Plastic granulator running state monitoring method and system based on multi-sensor fusion

Abstract

This invention discloses a method and system for monitoring the operating status of a plastic pelletizer based on multi-sensor fusion. The method includes: real-time acquisition of multi-dimensional operating data through a multi-sensor fusion module; outputting comprehensive monitoring data using an adaptive weighted fusion strategy; comparing and analyzing the comprehensive monitoring data with dynamic thresholds to identify abnormal trends; triggering a graded early warning mechanism to calculate a comprehensive severity value and determine the early warning level based on the degree of abnormal deviation, rate of change, and duration; and executing proactive adjustment control of process parameters, generating adjustment commands, verifying feasibility, and then issuing them. The system includes a multi-sensor fusion monitoring module, a knowledge base and fault feature database, a dynamic threshold comparison and analysis engine, an abnormal trend early warning mechanism, and a proactive adjustment control system for process parameters. This invention achieves comprehensive perception of the pelletizer's operating status, intelligent identification of anomalies, and automatic adjustment of process parameters, effectively reducing false alarm rates and ensuring production safety and stability.

Description

Technical Field

[0001] This application belongs to the field of industrial automation control, specifically relating to a method and system for monitoring the operating status of a plastic granulator based on multi-sensor fusion. Background Technology

[0002] Plastic granulators are key equipment in the plastics processing industry chain, and their operational stability directly affects product quality and production continuity. During granulator operation, operating parameters such as temperature, pressure, vibration, and current fluctuate significantly. Accurately identifying abnormal states and triggering reliable early warnings is a core prerequisite for ensuring production stability. Traditional methods often rely on manual inspections or offline monitoring, which makes it difficult to achieve real-time and comprehensive perception of the granulator's operating status.

[0003] Current solutions collect key operating parameters of the granulator by deploying single or limited combinations of sensors such as temperature, pressure, and vibration sensors, and then judge anomalies based on preset fixed thresholds. These solutions can play a certain monitoring role under relatively stable granulator operating conditions. However, in actual production, plastic granulators face multiple disturbances such as raw material batch differences, product specification changes, and ambient temperature variations, resulting in drastic fluctuations in operating conditions. Fixed thresholds are difficult to adapt to the dynamic range changes in normal operation, leading to a high false alarm rate and an inability to provide reliable anomaly warnings. Furthermore, existing solutions mostly focus on data acquisition and threshold alarms, lacking the ability to intelligently identify abnormal trends and proactively adjust process parameters.

[0004] In summary, existing technologies lack a monitoring scheme for the operating status of plastic granulators that can adapt to dynamic changes in their operating conditions, effectively reduce false alarm rates, and achieve automatic adjustment in abnormal states. This application aims to provide a method and system for monitoring the operating status of plastic granulators based on multi-sensor fusion, in order to solve the aforementioned technical problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for monitoring the operating status of a plastic granulator based on multi-sensor fusion, which can effectively solve the problems mentioned in the background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for monitoring the operating status of a plastic granulator based on multi-sensor fusion, the method comprising the following steps: The multi-sensor fusion monitoring module collects multi-dimensional operating data of the plastic granulator in real time, and adopts an adaptive weighted fusion strategy to preprocess and fuse the raw data collected by each sensor to output comprehensive monitoring data. The adaptive weighted fusion strategy monitors the consistency of the output data of each sensor in real time. When the data of a certain sensor deviates from the data of other sensors and the deviation exceeds a preset deviation threshold, the weight coefficient of the certain sensor is automatically reduced and the weight coefficients of other sensors are increased accordingly. The comprehensive monitoring data is compared and analyzed with a dynamic threshold to identify abnormal trends. The dynamic threshold is dynamically adjusted by a dynamic threshold algorithm based on the current operating status of the plastic granulator. The dynamic threshold algorithm combines a preset fixed threshold and a dynamic adjustment amount calculated based on historical operating data, and configures differentiated adjustment factors for different parameter types and fault types. After an abnormal trend is detected, a graded early warning mechanism is automatically triggered. The graded early warning mechanism calculates a comprehensive severity value by comprehensively considering the degree of abnormal deviation, the rate of change, and the duration of the current abnormal state, and determines the early warning level based on the comprehensive severity value. After the graded early warning mechanism is triggered, the process parameters are actively adjusted and controlled. Based on the identified fault type and the early warning level, the process parameters of the granulator are adjusted in a coordinated manner to eliminate the abnormal state. The process parameter active adjustment control performs a feasibility check after generating the parameter adjustment command and before issuing it. The feasibility check includes checking whether the adjusted parameter values ​​exceed the safe operating range of the equipment, and when the adjusted parameter combination falls into the parameter combination range corresponding to historical fault cases, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault cases.

[0007] On the other hand, the present invention provides a plastic pelletizer operation status monitoring system based on multi-sensor fusion, comprising: The multi-sensor fusion monitoring module is used to collect multi-dimensional operating data of the plastic granulator in real time, and to preprocess and fuse the raw data collected by each sensor using an adaptive weighted fusion strategy to output comprehensive monitoring data. The adaptive weighted fusion strategy is configured to monitor the consistency of the output data of each sensor in real time. When the data of a certain sensor deviates from the data of other sensors and the deviation exceeds a preset deviation threshold, the weight coefficient of the certain sensor is automatically reduced and the weight coefficients of other sensors are increased accordingly. The plastic granulator operation status knowledge base and fault feature base are used to store the normal operation parameter range, historical fault case feature data and expert experience rules, providing a data foundation for dynamic threshold calculation, abnormal trend identification and fault type identification. A dynamic threshold comparison and analysis engine is used to implement a dynamic threshold algorithm, which compares and analyzes the comprehensive monitoring data with dynamic thresholds to identify abnormal trends. The dynamic threshold algorithm is configured to combine a preset fixed threshold with a dynamic adjustment amount calculated based on historical operating data, and to configure differentiated adjustment factors for different parameter types and fault types. An abnormal trend early warning mechanism is used to automatically trigger a graded early warning after an abnormal trend is identified. It calculates a comprehensive severity value by combining the degree of abnormal deviation, the rate of change, and the duration, and determines the warning level based on the comprehensive severity value. The process parameter active adjustment control system is used to adjust the process parameters of the granulator in conjunction with the identified fault type and the warning level. The process parameter active adjustment control system is configured to perform a feasibility check after generating the parameter adjustment command and before issuing it. The feasibility check includes checking whether the adjusted parameter values ​​exceed the safe operating range of the equipment, and when the adjusted parameter combination falls into the parameter combination range corresponding to historical fault cases, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault cases.

[0008] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention significantly improves the accuracy and robustness of status monitoring through a multi-sensor fusion adaptive weighting strategy and a dynamic threshold algorithm. The adaptive weighting fusion monitors the consistency of data from each sensor in real time, automatically reducing the weight of abnormal sensors to avoid the impact of single-point failures or interference on the monitoring results. Simultaneously, the dynamic threshold adaptively adjusts according to changes in operating conditions, overcoming the high false alarm rate of fixed thresholds under fluctuating operating conditions, effectively reducing false alarms and missed alarms, and achieving accurate perception of the granulator's operating status.

[0009] 2. This invention constructs a complete closed-loop control link from anomaly identification and graded early warning to active adjustment of process parameters. It can not only detect anomalies in a timely manner, but also automatically adjust process parameters such as screw speed, heating temperature, and cooling intensity according to the type and severity of the fault. Furthermore, it performs feasibility verification before issuing adjustment commands to ensure that the adjustment operation does not exceed the equipment's safety range and does not cause other faults. This achieves automatic elimination of abnormal states and dynamic optimization of process parameters, significantly improving the automation level and safety of the production process.

[0010] 3. This invention establishes specific prevention and control strategies for typical faults such as pellet breakage, adhesion, and equipment overheating. It utilizes a pellet breakage precursor identification model, a three-dimensional temperature field distribution model, and a thermal balance dynamic model to achieve early identification and preventive adjustment of various fault precursors. At the same time, it combines a hierarchical structured knowledge base and fault feature database to support a complete data link from rule reasoning to feature matching and then to original data tracing, providing rich data support for fault diagnosis and decision-making, and comprehensively improving the intelligent level of operation and maintenance of plastic pelletizers. Attached Figure Description

[0011] Figure 1 A schematic diagram of the overall technical solution for a method of monitoring the operating status of a plastic granulator based on multi-sensor fusion; Figure 2 This is a schematic diagram illustrating the core principle of a multi-sensor fusion monitoring module. Figure 3 A schematic diagram illustrating the establishment and storage structure of the knowledge base and fault feature database for the operation status of a plastic granulator; Figure 4 Flowchart of a comparative analysis engine for intelligent identification of abnormal trends using dynamic threshold algorithms; Figure 5 A flowchart for automatically triggering an early warning mechanism for abnormal trends and handling fault level classification; Figure 6 The linkage control logic diagram for the active adjustment control system of process parameters; Figure 7 A diagram illustrating the multi-level interaction relationships and data flow between the monitoring terminal and the cloud-based knowledge base; Figure 8 A schematic diagram of the multi-level interaction relationship and data flow for the multi-channel distribution of early warning information and the linkage adjustment of process parameters. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 8 Specific embodiments are provided to further illustrate the present invention in detail.

[0013] Example 1 This invention provides a method and system for monitoring the operating status of a plastic granulator based on multi-sensor fusion. Through a multi-sensor fusion monitoring module, a knowledge base and fault feature database for the operating status of the plastic granulator, a dynamic threshold comparison and analysis engine, an abnormal trend early warning mechanism, an active adjustment and control system for process parameters, and preventive control strategies for typical faults such as pellet breakage, adhesion, and equipment overheating, it achieves comprehensive perception of the operating status of the plastic granulator, intelligent identification of abnormal trends, accurate identification of fault types, and active adjustment of process parameters.

[0014] Firstly, in the aforementioned method for monitoring the operating status of a plastic granulator based on multi-sensor fusion, the first step, S1, involves the multi-sensor fusion monitoring module acquiring multi-dimensional operating data of the plastic granulator in real time. It then preprocesses and fuses the raw data collected by each sensor using an adaptive weighted fusion strategy, outputting comprehensive monitoring data. The multi-sensor fusion monitoring module integrates five core sensor types: temperature sensor, pressure sensor, speed sensor, vibration sensor, and current sensor. Each sensor is deployed at a corresponding location on the granulator to collect operating parameters from different dimensions.

[0015] Temperature sensors are used to collect temperature parameters during the operation of the granulator. Their measurement range covers the entire operating temperature range of the granulator, and their response time meets real-time monitoring requirements. Pressure sensors are used to collect melt pressure parameters, and their range can withstand the peak melt pressure in the granulator. Speed ​​sensors employ a non-contact measurement method to meet the measurement needs of the high-speed rotating screw. Vibration sensors cover the main vibration frequency bands of the granulator and are used to collect equipment vibration parameters. Current sensors use the Hall effect principle to measure the current of the drive motor.

[0016] The multi-sensor fusion monitoring module preprocesses and fuses the raw data collected by each sensor using a data fusion algorithm. The algorithm employs an adaptive weighted fusion strategy. This strategy dynamically calculates weight coefficients based on the historical error statistics of each sensor. These weight coefficients are updated according to real-time data quality and used to adjust the contribution ratio of each sensor in the fusion result. The data fusion algorithm executes according to the following steps: Step S101 involves preprocessing the raw data collected by each sensor. The preprocessing operation includes three sub-steps. First, digital filtering is performed on the raw data, using a combination of median filtering and Kalman filtering to eliminate sensor noise and random interference.

[0017] Next, outlier detection and removal are performed on the filtered data. Data points exceeding the preset reasonable range are marked as outliers and removed. The preset reasonable range is pre-set based on the measurement range of each sensor and the parameter fluctuation range under normal operating conditions of the granulator.

[0018] Finally, the filtered and anomaly-removed data are standardized to unify the dimensions, sampling frequency, and data representation format of the data from each sensor.

[0019] Step S102: Dynamically assign weight coefficients to each sensor based on its measurement accuracy and reliability. The weight coefficients are calculated based on the historical error statistics of each sensor; sensors with higher measurement accuracy receive higher weight coefficients, and sensors with lower measurement accuracy receive lower weight coefficients. The weight coefficients are dynamically updated based on real-time data quality: when the data quality of a sensor deteriorates, its weight coefficient decreases accordingly.

[0020] This embodiment monitors the consistency of the output data from each sensor in real time. When the data from one sensor deviates from the data from other sensors and the deviation exceeds a preset deviation threshold, this embodiment automatically reduces the weighting coefficient of that sensor and correspondingly increases the weighting coefficients of the other sensors.

[0021] For example, when the amplitude of the vibration signal collected by the vibration sensor suddenly increases significantly while the data from other sensors does not change accordingly, this embodiment determines that the vibration data may be abnormal, and then reduces the weight coefficient of the vibration sensor to avoid abnormal data from having an excessive impact on the fusion result.

[0022] Step S103: The data from each sensor are fused using a weighted average method. The fusion calculation is based on the weighting coefficients assigned in step S102, and the preprocessed data from each sensor are weighted and summed to output comprehensive monitoring data, which includes temperature characteristics, pressure characteristics, vibration characteristics, and current characteristics.

[0023] The specific formula for weighted average fusion calculation is as follows:

[0024] Among them, each weight coefficient satisfies the normalization condition:

[0025] In the above formula, The total number of sensors participating in the fusion computing. The value is 5; For sensor serial number, Corresponding temperature sensor, Corresponding pressure sensor, Corresponding speed sensor, Corresponding vibration sensor, Corresponding current sensor; Sampling time; For the first Each sensor at time Preprocessed data values; For the first Each sensor at time Weighting coefficients; For a moment The integrated monitoring data value after fusion. It is a multidimensional vector containing temperature, pressure, vibration, and current components, each expressed in standardized units.

[0026] In this embodiment, the weighting coefficient The initial values ​​are preset based on the nominal accuracy of each sensor. The initial weights for temperature and pressure sensors are set to 0.22, for speed sensors to 0.20, and for vibration and current sensors to 0.18. The dynamic adjustment range of each weight coefficient does not exceed ±30% of the initial value.

[0027] Step S1 completes the entire processing flow from raw data acquisition to comprehensive monitoring data output. Step S101 removes noise and outliers from the raw data and standardizes the data format. Step S102 assigns fusion weights based on the data quality of each sensor. Step S103 synthesizes the multi-source data into comprehensive monitoring data with a unified representation. After the comprehensive monitoring data is output, it serves as input for subsequent dynamic threshold comparison analysis and anomaly trend identification steps.

[0028] For step S2 in this embodiment, a knowledge base and a fault feature base for the plastic granulator's operating status are established and stored. The knowledge base and fault feature base comprise three core components: normal operating parameter ranges, historical fault case feature data, and expert experience rules. Each component is established according to the following steps: Step S201: Establish the normal operating parameter range. The normal operating parameter range is derived from statistical analysis of a large amount of historical operating data, covering the safe fluctuation ranges for temperature, pressure, speed, vibration, and current parameters. The normal operating parameter range is configured differently according to different product specifications and manufacturing processes, with each product specification corresponding to an independent parameter range template.

[0029] Step S202: Store historical fault case feature data. Historical fault cases are stored in a structured format, including complete sensor data sequences at the time of each fault occurrence, fault type labels, fault development process descriptions, and corresponding handling measures. Historical fault cases are stored according to fault type, which includes at least particle breakage faults, adhesion faults, equipment overheating faults, bearing wear faults, and screw wear faults.

[0030] Multiple typical cases are stored under each fault type. Each typical case includes a complete sensor data sequence within the time window preceding the fault occurrence, fault type labeling, fault development process description, and corresponding handling measures. The time index uses the timestamp of the fault occurrence as the primary key, and also records the fault duration and fault recovery time. The time index is established to support weighted analysis of recent operating conditions.

[0031] Step S203: Encode and store expert experience rules. Expert experience rules encode fault diagnosis rules and parameter adjustment suggestions provided by domain experts in a computable form, represented in the form of production rules. Each production rule includes a condition part and an action part. The condition part defines the monitoring data conditions that must be met to trigger the production rule, and the action part defines the diagnostic conclusion or parameter adjustment operation that should be performed after triggering.

[0032] The inference and execution of production rules are calculated based on the matching degree, which measures the degree of conformity between the current comprehensive monitoring data and the conditional part of the production rule.

[0033] The matching degree is calculated as follows: The condition part of each production rule contains several condition items, each of which specifies a certain monitoring parameter and its corresponding numerical range or state; the current comprehensive monitoring data is compared with each condition item in the condition part of the production rule one by one. If the value of a parameter in the current comprehensive monitoring data falls within the numerical range specified by the corresponding condition item, the condition item is successfully matched; otherwise, the matching fails; the matching degree is defined as the proportion of the number of successfully matched condition items to the total number of condition items in the condition part of the production rule.

[0034] When the matching degree exceeds the preset matching threshold, the production rule is triggered to execute, and the execution content is the diagnostic conclusion or parameter adjustment operation specified in the action part of the production rule.

[0035] Step S204: Establish a hierarchical structured storage architecture. The knowledge base and fault feature base adopt a hierarchical structured storage design, comprising three layers: the raw data layer, the feature extraction layer, and the rule reasoning layer.

[0036] The raw data layer stores the raw data acquired by each sensor and timestamp information. The raw data layer is indexed using a composite primary key of sensor number and timestamp.

[0037] The feature extraction layer extracts features and vectorizes the raw data. It extracts time-domain and frequency-domain features from the preprocessed sensor data. Time-domain features include mean temperature, temperature fluctuation range, peak pressure, pressure pulsation amplitude, rotational speed stability index, RMS vibration value, RMS current value, and power factor. Frequency-domain features include the dominant vibration frequency components and their amplitudes.

[0038] The aforementioned time-domain and frequency-domain features are concatenated according to a preset dimensional order to form a multi-dimensional feature vector, which is the matching fault mode representation vector. The feature extraction layer establishes an index using the feature vector number and the start timestamp of the corresponding time window, and establishes a pointer to the sensor data in the original data layer.

[0039] In the feature extraction layer, the correspondence between the matching fault mode representation vector and the fault type is calibrated using historical fault cases. For each historical fault case, its corresponding sensor data sequence is converted into a multi-dimensional feature vector using the aforementioned feature extraction method, and the multi-dimensional feature vector is associated and stored with the fault type label of that historical fault case. When identifying the fault type, the current comprehensive monitoring data is converted into the current multi-dimensional feature vector, and the similarity between the current multi-dimensional feature vector and the multi-dimensional feature vectors of typical cases stored under each fault type is calculated. The fault type of the typical case with the highest similarity is determined as the current fault type.

[0040] Similarity calculation employs a weighted combination of Euclidean distance and cosine similarity. Euclidean distance measures the absolute distance between two multidimensional feature vectors in the multidimensional vector space, while cosine similarity measures the directional consistency between the two feature vectors. The weighted similarity value serves as the final similarity metric. When the final similarity exceeds a preset similarity threshold, the current multidimensional feature vector is considered to have successfully matched the typical case.

[0041] The rule reasoning layer stores fault diagnosis rules and reasoning logic based on expert experience. The rule reasoning layer is indexed by rule number and applicable fault type label, and establishes association pointers with feature vector patterns in the feature extraction layer.

[0042] Data is linked between layers through an indexing mechanism, supporting rapid retrieval and multi-dimensional queries of historical data. The data flow from the raw data layer to the feature extraction layer and then to the rule inference layer is clear: the raw data layer provides raw sensor data, the feature extraction layer performs feature transformation on the raw data to generate multi-dimensional feature vectors, and the rule inference layer performs fault diagnosis inference based on the multi-dimensional feature vectors. Through the aforementioned indexing mechanism, a data link is realized from rule triggering to feature matching and then to raw data tracing.

[0043] Step S2 establishes the complete architecture of the knowledge base and fault feature base. This knowledge base provides parameter boundaries for subsequent dynamic threshold calculation, provides historical case matching basis and expert rule reasoning basis for fault type identification, and provides hierarchical index support for multi-dimensional data query.

[0044] For step S3, the comprehensive monitoring data output from step S1 is compared and analyzed with the preset fixed threshold and historical operating condition data. This step uses a dynamic threshold algorithm to achieve intelligent identification of abnormal trends. The dynamic threshold algorithm dynamically adjusts the warning threshold according to the current operating status of the plastic granulator. The core of the dynamic threshold algorithm lies in establishing a dynamic threshold calculation model, which comprehensively considers the preset fixed threshold and the dynamic adjustment amount calculated based on historical operating condition data.

[0045] The calculation of the dynamic adjustment amount comprehensively considers the statistical characteristics of the operational data within the most recent time period and the degree of deviation between the current comprehensive monitoring data and historical data. An adjustment factor is used to achieve adaptive adjustment of the dynamic threshold within a preset range. The dynamic threshold algorithm is executed according to the following steps: Step S301: Calculate the dynamic adjustment amount. The dynamic adjustment amount is calculated based on the statistical characteristics of the operating data within the most recent time period and the degree of deviation between the current comprehensive monitoring data and historical data. The statistical characteristics of the operating data within the most recent time period include the average values ​​of temperature, pressure, rotational speed, vibration, and current parameters, as well as the standard deviation of each parameter. The degree of deviation between the current comprehensive monitoring data and historical data is measured by the ratio of the absolute value of the difference between the current comprehensive monitoring data and historical data for the same period to the standard deviation of historical data for the same period.

[0046] Step S302: Calculate the dynamic threshold by combining the preset fixed threshold and the dynamic adjustment amount. The formula for calculating the dynamic threshold is as follows:

[0047] in, For a moment The dynamic threshold; The preset fixed threshold is set in advance based on the parameter statistical distribution under normal operating conditions of the granulator; The adjustment factor is used to control the contribution of the dynamic adjustment amount to the dynamic threshold. For a moment The dynamic adjustment amount. Preset fixed threshold. Adjustment factor and dynamic adjustment amount The dimensions of all three are consistent with the dimensions of the monitored parameter, ensuring the uniformity of dimensions in the weighted summation operation.

[0048] In this embodiment, the adjustment factor The value is configured differently based on the parameter type and fault type. For temperature parameters, The value range is from 0.1 to 0.3; for pressure parameters, The value range is from 0.2 to 0.4; for vibration parameters, The value range is 0.4 to 0.6; for speed and current parameters, The value range is 0.2 to 0.5. This range is preset based on the fluctuation range and anomaly sensitivity of each parameter under normal operating conditions. Specific values ​​can be optimized using the statistical characteristics of historical data. For example, the standard deviation of parameter fluctuations under normal operating conditions in historical data can be used as a reference to select the value that minimizes the false alarm rate of the dynamic threshold. value.

[0049] Step S303 involves limiting the dynamic threshold calculation result. This limiting process applies a preset upper and lower limit boundary to the dynamic threshold calculated in step S302, preventing it from deviating excessively from a reasonable range. The upper and lower limits are determined based on the safe operating range of each parameter. When the dynamic threshold calculation result exceeds the upper limit boundary, the upper limit boundary is used as the final dynamic threshold; when the dynamic threshold calculation result is below the lower limit boundary, the lower limit boundary is used as the final dynamic threshold.

[0050] Step S304: Configure differentiated adjustment factors according to different fault types. The adjustment factor for the temperature parameter is configured to a smaller value to minimize fluctuations in the dynamic temperature threshold and prevent frequent warnings triggered by small temperature fluctuations. The adjustment factor for the vibration parameter is configured to a larger value to ensure that the dynamic vibration threshold maintains high sensitivity to rapid changes in the vibration signal. Refer to the configuration in step S302 for the value range of the adjustment factor corresponding to each fault type.

[0051] Step S305: Compare the comprehensive monitoring data output in step S1 with the dynamic threshold calculated in step S302 and obtained after amplitude limiting processing in step S303, and perform weighted analysis in conjunction with historical operating data. The weighted comparison of historical operating data adopts an exponential decay weighting method, assigning higher weight coefficients to recent historical data and lower weight coefficients to older historical data.

[0052] The weight decay coefficient is configured according to the operating condition change cycle of the granulator. A larger weight decay coefficient is configured for the production stage with faster operating condition changes to highlight the impact of recent data, while a smaller weight decay coefficient is configured for the production stage with slower operating condition changes to make use of more historical information.

[0053] In this embodiment, the weight decay coefficient The value is determined based on the operating condition change cycle. When the operating condition change cycle is less than 1 hour, The value is 0.1; when the operating condition change cycle is 1 to 4 hours, The value is 0.05; when the operating condition change cycle is greater than 4 hours, The value is set to 0.01. The above value can be adjusted based on actual production data, with the principle being that the dynamic threshold can keep up with changes in operating conditions in a timely manner without causing excessive fluctuations.

[0054] The specific calculation method for weighted alignment is as follows: For time... Comprehensive monitoring data value Calculate the deviation between it and the corresponding time in the historical operating condition data sequence. The formula for calculating the deviation is:

[0055] in, For a moment Comprehensive monitoring data and historical moments The deviation between historical operating condition data; This is the time lag, representing the time interval between the current moment and a historical moment. For a moment Comprehensive monitoring data values; For a moment Historical operating condition data values.

[0056] The formula for calculating the weighted bias is:

[0057] in, For a moment The weighted deviation; The length of the historical data time window used for weighted calculation; This is the time lag, with a value ranging from 1 to... ; Time lag The corresponding weighting coefficients are calculated using an exponential decay method. , These are the initial weighting coefficients. This is the weight decay coefficient. This is the base of the natural logarithm. The weighting coefficients for each time lag are... Satisfy normalization conditions According to the normalization condition, the initial weighting coefficients... From the formula The calculation determines the time window length. In this embodiment, the time window length is... The value ranges from 60 to 120 sampling periods, with the specific value determined based on the device's response speed and computing resources, in order to cover at least one complete operating condition change cycle.

[0058] Step S306: Based on the weighted deviation calculated in step S305, determine whether there is an abnormal trend in the current operating state. In this embodiment, when the comprehensive monitoring data continuously deviates from the dynamic threshold and the weighted deviation exceeds a preset anomaly determination threshold, an abnormal trend is identified. The determination condition for continuous deviation is: continuous... The weighted deviation at each sampling time exceeded the anomaly detection threshold, among which This is the threshold for the number of times the error continues.

[0059] In this embodiment, the anomaly detection threshold is set to 1.2 times the dynamic threshold width, and the dynamic threshold width is defined as half the difference between the upper and lower limits of the dynamic threshold; the continuous deviation count threshold... The value is taken from 3 to 5 times, and the specific value is determined according to the sampling frequency. A larger value is taken when the sampling frequency is higher, and a smaller value is taken when the sampling frequency is lower, in order to eliminate occasional fluctuation interference and take into account the timeliness of response.

[0060] Step S305 incorporates historical operating data into the comparative analysis using an exponential decay weighted method, ensuring that anomaly identification considers the relative changes between current comprehensive monitoring data and historical data. Step S306 eliminates false alarms caused by occasional data fluctuations through continuous deviation judgment criteria. These two steps work together to utilize the reference value of historical operating data while reducing the false alarm rate through continuous deviation conditions. Step S3, through dynamic threshold calculation and weighted comparison analysis, identifies abnormal trends in the granulator's operating status and outputs the abnormal trend judgment results for subsequent step S4 to trigger the graded early warning mechanism.

[0061] Regarding step S4 in the method of this embodiment, a graded early warning mechanism is automatically triggered after an abnormal trend is identified in step S3. This step performs graded processing based on the severity of the fault, calculating a comprehensive severity value by combining three dimensions: the degree of abnormal deviation, the rate of change, and the duration. Based on the comprehensive severity value, the early warning level is divided into four levels: normal, attention, warning, and severe. The graded early warning mechanism is executed according to the following steps: Step S401: Calculate the comprehensive severity value. The comprehensive severity value is calculated by weighted summation of features from three dimensions: degree of abnormal deviation, rate of change, and duration. The degree of abnormal deviation is defined as the ratio of the magnitude by which the current comprehensive monitoring data exceeds the dynamic threshold to the width of the dynamic threshold.

[0062] The rate of change is defined as the amount of change of the monitored parameter per unit time, obtained by differential calculation of the comprehensive monitoring data. The duration is defined as the cumulative number of sampling points where the comprehensive monitoring data continuously exceeds the dynamic threshold.

[0063] All three features are normalized before weighted summation. After normalization, the values ​​of the three features range from 0 to 1, and the units are unified as dimensionless values, ensuring the uniformity of units in the weighted summation operation. The weight coefficients are configured differently according to the parameter type and fault type.

[0064] When there is an abnormal temperature, the weighting coefficient corresponding to the degree of abnormal deviation is set higher to highlight the danger of excessively high temperatures; when there is an abnormal vibration, the weighting coefficient corresponding to the rate of change is set higher to respond quickly to sudden increases in vibration. The sum of the three weighting coefficients is 1.

[0065] Step S402: Determine the warning level based on the comprehensive severity value. The warning levels are divided into four levels: normal, attention, warning, and severe. When the comprehensive severity value is less than 0.3, the warning level is normal; when it is greater than or equal to 0.3 and less than 0.6, the warning level is attention; when it is greater than or equal to 0.6 and less than 0.8, the warning level is warning; and when it is greater than or equal to 0.8, the warning level is severe.

[0066] The above thresholds are calibrated based on the statistical distribution of the comprehensive severity values ​​in historical fault data, and can be adjusted according to the false alarm rate and false negative rate requirements in actual application scenarios.

[0067] Step S403: Determine the push strategy and distribute the warning information based on the warning level. The warning information is pushed to three channels—the on-site control panel, the mobile terminal, and the management backend—through a multi-channel distribution protocol. No warning information is pushed for the normal level; warning information is pushed only to the on-site control panel for the attention level; warning information is pushed to the on-site control panel and the management backend for the alert level; and severe information is pushed to the on-site control panel, the management backend, and the mobile terminal simultaneously. The warning information includes the fault type, fault location, overall severity value, warning level, and recommended actions.

[0068] Step S404: When the warning level is severe, the forced shutdown protection is triggered. The forced shutdown protection command is sent to the emergency stop circuit of the granulator to cut off the heating power supply and the drive motor power supply, while keeping the cooling system running and locking the operation authority until the fault is cleared and the authorized personnel can manually release it.

[0069] Step S4 follows the abnormal trend determination result of step S3, outputs graded early warning information and executes corresponding push and protection actions, providing trigger basis and fault level information for the process parameter active adjustment and control system in the subsequent step S5.

[0070] In step S5, after the early warning mechanism in step S4 is triggered, active adjustment control of process parameters is executed. This step automatically adjusts process parameters such as screw speed, heating temperature, and cooling intensity through the active adjustment control system to eliminate abnormal conditions and restore the system to its normal operating range. The active adjustment control system invokes appropriate preventative control strategies based on the fault type and severity, generates parameter adjustment commands, and sends them to the actuators. The active adjustment control system executes the following steps: Step S501: Determine the parameter combination to be adjusted and its adjustment sequence according to the fault type. The parameter adjustment linkage logic determines the parameter combination to be adjusted based on the fault type identified in step S4. For a pellet breakage fault, adjust the screw speed and feed rate, with the screw speed adjustment having a higher priority than the feed rate because the screw speed has a more direct impact on the melt pressure.

[0071] For adhesion faults, adjust the cooling water flow rate and cooling time, with the cooling water flow rate adjustment having higher priority than the cooling time. For equipment overheating faults, adjust the heating power and cooling water flow rate, with the heating power adjustment having higher priority than the cooling water flow rate. The priority of each parameter adjustment is determined based on the sensitivity of each parameter to the fault response. Sensitivity is quantified by the correlation coefficient between parameter changes and the fault occurrence time in historical fault data. The parameter with the largest absolute value of the correlation coefficient has the highest adjustment priority.

[0072] Step S502: Calculate the adjustment amount of each parameter based on the deviation between the current abnormal state and the target state. The adjustment amount of each parameter is calculated using a proportional adjustment method. The deviation value is calculated as the difference between the current parameter value and the target parameter value. The target parameter value is the center value of the normal operating parameter range. The normal operating parameter range is taken from the normal operating parameter range established in step S201. The adjustment amount is equal to the product of the deviation value and the proportional coefficient. The formula for calculating the adjustment amount is as follows:

[0073] in, For the first The adjustment amount of each parameter, For the first The proportionality coefficient of each parameter For the first The deviation value of each parameter. Deviation values ​​of each parameter The calculation formula is , For the first The current values ​​of the parameters, For the first Target values ​​for each parameter. Scale coefficient. The value is set according to the sensitivity of the fault adjustment based on each parameter.

[0074] Adjustment sensitivity is defined as the change in the monitored parameter caused by a unit adjustment of the parameter. It is calculated by comparing parameter adjustment records in historical fault data with the corresponding response data of the monitored parameter. Parameters with high adjustment sensitivity use a smaller proportional coefficient to prevent over-adjustment, while parameters with low adjustment sensitivity use a larger proportional coefficient to ensure adjustment effectiveness. In this embodiment, the proportional coefficient for the temperature parameter... The value is 0.15, which is the proportional coefficient for the speed parameter. The value is 0.3, which is the proportionality coefficient of the cooling water flow rate. The value is 0.25, which is the proportionality coefficient of the feed rate. The value is 0.2. All the above proportionality coefficients are dimensionless coefficients. and The dimensions must be consistent to ensure that the dimensions of the formula calculations are uniform.

[0075] Step S503: Perform a feasibility check before parameter adjustment. The feasibility check is performed before the parameter adjustment command is issued. The feasibility check includes a first check item and a second check item.

[0076] The first verification item is to check whether the adjusted parameter values ​​exceed the safe operating range of the equipment. The safe operating range of the equipment is determined according to the limit values ​​of each parameter specified in the granulator equipment manual. The upper limit of the safe operating range for temperature parameters is the rated maximum operating temperature of the equipment, and the lower limit of the safe operating range for temperature parameters is the rated minimum operating temperature of the equipment; the upper limit of the safe operating range for pressure parameters is the rated maximum melt pressure of the equipment; the upper limit of the safe operating range for speed parameters is the rated maximum screw speed of the equipment; and the upper limit of the safe operating range for heating power is the rated maximum heating power of the equipment. When the adjusted parameter values ​​exceed the safe operating range, the adjustment amount will be limited to the difference between the safe boundary value and the current parameter value.

[0077] The second verification item is to verify whether the adjustment operation may cause other types of abnormalities. When the adjusted parameter combination falls into the parameter combination range corresponding to any fault type in the historical fault cases stored in step S202, it is determined that it may cause other types of abnormalities. At this time, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault case.

[0078] In step S504, the parameter adjustment command is converted into a control signal and sent to the corresponding actuator. The actuator drive control algorithm generates the control signal for the corresponding actuator based on the parameter combination determined in step S501, the adjustment amount calculated in step S502, and the adjustment command verified in step S503.

[0079] The frequency converter receives speed regulation signals to adjust the screw motor speed. The speed regulation signal of the frequency converter is a current signal of 4 to 20 mA. 4 mA corresponds to the minimum speed and 20 mA corresponds to the maximum speed. The relationship between the speed regulation signal value and the target speed is a linear one.

[0080] The contactor array receives power adjustment signals to adjust the heating power of the heating elements. The contactor array consists of multiple contactors, each of which controls a group of heating elements. The heating power is adjusted in stages by controlling the number of contactors that are turned on. The on / off state of each contactor is controlled by a digital output module.

[0081] The electric regulating valve receives a flow regulation signal to adjust the cooling water flow. The flow regulation signal of the electric regulating valve is a current signal of 4 to 20 mA. 4 mA corresponds to the valve being fully closed, and 20 mA corresponds to the valve being fully open. The relationship between the flow regulation signal value and the valve opening is linear.

[0082] In step S505, the process parameters are smoothly adjusted according to a preset adjustment rate. After receiving the control signal issued in step S504, the actuator gradually adjusts the process parameters to the target value according to the preset adjustment rate. The adjustment rate is configured differently according to the parameter type. The adjustment rate of the temperature parameter is configured such that the change per minute does not exceed 10% of the difference between the current temperature and the target temperature; the adjustment rate of the speed parameter is configured such that the change per minute does not exceed 15% of the difference between the current speed and the target speed; and the adjustment rate of the cooling water flow rate is configured such that the change per minute does not exceed 20% of the difference between the current flow rate and the target flow rate.

[0083] The above-mentioned adjustment rate configuration is used to avoid sudden changes in process parameters from impacting the operation of the granulator. During the adjustment process, the actuator samples the current parameter value at fixed time intervals. The adjustment amount at each step is equal to the ratio of the total adjustment amount to the preset number of adjustment steps, until the parameter value reaches the target value. The formula for calculating the adjustment rate is as follows:

[0084] in, For the first The adjustment amount of each parameter at each step, The first one calculated in step S502 The absolute value of the parameter adjustment amount For the first The preset number of adjustment steps for each parameter. The preset number of adjustment steps for the temperature parameter. The value is 10 steps, which is the preset adjustment step for the speed parameter. The value is 6 steps, which is the preset adjustment step for cooling water flow. The set value is 4 steps. The execution time interval for each step is fixed at 1 second, so the total adjustment time is equal to the preset number of adjustment steps. Second. Dimensions and the first All parameters have the same dimensions. The number of steps is dimensionless. Dimensions and the first The dimensions of each parameter are consistent to ensure that the dimensions of the formula calculations are uniform.

[0085] Steps S501 to S503 complete the generation and verification of parameter adjustment commands, while steps S504 to S505 complete the conversion and execution of control signals. The active process parameter adjustment control system achieves a complete process from command generation to actuator action through these five steps. Step S5 receives the fault type and severity information output from step S4 and outputs control signals to each actuator of the granulator, realizing automatic elimination of abnormal states and dynamic optimization adjustment of process parameters.

[0086] In step S6, the breakage prevention and control strategy is implemented. This step establishes a breakage precursor identification model based on the coupling relationship between melt pressure and screw speed, and monitors the changing trends of melt pressure and screw speed in real time. The breakage precursor identification model is a nonlinear mapping model based on a multi-layer neural network. The model input is the melt pressure sequence and screw speed sequence within a preset time window before the current moment, and the output is the breakage risk probability. The breakage precursor identification model is established and executed according to the following steps: Step S601: Construct a training dataset for the pellet breakage precursor recognition model. The training dataset is derived from the historical operating records of the plastic pelletizer. Extract melt pressure data, screw speed data, and pellet breakage fault annotation data from the historical operating records.

[0087] The melt pressure and screw speed sequences within one time window before and after each pellet breakage failure are selected as positive samples, while data sequences of the same length are selected from the normal operation period as negative samples. The positive and negative samples are combined in a 1:1 ratio to form the training dataset.

[0088] Step S602: Train the pre-granulation precursor recognition model. A multi-layer neural network is used as the basic architecture of the model, and supervised training is performed using the training dataset constructed in step S601. The training objective is to minimize the binary cross-entropy loss between the model's output granulation risk probability and the true label. During training, the backpropagation algorithm is used to update the network parameters until the loss function converges.

[0089] Step S603: Monitor the changing trends of melt pressure and screw speed in real time. After the pellet breakage precursor recognition model is trained, deploy it to the field controller for operation. The field controller receives melt pressure data from the pressure sensor and screw speed data from the speed sensor in real time. It extracts real-time data sequences according to the same sampling interval and sequence length as the training data, forms an input feature vector, and inputs it into the pellet breakage precursor recognition model.

[0090] Step S604: Calculate the probability of particle breakage risk and extract the precursor features of particle breakage. Input the input feature vector formed in step S603 into the precursor identification model of particle breakage, and the precursor identification model of particle breakage outputs the probability of particle breakage risk at the current moment. At the same time, the decreasing trend of melt pressure and the abnormal fluctuation amplitude of screw speed are calculated in real time.

[0091] Step S605: Determine if there are any signs of pellet breakage. If the probability of pellet breakage exceeds a preset risk probability threshold, or the instantaneous drop in melt pressure exceeds a preset pressure drop threshold, or the screw speed fluctuation exceeds a preset speed fluctuation threshold, then it is determined that there are signs of pellet breakage. Preventive adjustment is triggered immediately if any of the above three determination conditions are met.

[0092] Step S606: When a precursor to pellet breakage is detected, the screw speed reduction ratio is calculated based on the difference between the current pellet breakage risk probability and the preset risk probability threshold. The adjustment amount of the feed rate is proportional to the screw speed reduction ratio. A screw speed adjustment command and a feed rate adjustment command are generated and sent to the actuator to gradually reduce the screw speed and feed rate until the melt pressure returns to the normal operating parameter range and the pellet breakage risk probability drops below the preset risk probability threshold.

[0093] Step S6, building upon the proactive adjustment and control of process parameters in Step S5, provides specific preventative control for grain breakage faults. This step establishes a grain breakage precursor identification model based on historical data, combines it with real-time monitoring data to determine the risk of grain breakage, and automatically executes preventative parameter adjustments when the risk occurs, thereby achieving early intervention and effective prevention of grain breakage faults.

[0094] Step S7 in this embodiment involves the execution of an adhesion failure prevention and control strategy. This step establishes an adhesion trend prediction model based on the relationship between temperature field distribution and discharge rate. It performs real-time modeling and analysis of the temperature field distribution in the granulator head plate area and comprehensively judges the likelihood of adhesion occurrence by combining the discharge rate parameter. The adhesion trend prediction model is established and executed according to the following steps: Step S701: Establish a three-dimensional temperature field distribution model for the head plate area of ​​the granulator. The three-dimensional temperature field distribution model is reconstructed using data from a multi-point temperature sensor array. The multi-point temperature sensor array is deployed at different spatial locations within the head plate area of ​​the granulator, and each temperature sensor synchronously collects temperature data according to a preset sampling period.

[0095] The three-dimensional temperature field distribution model uses the spatial coordinates of each temperature sensor and the corresponding temperature measurement values ​​as input, and reconstructs the complete three-dimensional temperature field distribution of the pellet mill head plate area through interpolation calculation methods. During the modeling process, cooling water flow rate parameters, ambient temperature parameters, and discharge rate parameters are introduced as boundary conditions affecting the temperature field distribution.

[0096] Step S702: Real-time evaluation of particle cooling status based on three-dimensional temperature field distribution model. The three-dimensional temperature field distribution model calculates the temperature value of each spatial point in the head plate area based on real-time collected multi-point temperature sensor data, current cooling water flow rate, current ambient temperature and current discharge rate, and generates a three-dimensional temperature field that reflects the temperature distribution in the head plate area.

[0097] The particle cooling status is evaluated in real time by analyzing the deviation between the temperature value of each spatial point in the head plate area and the preset target temperature range.

[0098] Step S703: Calculate the adhesion risk index. The adhesion trend prediction algorithm calculates the adhesion risk index by combining the temperature field characteristics output by the three-dimensional temperature field distribution model and the current discharge rate parameters. The temperature field characteristics include the average temperature of each spatial point in the head plate area, the proportion of the area of ​​the head plate area whose temperature is higher than the preset adhesion risk temperature threshold, and the proportion of the area of ​​the head plate area whose temperature gradient exceeds the preset gradient threshold.

[0099] The formula for calculating the adhesion risk index is a weighted sum of various temperature field characteristics and discharge rate parameters. Each weight coefficient is set according to the correlation coefficient between various characteristics and the occurrence of adhesion failure in historical failure data. The higher the correlation coefficient, the higher the weight coefficient of the characteristic.

[0100] Step S704: Determine whether to trigger cooling intensity adjustment. When the adhesion risk index calculated in step S703 exceeds the preset adhesion risk index threshold, cooling intensity adjustment is automatically triggered.

[0101] Step S705: Calculate and execute the adjustment amount of cooling intensity. The adjustment amount of cooling intensity is dynamically calculated based on the current temperature deviation and the current discharge rate. The temperature deviation is the difference between the average temperature of each spatial point in the head plate area and the center value of the preset target temperature range. The cooling intensity adjustment amount is equal to the product of the temperature deviation value and the cooling intensity proportionality coefficient, multiplied by the discharge rate influence factor.

[0102] The discharge rate influencing factor is the ratio of the current discharge rate to the standard discharge rate. After the adjustment amount is calculated, cooling water flow rate adjustment instructions and cooling time adjustment instructions are generated and sent to the electric regulating valve and cooling system controller to increase the cooling water flow rate and extend the cooling time until the adhesion risk index drops below the preset adhesion risk index threshold.

[0103] Steps S701 to S702 establish a temperature field distribution model for the head plate area and evaluate the cooling status in real time. Step S703 calculates the adhesion risk index. Step S704 determines whether adjustment is triggered. Step S705 executes cooling intensity adjustment and continuously monitors until the risk is eliminated. Step S7 provides specific preventive control for adhesion failures, which works in conjunction with the pellet breakage failure prevention control in step S6 and the equipment overheating failure prevention control in step S8 to jointly constitute a preventive control system for typical failures of plastic pelletizers.

[0104] Finally, step S8 involves the execution of the equipment overheating fault prevention and control strategy. This step utilizes multi-point temperature data collected by a temperature sensor array and a dynamic heat balance model to achieve overheating early warning and dynamic adjustment of heating power. The dynamic heat balance model calculates the granulator's heat input, heat accumulation, and heat loss in real time, establishes the dynamic relationship between temperature changes and various influencing factors, and updates parameters based on real-time sensor data to maintain prediction accuracy. The equipment overheating fault prevention and control strategy is executed according to the following steps: Step S801: Establish a dynamic thermal balance model. This model, based on the heat balance equation of the granulator, describes the dynamic relationship between heat input, heat accumulation, and heat loss during the granulator's operation. Heat input originates from the heating power provided by the heating elements; heat accumulation manifests as the temperature rise of various components of the granulator; heat loss includes natural heat dissipation to the ambient air through the equipment surface and heat carried away by the cooling water.

[0105] The thermal equilibrium dynamic model is mathematically expressed as a first-order differential equation concerning the granulator temperature. This equation includes terms for heating power, cooling water flow rate, ambient temperature, and current temperature. Model parameters include the heat transfer coefficient and heat capacity coefficient. Initial parameter values ​​are set according to the nominal values ​​in the granulator equipment manual. Subsequent online updates of the model parameters are based on real-time sensor data to minimize the deviation between the predicted temperature output by the model and the actual measured temperature.

[0106] Step S802: Real-time acquisition of multi-point temperature data and updating of the thermal balance dynamic model. Multi-point temperature data is acquired in real-time using an array of temperature sensors deployed at different locations within the granulator. The real-time acquired multi-point temperature data, current heating power, current cooling water flow rate, and current ambient temperature are input into the thermal balance dynamic model, which outputs the predicted temperature for each monitoring location.

[0107] The predicted temperature output by the model is compared with the actual temperature measurement value collected by the temperature sensor, and the prediction error is calculated. When the prediction error exceeds the preset error threshold, the model parameters are updated, and the heat transfer coefficient and heat capacity coefficient of the dynamic thermal equilibrium model are adjusted using a recursive least squares algorithm to bring the prediction error within the preset error threshold.

[0108] Step S803: Monitor the temperature rise rate and absolute temperature value to determine if there is an overheating trend. The overheating prediction algorithm obtains the predicted temperature and its changing trend at each monitoring location from the updated thermal balance dynamic model in step S802, and simultaneously obtains the actual temperature measurement value from the temperature sensor array. The temperature rise rate is obtained by performing a differential operation on the temperature measurement value.

[0109] The overheating prediction algorithm monitors two independent criteria: the rate of temperature rise and the absolute temperature value. When the rate of temperature rise exceeds a preset threshold, or the absolute temperature value at any monitoring location exceeds a preset absolute temperature threshold, it is determined that there is an overheating trend in the equipment.

[0110] Step S804: When an overheating trend is detected in the equipment, preventative control measures are implemented. These measures include reducing heating power and increasing cooling intensity, both of which are executed simultaneously. The reduction ratio of heating power is calculated based on the degree to which the current temperature exceeds the normal operating range. The degree to which the current temperature exceeds the normal operating range is measured as the ratio of the difference between the current temperature measurement and the upper limit of the normal operating parameter range to the width of the normal operating parameter range.

[0111] The reduction ratio of heating power is equal to the product of the above ratio and the preset maximum reduction ratio of heating power. The increased cooling intensity is achieved by increasing the cooling water flow rate, and the increase in cooling water flow rate is directly proportional to the reduction ratio of heating power. After the adjustment command is generated, it is sent to the contactor array and the electric regulating valve. The contactor array reduces the number of active contactors to reduce heating power, and the electric regulating valve increases the valve opening to increase the cooling water flow rate, until the temperature rise rate drops below the preset temperature rise rate threshold and the absolute temperature value at each monitoring location drops below the preset absolute temperature threshold.

[0112] Steps S801 to S802 establish a dynamic thermal balance model and continuously update it to maintain prediction accuracy. Step S803 identifies overheating trends through two independent judgment indicators. Step S804 simultaneously reduces heating power and enhances cooling intensity to eliminate overheating risks. Step S8 provides specific preventive control for equipment overheating faults, which, together with the pellet breakage fault prevention control in step S6 and the adhesion fault prevention control in step S7, constitute a preventive control system for typical faults of plastic granulators.

[0113] To facilitate understanding, in a specific application example, when the multi-sensor fusion monitoring module detects abnormal fluctuations in melt pressure and periodic impact signals collected by the vibration sensor at a certain moment, the system first performs fusion calculations on the data from each sensor using an adaptive weighted fusion algorithm to obtain comprehensive monitoring data including temperature, pressure, vibration, and current characteristics. Then, this comprehensive monitoring data is compared and analyzed with historical fault cases in the plastic granulator's operating status knowledge base. Simultaneously, a dynamic threshold algorithm is used to calculate the dynamic threshold corresponding to the current operating status. Based on the comparative analysis results, and following the method disclosed in step S6, the system determines that the current operating status exhibits pre-granulation characteristics using the pre-granulation warning identification model. This triggers the process parameter active adjustment control system to execute a granulation fault prevention control strategy, automatically reducing the screw speed and adjusting the feed rate. Simultaneously, the warning module pushes attention-level warning information to the field control panel and management backend. The entire process achieves automated closed-loop control from anomaly detection, fault identification, warning distribution to parameter adjustment.

[0114] In another specific application example, when the granulator operates at high load for an extended period, causing the temperature in the heating zone to rise continuously, the multi-point temperature data collected by the temperature sensor array shows an uneven temperature field distribution, with some areas approaching the safe upper limit. The dynamic heat balance model calculates the current heat input, accumulation, and dissipation status in real time, and the overheating prediction algorithm detects that both the temperature rise rate and the absolute temperature value are close to the threshold. The system determines that there is an overheating trend and immediately triggers the equipment overheating fault prevention and control strategy, automatically reducing the heating power while increasing the cooling water flow. At the same time, the abnormal trend early warning mechanism calculates the severity value based on the degree of temperature deviation and the rate of temperature rise, outputs a warning level warning message, and pushes it to a mobile terminal to notify maintenance personnel. The process parameter adjustment and control system continuously monitors the temperature change trend and gradually adjusts the heating power and cooling intensity until the equipment temperature returns to the normal operating range. This example demonstrates the collaborative working mechanism of multi-parameter linkage adjustment and overheating prevention and control.

[0115] In another specific application example, when the granulator discharge rate suddenly increases, the temperature field distribution in the granulator head plate area changes, and multi-point temperature sensor data shows an increase in the cooling zone temperature. The adhesion trend prediction algorithm calculates an adhesion risk index by combining temperature field characteristics and discharge rate parameters. When the risk index exceeds a preset threshold, the system automatically triggers a cooling intensity adjustment mechanism. The process parameter adjustment and control system increases the cooling water flow rate via an electric regulating valve and adjusts the cooling time to adapt to the new discharge rate. The system continuously monitors changes in the temperature field distribution, and when the adhesion risk index falls back to a safe range, the cooling intensity gradually returns to normal levels. This example illustrates the close cooperation mechanism between the adhesion prevention and control strategy and the process parameter adjustment and control system.

[0116] On the other hand, the multi-sensor fusion-based plastic pelletizer operation status monitoring system disclosed in this invention includes: The multi-sensor fusion monitoring module is used to collect multi-dimensional operating data of the plastic granulator in real time, and to preprocess and fuse the raw data collected by each sensor using an adaptive weighted fusion strategy to output comprehensive monitoring data. The adaptive weighted fusion strategy is configured to monitor the consistency of the output data of each sensor in real time. When the data of a certain sensor deviates from the data of other sensors and the deviation exceeds the preset deviation threshold, the weight coefficient of the certain sensor is automatically reduced and the weight coefficients of other sensors are increased accordingly. The plastic granulator operation status knowledge base and fault feature base are used to store the normal operation parameter range, historical fault case feature data and expert experience rules, providing a data foundation for dynamic threshold calculation, abnormal trend identification and fault type identification. The dynamic threshold comparison and analysis engine is used to implement the dynamic threshold algorithm, which compares and analyzes comprehensive monitoring data with dynamic thresholds to identify abnormal trends. The dynamic threshold algorithm is configured to combine a preset fixed threshold with a dynamic adjustment amount calculated based on historical operating data, and to configure differentiated adjustment factors for different parameter types and fault types. An abnormal trend early warning mechanism is used to automatically trigger tiered early warnings after an abnormal trend is identified. It calculates a comprehensive severity value by combining the degree of abnormal deviation, the rate of change, and the duration, and determines the warning level based on the comprehensive severity value. The process parameter active adjustment control system is used to adjust the process parameters of the granulator in conjunction with the identified fault type and warning level. The process parameter active adjustment control system is configured to perform a feasibility check after generating the parameter adjustment command and before issuing it. The feasibility check includes checking whether the adjusted parameter values ​​exceed the safe operating range of the equipment, and when the adjusted parameter combination falls into the parameter combination range corresponding to historical fault cases, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault cases.

[0117] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0118] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multi-sensor fusion based plastic granulator running condition monitoring method, characterized in that, Includes the following steps: The multi-sensor fusion monitoring module collects multi-dimensional operating data of the plastic granulator in real time, and adopts an adaptive weighted fusion strategy to preprocess and fuse the raw data collected by each sensor to output comprehensive monitoring data. The adaptive weighted fusion strategy monitors the consistency of the output data of each sensor in real time. When the data of a certain sensor deviates from the data of other sensors and the deviation exceeds a preset deviation threshold, the weight coefficient of the certain sensor is automatically reduced and the weight coefficients of other sensors are increased accordingly. The comprehensive monitoring data is compared and analyzed with a dynamic threshold to identify abnormal trends. The dynamic threshold is dynamically adjusted by a dynamic threshold algorithm based on the current operating status of the plastic granulator. The dynamic threshold algorithm combines a preset fixed threshold and a dynamic adjustment amount calculated based on historical operating data, and configures differentiated adjustment factors for different parameter types and fault types. After an abnormal trend is detected, a graded early warning mechanism is automatically triggered. The graded early warning mechanism calculates a comprehensive severity value by comprehensively considering the degree of abnormal deviation, the rate of change, and the duration of the current abnormal state, and determines the early warning level based on the comprehensive severity value. After the graded early warning mechanism is triggered, the process parameters are actively adjusted and controlled. Based on the identified fault type and the early warning level, the process parameters of the granulator are adjusted in a coordinated manner to eliminate the abnormal state. The process parameter active adjustment control performs a feasibility check after generating the parameter adjustment command and before issuing it. The feasibility check includes checking whether the adjusted parameter values ​​exceed the safe operating range of the equipment, and when the adjusted parameter combination falls into the parameter combination range corresponding to historical fault cases, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault cases.

2. The multi-sensor fusion based plastic granulator operating condition monitoring method according to claim 1, characterized in that, The method further includes: A knowledge base and a fault feature database for the operation status of a plastic granulator are established and stored. The knowledge base and the fault feature database for the operation status of the plastic granulator adopt a hierarchical structured storage design, which includes a raw data layer, a feature extraction layer and a rule reasoning layer. Data is linked between layers through an indexing mechanism to support the data chain from rule triggering to feature matching and then to tracing the original data; The rule reasoning layer stores fault diagnosis rules and reasoning logic based on expert experience. The fault diagnosis rules are represented in the form of production rules. The reasoning execution of production rules is calculated based on the matching degree, which is defined as the ratio of the number of successfully matched condition items to the total number of condition items in the condition part of the production rule.

3. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 2, characterized in that: The feature extraction layer extracts time-domain and frequency-domain features from the preprocessed sensor data and concatenates them in a preset dimensional order to form a multi-dimensional feature vector, which serves as a matching fault mode representation vector. When identifying fault types, the similarity between the current multidimensional feature vector converted from the current integrated monitoring data and the multidimensional feature vector of the typical cases stored under each fault type is calculated, and the fault type to which the typical case with the highest similarity belongs is determined as the current fault type. The similarity calculation uses a weighted combination of Euclidean distance and cosine similarity.

4. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 1, characterized in that: In the dynamic threshold algorithm, the dynamic adjustment amount is calculated based on the statistical characteristics of the operating data in the most recent time period and the degree of deviation between the current comprehensive monitoring data and historical data; The formula for calculating the dynamic threshold is: wherein, is a dynamic threshold value at the moment, is a preset fixed threshold value, is a dynamic adjustment amount at the moment, is an adjustment factor; The adjustment factor is differentiated configured according to parameter type and fault type, for temperature parameter, the value range is 0.1 to 0.3; for pressure parameter, the value range is 0.2 to 0.4; for vibration parameter, the value range is 0.4 to 0.6; for rotation speed parameter and current parameter, the value range is 0.2 to 0.

5.

5. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 1, characterized in that: The comprehensive severity value is obtained by normalizing and weighting the features of the three dimensions of abnormal deviation, change rate and duration. The weight coefficients are configured differently according to the parameter type and fault type. When the temperature is abnormal, the weight coefficient corresponding to the abnormal deviation is higher. When the vibration is abnormal, the weight coefficient corresponding to the change rate is higher. The sum of the three weight coefficients is 1. The warning levels are divided into four levels: normal, attention, warning, and severe. When the warning level is severe, a forced shutdown protection is triggered. The forced shutdown protection includes cutting off the heating power supply and the drive motor power supply, while keeping the cooling system running.

6. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 1, characterized in that: The active adjustment control of process parameters determines the combination of parameters to be adjusted and their adjustment order based on the identified fault type. The adjustment priority of each parameter is determined based on the sensitivity of each parameter to the fault response. The sensitivity is quantified by the correlation coefficient between parameter changes and fault occurrence time in historical fault data. The parameter with the largest absolute value of the correlation coefficient has the highest adjustment priority. The adjustment amount of each parameter is calculated based on the deviation between the current abnormal state and the target state. The formula for calculating the adjustment amount is as follows: wherein, is an adjustment amount for the th parameter, is a proportional coefficient for the th parameter, is a deviation value for the th parameter, , is a current value for the th parameter, is a target value for the th parameter.

7. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 1, characterized in that, The step of identifying abnormal trends also includes: The comprehensive monitoring data and historical operating data are compared and analyzed with weighted values. An exponential decay weighting method is used to assign higher weight coefficients to recent historical data and lower weight coefficients to distant historical data. The weight decay coefficients are configured according to the operating condition change cycle of the granulator. When the comprehensive monitoring data continuously deviates from the dynamic threshold and the weighted deviation from the historical operating condition data exceeds the preset anomaly judgment threshold, it is identified as having an abnormal trend.

8. The method for monitoring the operating status of a plastic granulator based on multi-sensor fusion according to claim 1, characterized in that, The method further includes: A pellet breakage prevention and control strategy is implemented. This strategy is based on the coupling relationship between melt pressure and screw speed. It utilizes a pre-trained pellet breakage precursor identification model based on a multi-layer neural network to monitor the changing trends of melt pressure and screw speed in real time and output the probability of pellet breakage risk. When the probability of pellet breakage exceeds a preset risk probability threshold, or the instantaneous drop in melt pressure exceeds a preset pressure drop threshold, or the fluctuation range of screw speed exceeds a preset speed fluctuation threshold, it is determined that there is a precursor to pellet breakage. Preventive adjustments are automatically executed to reduce screw speed and feed rate until the melt pressure returns to the normal operating parameter range and the probability of pellet breakage drops below the preset risk probability threshold.

9. The method for monitoring the operating status of a plastic pelletizer based on multi-sensor fusion according to claim 1, characterized in that, The method further includes: implementing at least one of an adhesion fault prevention and control strategy and an equipment overheating fault prevention and control strategy; The adhesion failure prevention and control strategy is based on multi-point temperature sensor array data. It reconstructs the three-dimensional temperature field distribution of the pellet mill head plate area through interpolation calculation method to evaluate the pellet cooling status in real time. It calculates the adhesion risk index by combining the temperature field characteristics output by the three-dimensional temperature field distribution model and the current discharge rate parameter. When the adhesion risk index exceeds the preset adhesion risk index threshold, it automatically triggers the cooling intensity adjustment. The temperature field features include the average temperature of each spatial point in the headplate region, the proportion of the area of ​​the headplate region whose temperature is higher than the preset adhesion risk temperature threshold in the temperature field to the total area of ​​the headplate region, and the proportion of the area of ​​the headplate region whose temperature gradient exceeds the preset gradient threshold in the temperature field to the total area of ​​the headplate region. The equipment overheating fault prevention and control strategy is based on multi-point temperature data collected by a temperature sensor array and a dynamic thermal balance model. The dynamic thermal balance model calculates the heat input, heat accumulation and heat loss of the granulator in real time, and uses a recursive least squares algorithm to update the model parameters online based on real-time sensor data. When the rate of temperature rise or the absolute temperature value exceeds the preset threshold, it is determined that there is an overheating trend of the equipment, and preventive control measures such as reducing heating power and increasing cooling intensity are implemented.

10. A monitoring system for the operating status of a plastic pelletizer based on multi-sensor fusion, characterized in that, The system for performing the method as described in any one of claims 1 to 9 includes: The multi-sensor fusion monitoring module is used to collect multi-dimensional operating data of the plastic granulator in real time, and to preprocess and fuse the raw data collected by each sensor using an adaptive weighted fusion strategy to output comprehensive monitoring data. The adaptive weighted fusion strategy is configured to monitor the consistency of the output data of each sensor in real time. When the data of a certain sensor deviates from the data of other sensors and the deviation exceeds a preset deviation threshold, the weight coefficient of the certain sensor is automatically reduced and the weight coefficients of other sensors are increased accordingly. The plastic granulator operation status knowledge base and fault feature base are used to store the normal operation parameter range, historical fault case feature data and expert experience rules, providing a data foundation for dynamic threshold calculation, abnormal trend identification and fault type identification. A dynamic threshold comparison and analysis engine is used to implement a dynamic threshold algorithm, which compares and analyzes the comprehensive monitoring data with dynamic thresholds to identify abnormal trends. The dynamic threshold algorithm is configured to combine a preset fixed threshold with a dynamic adjustment amount calculated based on historical operating data, and to configure differentiated adjustment factors for different parameter types and fault types. An abnormal trend early warning mechanism is used to automatically trigger a graded early warning after an abnormal trend is identified. It calculates a comprehensive severity value by combining the degree of abnormal deviation, the rate of change, and the duration, and determines the warning level based on the comprehensive severity value. The process parameter active adjustment control system is used to adjust the process parameters of the granulator in conjunction with the identified fault type and the warning level. The process parameter active adjustment control system is configured to perform a feasibility check after generating the parameter adjustment command and before issuing it. The feasibility check includes checking whether the adjusted parameter values ​​exceed the safe operating range of the equipment, and when the adjusted parameter combination falls into the parameter combination range corresponding to historical fault cases, the adjustment amount of each parameter is reduced one by one in order of adjustment priority from low to high, until the adjusted parameter combination does not fall into the parameter combination range corresponding to any historical fault cases.

Images