshold comparison operation is performed. Directed edges with transition probability weight values ​​lower than a preset threshold are marked as low-confidence paths and removed from the topology to eliminate low-probability connection paths that interfere with high-frequency evolution analysis.

[0131] The connectivity of the directed graph structure after removing low-confidence paths is checked. The adjacency matrix expansion breadth-first search algorithm is used to identify local subgraphs with isolated nodes or broken links, and the neighborhood reconstruction strategy to complete the connection path is invoked to restore topological continuity.

[0132] A dynamic update mechanism is implemented for the reconstructed directed graph structure, which reconstructs the trend phase transition events in the latest sliding time window into the connection relationship between nodes, and uses an incremental statistical model to correct the transition probability weight value in real time, so as to realize the graph's adaptive update capability to changes in operating conditions.

[0133] The dynamically updated directed graph structure is compared with the historical high-frequency path set, and the path consistency judgment logic is called to ensure that the final trend phase transition map only contains the set of high-frequency evolution paths that meet the high confidence condition and have temporal continuity.

[0134] By using high-frequency path filtering and structural optimization, the initial trend phase transition map results from the previous step are transformed into a final trend phase transition map that records only high-frequency evolution paths and has dynamic update capabilities, thereby achieving stability in trend evolution pattern recognition and accuracy in control parameter mapping.

[0135] For example, in the online monitoring scenario of the wet mixing process of grinding wheel forming material, the initial trend phase transition map contains 42 directed edges with weight values ​​ranging from 0.01 to 0.87, and the preset high-frequency path screening threshold is set to 0.25. Eighteen directed edges with weight values ​​less than 0.25 are removed to eliminate the risk of low-frequency connection paths interfering with the control strategy. Topological connectivity is checked on the nodes corresponding to the remaining 24 directed edges, revealing three isolated nodes and two broken links. Neighborhood reconstruction strategies are then invoked to complete the connections, restoring the node degree distribution to a stable state with a mean of 2.8. In the dynamic update mechanism, trend phase transition event data within the most recent 300-second sliding time window is used as input, and an incremental statistical model is employed to calculate the directed edge weights using the formula: in, The updated weight values, The weight values ​​before the update. To increase the number of transfer events, The historical cumulative event count is used to achieve a smooth adjustment of the weight values. In this embodiment, the average weight value of the directed edges after the update is significantly improved compared to the previous value. The final trend phase transition map retains only 17 high-confidence paths, which significantly improves the response accuracy of the spray strategy and the robustness of the system's closed-loop control in subsequent control parameter mapping.

[0136] Step S5: Based on the typical evolutionary chains in the trend phase transition spectrum where the cumulative occurrence exceeds a preset threshold, define a dedicated spray control parameter response template library containing parameter adjustment directionality strategies and step-by-step incremental rules. Specifically, this includes: S5.1: Threshold filtering is performed on the weight attributes of each directed edge in the trend phase transition map to extract typical evolution chain sequences whose cumulative occurrence exceeds a preset threshold, thereby establishing a set of high-confidence working condition evolution paths.

[0137] The weight attribute data of each directed edge in the final trend phase transition map is obtained, and the cumulative occurrence count of each directed edge is determined as the path statistics input. For the cumulative occurrence count of each directed edge, a threshold filtering algorithm is called, and a comparison operation is performed based on a preset threshold parameter. Paths with a cumulative occurrence count higher than the threshold are marked as candidate high-confidence paths. Phase chain recombination processing is performed on the marked candidate high-confidence paths, and the phase codes are continuously connected to form a multi-node sequence to form a typical evolutionary chain structure set that conforms to evolutionary logic. The uniqueness verification and redundancy removal of the formed typical evolutionary chain structure set are performed. Repeated or highly overlapping chains are removed by calculating the similarity of relationships between nodes, ensuring the independence and representativeness of the chains within the set. The final typical evolutionary chain structure set is used as the direct input object for subsequent parameter adjustment directional strategy matching and step-by-step incremental rule calculation. Through the extraction method of the high-confidence path set, the trend phase transition map result of the previous step is transformed into stable operating condition evolution data for spray control logic mapping, achieving accurate operating condition selection in the control strategy design stage.

[0138] For example, in the continuous operation scenario of wet mixing process of grinding wheel forming material, the trend phase transition map contains 180 directed edges, and the cumulative occurrence count of each edge ranges from 5 to 420. The preset threshold parameter is 100, which means that the path with a cumulative occurrence count of not less than 100 is identified as a high-confidence path. A filtering operation is performed on the cumulative occurrence count data. For example, for a directed edge with a starting phase code of 3 and an ending phase code of 5, its cumulative occurrence count is 265, which is greater than the threshold of 100, so it is included in the candidate high-confidence path. This path is connected sequentially with the subsequent paths with phase codes of 5 to 7 (cumulative occurrence count of 198) and the path with phase codes of 7 to 2 (cumulative occurrence count of 156) to form a typical three-node evolution chain [3→5→7→2]. For the multiple evolutionary chain structures formed, identical chains are eliminated by comparing the hash index of the phase sequence. Chains with the same sequence length and a node difference rate of less than 0.1 are merged, ultimately resulting in a set of typical evolutionary chains with high independence and strong representativeness. For example, one evolutionary chain in the set [1→4→6→3] has occurred 105, 112, and 118 times in historical operating conditions, with a total chain frequency of 435, providing a stable trend basis for subsequent matching of spray flow and pressure adjustment direction strategies.

[0139] S5.2: Based on the initial phase encoding and the final phase encoding in the typical evolutionary chain sequence, perform phase transition feature mapping analysis to generate a phase transition feature vector characterizing the coupling relationship between the rate of change of water ratio and the acceleration.

[0140] Based on the typical evolutionary chain sequences filtered by S5.1, the initial and final phase codes of each evolutionary chain are used as input conditions, and a mapping relationship is established with the velocity-dominant and acceleration-dominant component values ​​recorded in the corresponding trend phase space. The difference between the trend velocity component value corresponding to the initial phase code and the trend velocity component value corresponding to the final phase code is calculated to obtain the velocity change component Δv; simultaneously, the difference between the trend acceleration component value corresponding to the initial phase code and the trend acceleration component value corresponding to the final phase code is calculated to obtain the acceleration change component Δa. Δv and Δa are normalized using the maximum absolute value normalization method to ensure consistent numerical scales between different chains, so that subsequent coupling analysis is not affected by dimensions. The normalized Δv and Δa are used as two-dimensional vector components to construct the coupling relationship unit, and the coupling strength parameter K is generated using the product interaction term; this calculation follows the following formula: Here, Δv is the normalized velocity change component, and Δa is the normalized acceleration change component. K and the original components form a triple <Δv, Δa, K> as the core element of the phase transition feature vector. This triple is then combined with the directionality of the phase transition in the evolutionary chain (the increase or decrease of the horizontal and vertical coordinates from the initial phase to the final phase) to encode the direction label, forming a complete phase transition feature vector data structure. Through this processing method, the typical evolutionary chain structure features from the previous step are transformed into a quantitative coupling relationship vector, realizing the trend dynamic feature input required for adjusting the spray control strategy parameters.

[0141] For example, in a typical evolution chain of a wet mixing process, the trend rate component value corresponding to the initial phase encoding is set to 0.45, the trend rate component value corresponding to the final phase encoding is set to 0.60, the trend acceleration component value corresponding to the initial phase encoding is set to -0.20, and the trend acceleration component value corresponding to the final phase encoding is set to 0.10. The difference is calculated to yield Δv = 0.15 and Δa = 0.30. Maximum absolute value normalization is performed on Δv and Δa. Assuming the maximum absolute values ​​are 0.20 and 0.50 respectively, the normalized result Δv... norm It is 0.75, Δa norm The value is 0.60. Substituting the normalized result into the MathML formula above... Calculations show that the coupling strength parameter K is 0.45. Δv norm , Δa norm Together with K, they form a feature vector {0.75, 0.60, 0.45}, with the directional labels "increasing rate, increasing acceleration" appended, forming a complete phase transition feature vector. This feature vector can significantly improve the matching accuracy of the spray flow and pressure adjustment direction when querying the strategy rule base in S5.3, and ensure that the control strategy can intervene in advance when the rate and acceleration increase simultaneously, significantly improving the foresight and accuracy of the closed-loop response in the wet mixing process.

[0142] S5.3: Utilize the phase transition feature vector to query the preset strategy rule base, perform parameter adjustment directionality strategy matching operation, and generate spray flow regulation direction identifier and spray pressure regulation direction identifier for the current typical evolution chain.

[0143] For the phase transition feature vector output by S5.2, a pre-set policy rule base is loaded and the feature vector is used as the retrieval input condition. The feature matching algorithm is called to perform the preliminary screening operation of the policy template candidate set, and template entries with the same combination of rate dominance sign and acceleration dominance sign are included in the candidate set.

[0144] The policy rule base is a pre-defined, structured set of condition-action mapping knowledge used to quantify abstract process state characteristics into specific control commands. Its core components are a large number of decision rules derived from the experience of domain experts and historical best-case data analysis. Each rule explicitly specifies the direction of the system's spray flow rate and spray pressure adjustment—such as increasing, decreasing, or maintaining—when the process dynamics represented by the input phase transition feature vector, such as transition severity, energy distribution shift, and dwell stability, meet a set of pre-defined quantification conditions (threshold range).

[0145] For each template entry in the candidate set, feature vector similarity is calculated. The weighted Euclidean distance formula is used to quantify the matching degree between the current feature vector and the template feature vector. The weights of each component are adjusted based on historical operating parameters to adjust sensitivity settings. The specific calculation formula is as follows: in, This is the similarity distance value. The weights of the components of the current feature vector in the i-th dimension are given. This represents the component value of the current feature vector in the i-th dimension. Let be the component value of the template feature vector in the i-th dimension, and n be the total number of dimensions of the feature vector.

[0146] Template entries with similarity distance values ​​lower than the preset matching threshold are subjected to secondary matching verification. Their historical execution effects are analyzed through logical rules to evaluate the effectiveness of their directional adjustments under the current evolutionary chain conditions.

[0147] Based on the verification results, select the template item with the highest matching degree and the directional adjustment effectiveness score greater than the safety threshold, and extract the spray flow adjustment direction identifier and spray pressure adjustment direction identifier of the item.

[0148] The extracted directional identifiers are verified by the symbol consistency check logic to ensure that the spray flow regulation direction and the pressure regulation direction form a consistent trend response strategy under the coupling relationship of rate dominance and acceleration dominance.

[0149] Through the above matching and verification process, the phase transition feature vector from the previous step is transformed into a directional parameter identifier that can directly characterize the adjustment tendency of the spray actuator, thereby achieving the technical effect of adaptive decision-making for spray flow and spray pressure direction.

[0150] For example, in the wet mixing process monitoring system, the strategy rule base contains 120 template entries, with the rate-dominant component weight set to 1.5, the acceleration-dominant component weight set to 2.0, and the matching threshold set to 0.8. The input phase transition feature vector is [0.6, -0.9]. The distance values ​​calculated between each component and the template vector in the weighted Euclidean distance formula are 0.45, 0.72, 0.68, etc. Templates with distance values ​​lower than 0.8 are included in the secondary verification. In the secondary verification, the historical execution effect score of this feature vector and template entry #47 reaches 0.92, and the directional adjustment strategy is "reduced" for spray flow adjustment direction and "increased" for spray pressure adjustment direction. After sign consistency verification, this combination can effectively suppress further decrease in moisture ratio under the transition condition of acceleration descent → inflection point. Finally, the output directional indicator data command is used by S5.4 for calling. The application effect is to maintain stable material moisture content in high humidity disturbance scenarios and significantly improve spray control accuracy.

[0151] S5.4: Based on the distribution data of the average residence time of the typical evolutionary chain sequence under historical operating conditions, perform step-by-step incremental calculation to generate a pressure reference value increment coefficient sequence that increases step by step with the extension of phase residence time.

[0152] Obtain the distribution data of the average residence time of typical evolutionary chain sequences under historical operating conditions, and use it as the time reference for calculating the incremental pressure benchmark value.

[0153] The average dwell time distribution data is segmented into intervals. Based on the dwell time statistics of each phase type, a set of phase dwell time interval boundaries is generated as the trigger node for the step-by-step increasing rule.

[0154] The residence time of the current phase type is input into the interval boundary set for boundary matching calculation. The interval level of the residence time is determined, and a pressure increment amplification coefficient index is generated based on the level.

[0155] The incremental coefficient calculation formula is used to amplify the pressure reference value through multi-level multiplier factors. The formula is as follows: Where ΔP is the pressure baseline increment, K is the multiplier factor corresponding to the interval level, Δt is the difference between the current residence time and the average residence time of the phase, and α is the phase type sensitivity coefficient.

[0156] A stepped pressure increment coefficient sequence is constructed based on the ΔP values ​​calculated for each interval level, and this sequence is arranged in ascending order over time to maintain the stability of the execution logic.

[0157] Through the above processing method, the typical evolutionary chain residence time data in the previous step is transformed into a pressure benchmark value increment coefficient sequence that increases step by step with the extension of residence time, so as to achieve the expected technical effect of automatic adjustment of spray pressure with trend.

[0158] For example, during the wetting process of grinding wheel forming material in a wet-mixing process, the average residence time of the accelerated rising phase under historical working conditions is obtained as 4.5 seconds. The residence time interval boundaries are set as three level intervals: [0–3 seconds], [3–5 seconds], and [>5 seconds], corresponding to K values ​​of 0.8, 1.0, and 1.3, respectively. When the current residence time is 5.8 seconds, Δt is calculated as 1.3 seconds, and the sensitivity coefficient α is set to 0.5, the pressure increment ΔP is calculated as follows: The calculated result is 0.845, indicating that the pressure reference value will be increased by 0.845 pressure step units at that moment. For cases where the dwell time falls within different ranges, such as a dwell time of 2.7 seconds and Δt of -1.8 seconds, the corresponding K value is 0.8, and the calculated ΔP is -0.72. The system will then decrease the spray pressure reference value by 0.72 step units. This rule has been verified under various environmental conditions, significantly improving the response and accuracy of spray pressure adjustment in high humidity disturbance scenarios.

[0159] S5.5: Integrate the spray flow rate adjustment direction indicator, spray pressure adjustment direction indicator, and pressure reference value increment coefficient sequence, and perform a dedicated spray control parameter response template encapsulation operation to generate a dedicated spray control parameter response template library containing complete dynamic tuning logic.

[0160] Step S6: Monitor the residence time of the basic phase state code in the trend phase space at the current moment in real time, and combine the historical average residence time comparison results with the multi-scale fluctuation energy ratio to generate a triplet control command containing the phase code, residence level, and evolutionary chain confidence. Specifically, this includes: S6.1: Obtain the basic phase state code at the current moment and the starting timestamp of entering the trend phase space, calculate the current phase dwell time based on the current system clock, and use the current phase dwell time as the benchmark input data for subsequent time series comparison analysis.

[0161] S6.2: Read the historical average dwell time data corresponding to the basic phase state code at the current time in the trend phase transition map, and use the time difference calculation algorithm to compare the current phase dwell time with the historical average dwell time to generate a dwell time deviation coefficient that characterizes the strength of the current trend persistence.

[0162] The historical average dwell time data corresponding to the basic phase state code at the current moment in the trend phase transition map is read as a reference benchmark value for time series comparison.

[0163] The current phase dwell time calculated by step S6.1 and the historical average dwell time are simultaneously input into the time difference calculation module.

[0164] This module uses a difference calculation logic to subtract the values ​​of the two to form the original dwell time difference data.

[0165] The original dwell time difference data was normalized, and the historical average dwell time was used as the normalization denominator to eliminate the influence of different time scales, thus obtaining the relative deviation value of dwell time.

[0166] The relative deviation value of the dwell time is input into the dwell duration strength determination unit, and a quantified dwell time deviation coefficient is generated by threshold comparison.

[0167] The dwell time deviation coefficient is calculated using the following formula: Where ΔT is the dwell time deviation coefficient, T is the current phase dwell time, and T avg This represents the historical average dwell time.

[0168] By using the above processing method, the current phase dwell time in the previous step is transformed into a dwell time deviation coefficient that characterizes the strength of trend persistence, thereby achieving the expected technical effect of quantitatively determining trend persistence.

[0169] For example, during the operation of the wet mixing process, the current phase dwell time obtained from a certain monitoring is 14.6 seconds, and the historical average dwell time of this phase in the trend phase transition spectrum is 10.0 seconds. Inputting these two values ​​into the time difference calculation module yields a raw dwell time difference of 4.6 seconds. In the normalization process, using 10.0 seconds as the normalization denominator, the relative deviation of the dwell time is calculated to be 0.46. This relative deviation value is input into the persistence strength determination unit, with a determination threshold of 0.30. Since 0.46 is greater than the threshold, the deviation coefficient ΔT is determined to be of a high persistence level. Combined with subsequent multi-scale fluctuation energy ratio verification, this high persistence level is used to improve the evolution chain confidence score. The dwell level label in the final output triplet control command is set to enhanced mode, which helps the spray control module to expand the pressure adjustment range in advance, improving the response speed and control accuracy to trend changes in the wet mixing process.

[0170] S6.3: Extract the original moisture ratio signal sequence of the short-time window containing the basic phase state code of the current moment, apply the multi-scale wavelet decomposition algorithm to extract the signal fluctuation components at different scales, and calculate the multi-scale fluctuation energy ratio based on the energy spectral density of each fluctuation component to quantify the noise interference level and abrupt change characteristics of the current trend signal.

[0171] The multi-scale fluctuation energy ratio directly characterizes the uniformity and stability of the energy distribution of the signal in the multi-scale frequency band: the lower the check value, the more stable the energy distribution at each scale, and the less the signal is affected by noise interference or sudden disturbances; the higher the check value, the more drastic the energy distribution fluctuations, and the more significant the abnormal disturbances or state change characteristics in the signal.

[0172] Based on the acquired basic phase state code at the current moment and its short-time window original moisture ratio signal sequence, the time span of the sequence is determined as the input time domain range for multi-scale analysis.

[0173] Discrete wavelet decomposition is performed on the original moisture ratio signal sequence of the short time window. A Daubechies-type wavelet basis with tight support and smoothness is selected to decompose the signal at a preset scale level, separating the signal into approximate components and detail components, and outputting the coefficient set scale by scale.

[0174] The energy spectral density is calculated separately for the approximate and detail components at each scale. The summation is performed using squared coefficients and normalized to the unit time. The formula for calculating the energy spectral density is as follows: Where c s,i represents the wavelet coefficients at this scale, s is the scale index, and N is the total number of wavelet coefficients at scale s.

[0175] Perform inter-scale ratio calculations on energy spectral densities at different scales to establish a multi-scale fluctuation energy ratio vector. The calculation formula is as follows: in For the current scale index, For reference scale index, This represents the energy spectral density.

[0176] The statistical mean and standard deviation of the multi-scale fluctuation energy ratio vector are calculated, and the maximum difference from the reference threshold is extracted as a comprehensive index to quantify the noise interference level and abrupt change characteristics of the current trend signal.

[0177] Through the above calculation and statistical analysis of the multi-scale fluctuation energy ratio, the dwell time deviation coefficient obtained in the previous step is quantified and supplemented into trend noise judgment data, realizing a numerical description of the stability and abruptness of the trend signal, and outputting the multi-scale fluctuation energy ratio index for subsequent trend credibility verification.

[0178] For example, the short-time window moisture ratio signal collected by the near-infrared moisture sensor at the outlet of the wet mixer is set to 256 sampling points with a time span of 2.56 seconds. A 5-level discrete wavelet decomposition using the Daubechies-4 wavelet basis is performed to obtain the approximation coefficients and detail coefficients at each scale. The energy spectral density from scale 1 to scale 5 is calculated, where the energy value of scale 1 is 0.85, the energy value of scale 5 is 0.12, and the reference scale is set as scale 3 with an energy value of 0.45. The energy ratio at scale 1 is calculated as follows: The energy ratio at scale 5 is The energy ratio vector was calculated to have a mean of 1.02, a standard deviation of 0.65, and a maximum deviation of 1.44, which served as a comprehensive indicator for judging noise interference and trend abrupt changes. During the wet mixing process of different batches of molding materials, if the maximum deviation exceeded the preset abrupt change threshold of 1.2, the probability of a trend abrupt change was significantly increased; otherwise, a high-confidence evolution chain score was maintained. Finally, a multi-scale fluctuation energy ratio index was output and passed on to subsequent steps, significantly improving the coordination between trend analysis and control decisions.

[0179] S6.4: Determine the initial residence level label based on the residence duration deviation coefficient, and perform trend confidence verification logic in conjunction with the multi-scale fluctuation energy ratio. If the multi-scale fluctuation energy ratio exceeds the preset mutation threshold, reduce the evolution chain confidence score; otherwise, maintain a high confidence score, thereby generating an evolution chain confidence index that has been corrected for noise robustness.

[0180] Based on the input residence duration deviation coefficient data, a residence level mapping table is invoked for numerical indexing, mapping the intervals where the deviation coefficients fall to the initial residence level labels to complete the pre-coding of the trend persistence level. Threshold comparison processing is performed on the acquired multi-scale fluctuation energy ratios, using a mutation threshold criterion matrix to determine whether the energy distribution of the current trend curve at each scale shows significant anomalies. The threshold comparison results are logically combined with the residence level labels to construct a trend confidence verification condition set. If the energy ratio anomaly criterion is met, a confidence decay function is invoked to generate a reduced evolutionary chain confidence score. When executing the confidence decay function, a linear decay mode is used, and the decay coefficient is mathematically calculated. Where C is the original confidence score, k is the attenuation ratio coefficient, E is the actual multi-scale fluctuation energy ratio, and T is the mutation threshold. When the energy comparison result does not trigger the anomaly criterion, the original confidence score is maintained and marked as a high-confidence state. Through the above adaptive attenuation or maintenance processing method, the residence level label from the previous step is combined with the multi-scale fluctuation energy ratio to transform it into an evolutionary chain confidence index that includes noise robustness correction, achieving the expected technical effect of dynamic assessment of trend confidence.

[0181] For example, in a scenario of continuous and stable operation of a wet-mixing process, the current phase dwell time is 45 seconds, and the historical average dwell time is 40 seconds. The dwell time deviation coefficient calculated using the time difference is 0.125. Indexing the dwell level mapping table, the deviation coefficient falls within the range of 0.1 to 0.15, corresponding to an initial dwell level label set to "Level 2". The measured value of the multi-scale fluctuation energy ratio within the short-term window is 1.35, and the mutation threshold is set to 1.50. The comparison result shows that the energy ratio is lower than the mutation threshold, and no anomaly criterion is triggered; the confidence score remains at its original value of 0.92. After encapsulating the confidence index, the confidence field of the triplet is 0.92, the dwell level field is "Level 2", and the phase encoding is a stable maintenance state. In another batch of wet mixing, the residence time deviation coefficient was 0.2, corresponding to the initial residence level label set to "Level 3". The measured value of the multi-scale fluctuation energy ratio was 1.65, and the mutation threshold was also set to 1.50. The anomaly criterion was triggered, and the decay function was called to calculate the decay coefficient as (1-0.05×(1.65-1.50))=0.9925. The original confidence score of 0.88 decayed to 0.8724, and the confidence field after encapsulation was 0.8724. This ensured that the impact of noise mutation was effectively smoothed, improving the stability of trend recognition and the reliability of spray control decisions.

[0182] S6.5: Integrate the basic phase state code, dynamic dwell stability label, and evolution chain confidence index at the current moment, and assemble them according to the predefined triplet data structure encapsulation protocol to output a triplet control command containing phase code, dwell level, and evolution chain confidence, which serves as a direct decision variable for dynamically tuning the spray control parameters. The dynamic dwell stability label is obtained by modifying the initial dwell level label.

[0183] The initial residence level label is modified to obtain the dynamic residence stability label. Specifically, if the multi-scale fluctuation energy ratio exceeds a preset mutation threshold, it indicates that there is a violent disturbance or an imminent transition within the current phase state. The system will then lower the confidence level that the current state will remain stable, thereby correcting the initial residence level label downwards, for example, from stable to on the verge of instability. If the multi-scale fluctuation energy ratio does not exceed the threshold, the initial residence level label is maintained.

[0184] Step S7: Based on the triplet control command, retrieve matching items from the dedicated spray control parameter response template library, and combine the nozzle response delay upper limit and the minimum adjustable pressure step physical constraint boundary to perform parameter safety mapping to generate a dynamically tuned spray control parameter set. Specifically, this includes: S7.1: Obtain the triplet control command containing phase encoding, residence level and evolution chain confidence. Use the phase encoding as the index key to perform a hash retrieval operation on the dedicated spray control parameter response template library to extract the original parameter adjustment directional strategy and step-by-step incremental rule set that match the evolution path of the current working condition.

[0185] S7.2: Based on the original parameters, adjust the directional strategy and the step-by-step incremental rule set, and perform linear interpolation calculations in combination with the dwell level values ​​in the triplet control command to generate a preliminary spray flow reference value and a preliminary spray pressure adjustment amount without physical constraint correction.

[0186] S7.3: Obtain the physical constraint boundary data of the nozzle response delay upper limit and the minimum adjustable pressure step size, use the saturation limiting algorithm to perform threshold truncation processing on the preliminary spray pressure adjustment amount, and perform time-series matching verification between the truncated pressure change rate and the nozzle response delay upper limit to generate a limited spray pressure adjustment amount that conforms to the dynamic characteristics of the hardware.

[0187] Obtain the physical constraint parameter dataset of the nozzle hardware, including quantitative values ​​of the upper limit of nozzle response delay and the minimum adjustable pressure step, and use it along with the initial spray pressure adjustment amount without physical constraint correction as input objects.

[0188] The initial spray pressure adjustment is compared with the upper limit of the response delay. A saturation limiting algorithm is used to truncate the adjustment that exceeds the physical threshold, ensuring that the adjustment range does not exceed the safe response range of the nozzle.

[0189] Construct the adjustment limit formula in the saturation limiting algorithm, for example: Among them, P lim P is the pressure regulation amount after cutoff. adj P is the initial pressure regulation amount. max The maximum safe pressure change rate value corresponding to the upper limit of response delay.

[0190] The truncated pressure regulation amount is time-matched and verified with the upper limit of the nozzle response delay. The theoretical pressure response curve is calculated based on the hardware delay time Δt and the rate of change of the regulation amount, using the formula: Among them, R rateΔt represents the pressure change rate and the nozzle response delay time. The check is to determine whether the change rate is within the acceptable dynamic characteristic range of the hardware.

[0191] If the rate of change exceeds the hardware's allowable range, a secondary limiting process is performed to reduce the rate of change to the upper limit acceptable to the device, generating a limited spray pressure adjustment amount that conforms to the hardware's dynamic characteristics.

[0192] By using saturation limiting and timing matching verification, the initial pressure adjustment amount from the previous step is transformed into a restricted spray pressure adjustment amount that meets the constraints of nozzle response delay and minimum adjustable pressure step size, thereby ensuring the safety and stability of control parameters during dynamic execution.

[0193] S7.4: Based on the physical constraint boundary of the limited spray pressure adjustment amount and the minimum adjustable pressure step size, perform discretization quantization and rounding operation to map the continuous pressure values ​​to standard pressure step levels that can be recognized by the device controller, so as to generate the final standardized spray pressure control parameters.

[0194] Based on the physical constraints of the limited spray pressure adjustment amount and the minimum adjustable pressure step size, when loading the parameter mapping algorithm for discretization quantization and rounding, the limited spray pressure adjustment amount is used as the initial condition for establishing the mapping process as a continuous input value.

[0195] For the preset pressure step level table and corresponding control register encoding of the continuous input value reading system, the discretization and quantization module is called to form a layered resolution interval based on the step size boundary setting, and the center value of the current interval is matched with the numerical interval of the limited spray pressure adjustment as the candidate level pressure value.

[0196] After the candidate pressure values ​​are formed, a quantization and rounding operation is performed. The candidate values ​​are fully aligned to the physically achievable standard pressure values ​​through rounding or rounding down rules, ensuring that the pressure adjustment is compatible with the hardware controller.

[0197] The pressure gear value mapping table is invoked to convert the rounded pressure gear value into a register input code that the device controller can recognize, and a standardized spray pressure control parameter data field is generated.

[0198] Perform a consistency check on the generated standardized spray pressure control parameters to confirm that they simultaneously meet the minimum adjustable pressure step size constraint and the nozzle response delay constraint, ensuring that the mapped parameters are both executable and do not cause hardware anomalies.

[0199] By using discretization, quantization, rounding, and register encoding mapping, the restricted spray pressure adjustment amount from the previous step is transformed into the final standardized spray pressure control parameters that conform to the device controller protocol, thereby achieving safe and controllable output of spray pressure under physical constraints.

[0200] S7.5: Integrate the preliminary spray flow reference value with the final standardized spray pressure control parameters, and assemble the data structure according to the industrial fieldbus communication protocol encapsulation format to generate a dynamically tuned spray control parameter set that can directly drive the spray actuator.

[0201] Step S8: Drive the spray actuator using the dynamically tuned spray control parameter set, and re-input the new moisture ratio feedback signal after the action to the sliding interception stage to complete the iterative update of the closed-loop control process. Specifically, this includes: S8.1: Obtain the dynamically tuned spray control parameter set as input conditions, and use the industrial fieldbus protocol to parse the flow reference value, pressure regulation amount and cycle time interval to generate a digital control instruction sequence that conforms to the nozzle hardware communication format.

[0202] S8.2: Based on the digital control command sequence, the electric regulating valve and high-pressure pump group of the spray actuator are driven by the pulse width modulation algorithm to convert the abstract parameter values ​​into a physical spray flow field with a specific opening ratio and output pressure.

[0203] S8.3: The physical spray flow field is applied to the surface of the molding material during the wet mixing process to perform real-time mixing intervention, and the local water content is changed by the droplet diffusion and material absorption mechanism to generate a mixed material flow containing the latest humidity information.

[0204] S8.4: Based on the online moisture acquisition module, the dielectric constant of the mixed material flow is scanned non-contactly, and the real-time moisture content data is extracted using the principle of high-frequency electromagnetic wave reflection to generate a new moisture ratio feedback signal characterizing the current working condition.

[0205] S8.5: The new moisture ratio feedback signal is re-injected into the sliding intercept processing unit as an input data stream to trigger the next round of local moisture ratio data segment update operation, so as to complete the closed-loop control process iterative update from parameter execution to state awareness.

[0206] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0207] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0208] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material, characterized in that, Specifically, it includes: S1: Obtain the original moisture ratio signal sequence within a continuous time window during the wet mixing process, and perform sliding truncation processing on the original moisture ratio signal sequence to generate local moisture ratio data segments. S2: Calculate the first-order difference symbol sequence, the second-order difference polarity inflection point, and the local curvature change rate based on the local moisture ratio data segment to generate the basic phase state code; S3: Map the basic phase state encoding to a two-dimensional coordinate system composed of the rate of change dominant horizontal axis and the acceleration dominant vertical axis to construct a trend phase space coordinate point set; S4: Based on the frequency statistics of the transitions between the trend phase spatial coordinate points under historical operating conditions, construct a directed graph structure and generate a trend phase transition map; S5: Define a dedicated spray control parameter response template library based on typical evolution chains in the trend phase transition map that have accumulated more than a preset threshold number of occurrences; S6: Real-time monitoring of the residence time of the current basic phase state code in the trend phase space, and combined with the historical average residence time comparison results and the multi-scale fluctuation energy ratio to generate triplet control commands; S7: Based on the triplet control command, retrieve matching items from the dedicated spray control parameter response template library, and combine the physical constraint boundary to perform parameter safety mapping to generate a spray control parameter set.

2. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 1, characterized in that, Step S7 is followed by: S8: The spray actuator is driven to move using the dynamically tuned spray control parameter set, and the new moisture ratio feedback signal after the movement is re-inputted to the sliding interception link to complete the iterative update of the closed-loop control process.

3. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 1, characterized in that, Step S4 specifically includes: The set of trend phase spatial coordinate points in the continuous time series is processed by extracting adjacent pairs in chronological order to generate the original phase transition event sequence. Based on the original phase transfer event sequence, perform similar path aggregation statistical processing, calculate the cumulative number of occurrences of each unique phase transfer path under historical operating conditions, and generate a phase transfer frequency statistics table. Based on the cumulative occurrence count in the phase transition frequency statistics table, a weight normalization calculation is performed. The cumulative occurrence count is converted into a transition probability weight value that represents the possibility of path occurrence using a frequency probability mapping mechanism, thereby generating a phase transition weighted dataset. The phase transition weighted dataset is used to construct a topological connection relationship, the phase encoding is mapped to nodes and the transition probability weight value is mapped to the directed edge weight, and an initial trend phase transition map is generated. The initial trend phase transition map is subjected to high-frequency path filtering and structural optimization processing. Low-probability transition edges are filtered out and high-confidence connection paths are retained based on a preset threshold to generate the trend phase transition map.

4. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 3, characterized in that, The original phase transition event sequence includes: a start phase code and a stop phase code.

5. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 4, characterized in that, The dedicated spray control parameter response template library includes: parameter adjustment directional strategies and step-by-step incremental rules.

6. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 5, characterized in that, Step S5 specifically includes: The weight attributes of each directed edge in the trend phase transition map are subjected to threshold filtering to extract typical evolution chain sequences whose cumulative occurrence exceeds a preset threshold, and a set of working condition evolution paths is established. Based on the initial phase encoding and the final phase encoding in the typical evolutionary chain sequence, phase transition feature mapping analysis is performed to generate a phase transition feature vector to characterize the coupling relationship between the rate of change of water ratio and the acceleration. The phase transition feature vector is used to query a preset strategy rule base, and a parameter adjustment directionality strategy matching operation is performed to generate a spray flow rate adjustment direction identifier and a spray pressure adjustment direction identifier. Based on the distribution data of the average residence time of the typical evolutionary chain sequence under historical working conditions, a step-by-step incremental rule calculation process is performed to generate a pressure benchmark value increment coefficient sequence. By integrating the spray flow rate adjustment direction indicator, spray pressure adjustment direction indicator, and pressure reference value increment coefficient sequence, a dedicated spray control parameter response template encapsulation operation is performed to generate the dedicated spray control parameter response template library.

7. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 1, characterized in that, Step S6 specifically includes: Obtain the basic phase state code at the current moment and its starting timestamp for entering the trend phase space, and calculate the current phase dwell time based on the system's current running clock; Read the historical average dwell time data corresponding to the basic phase state code at the current moment in the trend phase transition map, compare the current phase dwell time with the historical average dwell time, and generate a dwell time deviation coefficient. Extract the original moisture ratio signal sequence of a short-time window containing the basic phase state code at the current moment, extract the signal fluctuation components at different scales, and calculate the multi-scale fluctuation energy ratio based on the energy spectral density of each fluctuation component. The initial residence level label is determined based on the residence duration deviation coefficient. If the multi-scale fluctuation energy ratio exceeds the preset mutation threshold, the evolution chain confidence score is reduced; otherwise, the high confidence score is maintained, and an evolution chain confidence index is generated. The basic phase state code, dynamic residence stability label, and evolution chain confidence index at the current moment are integrated and assembled according to a predefined triplet data structure encapsulation protocol to output the triplet control command, wherein the dynamic residence stability label is obtained by modifying the initial residence level label.

8. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 7, characterized in that, The correction of the initial residency level label includes: If the multi-scale fluctuation energy ratio exceeds the preset mutation threshold, the level of the initial residence level label will be corrected downward. If the multi-scale fluctuation energy ratio does not exceed the preset mutation threshold, the initial residency level label remains unchanged.

9. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 1, characterized in that, The triplet control command includes phase encoding, residency level, and evolution chain confidence.

10. The spray humidification control method based on online monitoring of moisture ratio during the wet mixing process of grinding wheel forming material according to claim 1, characterized in that, The physical constraint boundaries include the upper limit of nozzle response delay and the minimum adjustable pressure step.

Images