A heat pump oven fault diagnosis method

Abstract

The application discloses a kind of heat pump curing room fault diagnosis methods, it is related to intelligent control and equipment health management technical field, including S1, through sensor network from curing room interior heater fan and humidity regulator Multivariate real-time data including temperature humidity airflow is collected, multivariate real-time data collected is handled using neural network model to identify the fluctuation mode of current production environment, obtain the initial control scheme of dynamic parameter;S2, according to the initial control scheme of dynamic parameter obtained, obtain external environmental variable and batch change information, using clustering algorithm to group matching is carried out to external environmental variable and initial control scheme, determine the optimization control instruction suitable for the characteristics of current material;The heat pump curing room fault diagnosis and residual life evaluation calculation method, realizes the intelligent health evaluation and adaptive operation optimization of curing room equipment, improves the security, reliability and overall production efficiency of drying process.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and equipment health management technology, specifically to a method for diagnosing faults in a heat pump drying oven. Background Technology

[0002] As crucial equipment in agricultural product processing, food production, and industrial drying, the precise control and fault correlation analysis of drying barns are essential for improving production efficiency, ensuring product quality, and reducing maintenance costs. The operation of drying barns involves complex thermodynamic processes and equipment coordination, requiring precise control of parameters such as temperature, humidity, and airflow. Simultaneously, fault analysis optimizes maintenance decisions to reduce downtime and resource waste. Research in this field is not only a cornerstone for promoting the intelligent transformation of agriculture and industry but also a vital guarantee for achieving efficient and sustainable production. Existing drying barn control and fault analysis methods have significant limitations in practical applications. Traditional solutions often rely on manual experience or simple rule-based control, making it difficult to adapt to changes in dynamic production environments. For example, when processing different materials or batches, environmental parameters fluctuate significantly, making it difficult for existing methods to adjust control strategies in real time, leading to increased energy consumption or unstable product quality. Furthermore, fault diagnosis is usually based on historical data from individual devices, ignoring the mutual influence between devices and failing to capture potential correlation problems in complex systems, thus affecting the accuracy of maintenance decisions. The core technical challenge of precise control and fault correlation analysis lies in achieving dynamic parameter control and system-level fault correlation. During the operation of a drying barn, parameters such as temperature and humidity need to be adjusted in real time according to the characteristics of the materials and the external environment. However, existing control systems struggle to respond quickly to changes in multiple coupled variables, leading to control lags. For example, in the drying of agricultural products, excessively high temperatures or improper humidity control can cause over-drying or a decline in quality. More complexly, the operating states of various devices in the drying barn system (such as heaters, fans, and sensors) influence each other. A minor malfunction in one component can trigger a chain reaction, but current technology struggles to extract correlations between devices from massive amounts of operational data, making it impossible to accurately predict the root cause and scope of the malfunction. Therefore, the lag in dynamic parameter control and the lack of system-level fault correlation make it difficult for drying barns to achieve efficient and stable production in actual operation. Taking agricultural product drying as an example, when the drying task switches to different types of materials, the control system cannot optimize parameters in a timely manner, resulting in energy waste or uneven drying. Furthermore, when a piece of equipment malfunctions, maintenance personnel lack system-level correlation analysis, making it difficult to quickly locate the root cause and prolonging downtime. These two problems are intertwined and become key issues in achieving efficient operation and intelligent maintenance of drying barns. Summary of the Invention

[0003] The purpose of this invention is to provide a method for diagnosing faults in heat pump ovens, thereby solving the problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for fault diagnosis of a heat pump drying oven, comprising: S1, collecting real-time multivariate data including temperature, humidity, and airflow from the heater fan and humidity regulator inside the drying oven via a sensor network, and processing the collected real-time multivariate data using a neural network model to identify the fluctuation pattern of the current production environment and obtain an initial control scheme for dynamic parameters; S2, based on the obtained initial control scheme for dynamic parameters, acquiring external environmental variables and batch change information, and using a clustering algorithm to group and match the external environmental variables with the initial control scheme to determine an optimized control command adapted to the current material characteristics; S3, if the temperature value in the optimized control command exceeds a preset threshold, adjusting the output power of the humidity regulator through a feedback loop mechanism to balance the coupled changes of multiple variables, and judging the stability of the control command. S4. Extract equipment operation log data from the control commands in the stable state, and use a neural network model to analyze the abnormal patterns in the equipment operation log data to obtain a preliminary correlation map of potential faults; S5. For the obtained preliminary correlation map of potential faults, obtain historical fault datasets, and use a clustering algorithm to compare the similarity between the historical fault datasets and the preliminary correlation map to determine the location of the fault root cause under the influence of system equipment; S6. If the determined fault root cause location involves the fan component, calculate the impact path of this location on the heater and sensor through a simulated chain propagation model to obtain the predicted range of the chain reaction; S7. Based on the obtained predicted range of the chain reaction, generate a maintenance priority sequence and send an alarm signal to the control system to determine the recovery path of the overall curing barn operation in order to achieve efficient parameter control and fault correlation analysis;

[0005] S2 includes:

[0006] Raw data on material type and batch variation are acquired from a sensor network outside the baking oven. Data cleaning methods are used to denoise and unify the format of the raw data to obtain a standardized dataset of external environmental variables.

[0007] The K-means clustering algorithm is used to group and match the standardized external environmental variable dataset with the initial control scheme parameter set. Clustering is performed based on the feature vectors of material type and batch change to obtain the group feature set corresponding to the current material characteristics.

[0008] If the matching degree between the material type or batch change data in the group feature set and the parameter set of the initial control scheme is lower than a preset threshold, the temperature, humidity and airflow parameters in the group feature set are adjusted by a weighted average algorithm to obtain an optimized control instruction set.

[0009] The operating parameters of heater control, humidity regulation and fan operation are adjusted according to the optimized control instruction set. The adjusted material type and batch change data are collected through the sensor network to obtain the updated external environmental variable dataset.

[0010] S6 includes:

[0011] Real-time data acquisition results are obtained from the operation logs of the wind turbine components. The data acquisition results are processed using time series analysis tools to extract the operating status characteristics of the wind turbine components and obtain a time series dataset.

[0012] If the operating status features in the time series dataset exceed a preset threshold, an association rule mining tool is used to analyze the operating status features and the historical operating data of the heater and sensor to determine the set of association rules.

[0013] Based on the set of association rules, a chain propagation model is used to calculate the propagation path of wind turbine component failures to heaters and sensors, and the path weight distribution is obtained.

[0014] For the path weight distribution, a prediction model is used to analyze the matching degree between the path weight distribution and historical chain reaction data to determine the final scope of influence.

[0015] S7 includes:

[0016] The chain reaction prediction range is obtained from sensor data and historical data, and the operating status of each device is determined by a pre-established decision tree algorithm to obtain the probability value of the chain reaction occurring.

[0017] A maintenance priority sequence is generated based on the probability value of the chain reaction and the operating status. If the priority is higher than a preset threshold, the maintenance priority sequence is determined.

[0018] A list of high-priority devices is obtained from the maintenance priority sequence, and an alarm signal is sent to the control system. If the device priority is higher than a preset threshold, the target of the alarm signal is determined through real-time monitoring.

[0019] Based on the alarm signal and real-time monitoring data, a preset recovery path algorithm is used to generate a recovery path for the operation of the drying room, thereby obtaining the optimized parameter control scheme.

[0020] Preferably, step S1 includes acquiring temperature, humidity, and airflow data from sensors in the heater, fan, and humidity regulator inside the drying oven; integrating the data into a unified multivariate real-time dataset using a data fusion algorithm; processing the multivariate real-time dataset using a neural network; extracting features from the temperature, humidity, and airflow data using a pre-established neural network model to obtain the fluctuation pattern feature set; if the temperature or humidity data in the fluctuation pattern feature set exceeds a preset threshold, calculating dynamic parameter control values ​​based on the fluctuation pattern feature set to obtain the initial control scheme parameter set; adjusting the operating parameters of the heater control, humidity regulation, and fan operation based on the initial control scheme parameter set; and feeding back the adjusted temperature, humidity, and airflow data through a sensor network to obtain an updated multivariate real-time dataset.

[0021] Preferably, step S3 includes acquiring real-time data on material type, batch changes, and current temperature values ​​from a sensor network; processing the data using a data standardization method to obtain a standardized environmental variable dataset; if the temperature value in the standardized environmental variable dataset exceeds a preset threshold, calculating the coupling deviation between the temperature value, humidity value, and airflow parameters through a feedback loop mechanism to obtain a deviation adjustment value; adjusting the output power of the humidity regulator based on the deviation adjustment value to generate a new control command set and determining the updated humidity value and airflow parameters; collecting data on material type, batch changes, and temperature values ​​after the control command set is executed through a sensor network; evaluating the stable state of the control command set using a threshold judgment method to obtain a stable state result.

[0022] Preferably, step S4 includes obtaining timestamps, device identifiers, and operating parameters from control instructions; performing structured processing on the control instructions using instruction parsing technology to generate a structured operating log dataset; and determining the completeness of the operating log dataset. If the operating log dataset is complete, log analysis technology is used to clean the operating log dataset, removing noise data to generate a cleaned log dataset, resulting in standardized operating log data. For the cleaned log dataset, a neural network is used with a multilayer perceptron algorithm to analyze the fluctuation patterns of operating parameters, generating an abnormal pattern dataset and identifying anomalies in the abnormal pattern dataset. Based on the abnormal pattern dataset, association analysis technology is used to perform frequent itemset mining on the anomalies, generating association rules between potential faults and abnormal patterns, and obtaining a preliminary association map of the potential faults.

[0023] Preferably, step S5 includes: acquiring fault record data from a historical fault dataset; grouping the fault record data using a clustering algorithm; calculating the similarity between each group and the preliminary association graph to obtain the fault record data group with the highest similarity; extracting the device identifier and fault timestamp contained in the fault record data group; analyzing the device identifier and fault timestamp using an association rule mining tool to determine a candidate device set; if the candidate device set contains at least one device, acquiring the operation log data of each device in the candidate device set; comparing the operation log data with the fault pattern of the preliminary association graph using a time series analysis tool to determine the most matching device and obtain a preliminary judgment result; and based on the preliminary judgment result, acquiring relevant fault repair records from the historical fault dataset; comparing the fault repair records with the preliminary association graph using a pattern matching tool to determine the final fault root cause location.

[0024] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0025] This method for fault diagnosis and remaining life assessment of heat pump drying ovens, by constructing an intelligent diagnosis and life assessment approach that integrates multi-source sensor data, neural network analysis, and cluster modeling, achieves dynamic control of the drying oven operation and system-level fault correlation identification. Compared with traditional methods relying on manual experience or rule-based control, this method can automatically generate and optimize control strategies based on real-time collected parameters such as temperature, humidity, and airflow, as well as external material characteristics, significantly improving the response speed and stability of temperature and humidity control. Simultaneously, through deep learning and cluster analysis of equipment operation logs, the system can identify potential associated faults and trace them back to the root cause components, supporting early warning and maintenance decisions, and avoiding cascading downtime caused by localized faults. Combined with a remaining life prediction model based on historical data, this invention achieves intelligent health assessment and adaptive operation optimization of drying oven equipment, reducing energy consumption and maintenance costs, and improving the safety, reliability, and overall production efficiency of the drying process. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method for diagnosing faults and assessing remaining life of a heat pump oven according to the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] like Figure 1As shown, the present invention provides a technical solution: a method for diagnosing faults in a heat pump drying room, comprising:

[0029] S1. Collect multivariate real-time data including temperature, humidity and airflow from the heaters, fans and humidity regulators inside the drying oven through a sensor network. Use a neural network model to process the collected multivariate real-time data to identify the fluctuation pattern of the current production environment and obtain the initial control scheme of dynamic parameters.

[0030] S2. Based on the initial control scheme of the obtained dynamic parameters, obtain the external environmental variables and batch change information, and use a clustering algorithm to group and match the external environmental variables with the initial control scheme to determine the optimized control instructions that are suitable for the current material characteristics.

[0031] S3. If the temperature value in the optimized control command exceeds the preset threshold, the output power of the humidity regulator is adjusted through a feedback loop mechanism to balance the multivariate coupling changes and determine the stable state of the control command.

[0032] S4. Extract equipment operation log data from the control commands in a stable state, and use a neural network model to analyze the abnormal patterns in the equipment operation log data to obtain a preliminary correlation map of potential faults.

[0033] S5. Based on the preliminary correlation map of potential faults, obtain the historical fault dataset, and use a clustering algorithm to compare the similarity between the historical fault dataset and the preliminary correlation map to determine the location of the root cause of the fault under the influence of system equipment.

[0034] S6. If the determined root cause of the fault involves the wind turbine components, the impact path of that location on the heater and sensor is calculated by simulating the chain propagation model to obtain the predicted range of the chain reaction.

[0035] S7. Based on the predicted range of the chain reaction, generate a maintenance priority sequence and send an alarm signal to the control system to determine the recovery path of the overall curing barn operation in order to achieve efficient parameter control and fault correlation analysis.

[0036] This implementation uses a multi-sensor network to continuously monitor the multivariate environment (temperature, humidity, airflow, etc.) inside the drying oven. Neural network algorithms are used to learn and analyze this real-time data, extracting potential patterns of dynamic environmental changes and initially formulating a control plan. Based on this, and considering external environmental conditions and batch-to-batch material differences, a clustering algorithm is used to establish a matching mechanism between the initial plan and the actual environment to optimize the generation of control commands. When the temperature abnormally rises in the control plan, a feedback mechanism dynamically adjusts the humidity output power to restore the system to a stable state. Furthermore, neural network analysis of the equipment operation logs under stable conditions reveals preliminary correlation paths between abnormal behavior and potential faults. This path, combined with historical fault database information, is further compared and analyzed through clustering to accurately locate the root cause of the fault. If the root cause is located in the fan assembly, the system further assesses its potential impact on other key system nodes, such as heaters and sensors, based on a simulated chain reaction model, defining the risk propagation range. Finally, a maintenance priority strategy is formulated based on the fault propagation impact path, and alarms and recovery suggestions are issued to the control system, achieving intelligent self-recovery and optimization of the entire system operation.

[0037] This method integrates artificial intelligence and big data processing technologies into the traditional heat pump oven system, enhancing its adaptability to complex environmental changes. Through multivariate data fusion analysis and feedback control mechanisms, the system can promptly detect and correct operational deviations, effectively preventing fault propagation. A multi-level fault diagnosis process based on neural networks and clustering algorithms enhances the system's ability to identify and accurately trace faults in their early stages, improving the accuracy and timeliness of maintenance responses. The introduction of a chain propagation model further enhances the depth and breadth of fault prediction, achieving impact assessment from specific points to a broader scope. The final maintenance prioritization strategy not only improves maintenance efficiency but also reduces system downtime risk, enhancing the overall system reliability and lifespan.

[0038] S1 includes acquiring temperature, humidity, and airflow data from sensors in the heaters, fans, and humidity regulators inside the drying oven; integrating the data into a unified multivariate real-time dataset using a data fusion algorithm; processing the multivariate real-time dataset using a neural network; extracting features from the temperature, humidity, and airflow data using a pre-established neural network model to obtain a fluctuation pattern feature set; if the temperature or humidity data in the fluctuation pattern feature set exceeds a preset threshold, calculating dynamic parameter control values ​​based on the fluctuation pattern feature set to obtain an initial control scheme parameter set; adjusting the operating parameters of the heater control, humidity regulation, and fan operation based on the initial control scheme parameter set; and feeding back the adjusted temperature, humidity, and airflow data through a sensor network to obtain an updated multivariate real-time dataset.

[0039] In one possible implementation, step S1 includes the following: First, multiple temperature sensors, humidity sensors, and airflow sensors are installed inside the heat pump oven, mounted on heaters, fans, and humidity regulators. Each sensor has real-time sampling and signal transmission capabilities. The temperature sensor detects the instantaneous temperature of the air inside the oven; the humidity sensor detects the relative humidity of water vapor in the air; and the airflow sensor detects the airflow speed and direction. Each sensor operates continuously at a sampling frequency of once per second, and transmits the collected raw temperature data, humidity data, and airflow data to the main processing unit of the control system via a sensor network.

[0040] After receiving the three types of raw data, the main processing unit first executes a data fusion algorithm to form a unified multivariate real-time dataset. This fusion algorithm is completed through three steps: time alignment, noise removal, and weighted averaging. The time alignment step ensures data synchronization by calibrating the timestamps of all sensors to a unified system clock. The noise removal step uses a sliding window method to calculate the average value of each data set over 3 seconds, thereby eliminating abnormal readings caused by instantaneous fluctuations. The weighted averaging step determines the weights according to the representativeness of the sensor locations: sensors near the center of the drying chamber are weighted at 0.4, sensors near the air outlet at 0.3, and sensors near the air inlet at 0.3. The weighted calculation yields representative values ​​for overall temperature, humidity, and airflow. After completing these three steps, a unified multivariate real-time dataset is obtained.

[0041] The control system then inputs this multivariate real-time dataset into a pre-built and trained neural network model. This neural network model consists of an input layer, hidden layers, and an output layer. The input layer receives time-series data on temperature, humidity, and airflow; the hidden layer extracts correlation features between the data using a nonlinear function; and the output layer outputs a set of fluctuation pattern features. This feature set is a set of feature values ​​representing the changing trends of temperature, humidity, and airflow, used to reflect the dynamic changes in the internal environment of the drying chamber.

[0042] When the system detects that the temperature or humidity data in the characteristic set of the fluctuation pattern exceeds a preset threshold, the system enters the dynamic parameter calculation phase. The threshold is determined as follows: during the initial system run, temperature, humidity, and airflow data are continuously collected for 72 hours, and their average value and standard deviation are calculated. The temperature threshold is set as the average value plus twice the standard deviation, and the humidity threshold is set as the average value plus 1.5 times the standard deviation. Taking a common tobacco drying condition as an example, if the average temperature is 55 degrees Celsius and the standard deviation is 3 degrees Celsius, then the temperature threshold is 61 degrees Celsius; if the average humidity is 70% and the standard deviation is 5%, then the humidity threshold is 77.5%. Exceeding this threshold is considered an abnormal fluctuation in the environment.

[0043] Upon detecting an anomaly, the control system calculates dynamic parameter control values ​​based on the fluctuation pattern feature set. The calculation process comprises three stages: First, it extracts the temperature, humidity, and airflow change rates over the past 30 seconds as short-term fluctuation indicators; second, it extracts the average values ​​from the past 10 minutes as long-term stability indicators; third, it compares the short-term fluctuation indicators with the long-term stability indicators and calculates the deviation. The larger the absolute value of the deviation, the more severely the system deviates from its normal state. The control system then generates a proportional control amount based on the deviation value, serving as the dynamic parameter control value.

[0044] These dynamic parameter control values ​​constitute the initial control scheme parameter set. The parameter set includes three sub-parameters: temperature control value, humidity control value, and airflow control value. The temperature control value adjusts the heater's output power, the humidity control value regulates the humidity regulator's output voltage, and the airflow control value controls the fan speed. For example, when the temperature exceeds a threshold, the temperature control value becomes negative, and the control system automatically reduces the heater's output power by approximately 10%; simultaneously, the humidity control value slightly increases, raising the humidity regulator's output power by approximately 5%, thereby increasing the evaporation equilibrium temperature; the airflow control value increases by approximately 8% to enhance airflow and accelerate temperature diffusion.

[0045] After the above adjustments are completed, the control system re-collects real-time data on temperature, humidity, and airflow through a sensor network, forming an updated multivariate real-time dataset. This dataset is then input into the neural network model for real-time analysis to verify whether the control effect has stabilized. If the temperature and humidity fluctuations are both less than 2% and remain stable for more than 5 minutes, the system determines that an equilibrium state has been reached; if fluctuations still exist, the system automatically repeats the above calculation and control process until the environment returns to normal.

[0046] S2 includes acquiring raw data on material type and batch changes from a sensor network outside the baking oven, and using data cleaning methods to denoise and unify the format of the raw data to obtain a standardized dataset of external environmental variables.

[0047] The K-means clustering algorithm is used to group and match the standardized external environmental variable dataset with the initial control scheme parameter set. Clustering is performed based on the feature vectors of material type and batch change to obtain the group feature set corresponding to the current material characteristics.

[0048] If the matching degree between the material type or batch change data in the group feature set and the parameter set of the initial control scheme is lower than a preset threshold, the temperature, humidity and airflow parameters in the group feature set are adjusted by a weighted average algorithm to obtain an optimized control instruction set.

[0049] The operating parameters of heater control, humidity regulation and fan operation are adjusted according to the optimized control instruction set. The adjusted material type and batch change data are collected through the sensor network to obtain the updated external environmental variable dataset.

[0050] In one possible implementation, step S2 includes the following: First, the control system acquires raw data on material type and batch variations through a sensor network installed outside the curing barn. The external sensor network includes a barcode recognition sensor for identifying material type, a humidity sensor for detecting material moisture content, a temperature sensor for detecting ambient temperature, and a batch recording unit for recording batch number and weight. The barcode recognition sensor scans the barcode information on the material surface before each batch enters the curing barn to obtain the material name and number, such as tobacco leaf number 001, tea leaf number 002, and timber number 003. The humidity sensor collects external air humidity data once per second, the temperature sensor collects external ambient temperature data once per second, and the batch recording unit records loading time, weight, and material number. All data collected by the sensors is synchronously transmitted to the main processing unit of the control system via a data communication module to form a raw dataset.

[0051] After acquiring the raw data, the control system performs data cleaning. Data cleaning includes three steps: noise reduction, missing data completion, and format standardization. First, the noise reduction step calculates the average value over a continuous 5-second period using a sliding time window method. If the deviation between the current sampled value and the 5-second average exceeds 10%, the data is considered abnormal and discarded. Second, the missing data completion step automatically fills in missing batch numbers or weight information by calling the average value of the previous batch of similar materials. Third, the format standardization step converts all temperature data to degrees Celsius, humidity data to percentages, times to seconds, and material and batch numbers to Arabic numerals, forming a standardized dataset of external environmental variables with a consistent format.

[0052] Next, the control system invokes the K-means clustering algorithm to group and match the standardized dataset of external environmental variables with the initial control scheme parameter set. The execution process of the K-means clustering algorithm includes four stages. The first stage determines the number of cluster centers. Based on the common material categories in the curing barn, three cluster centers are set, corresponding to tobacco leaves, wood, and tea leaves respectively. The second stage randomly selects three data points as initial cluster centers. The third stage calculates the distance difference between each data point and the three cluster centers. This distance difference is calculated by summing the absolute differences of temperature, humidity, and airflow parameters; the smaller the value, the closer the data point is to the cluster center. The fourth stage assigns each data point to the group corresponding to the cluster center with the smallest distance. The average temperature, average humidity, and average airflow values ​​of each group are calculated as new cluster centers, and the above process is repeated until the clustering results stabilize. Finally, a grouped feature set corresponding to the current material characteristics is obtained.

[0053] The grouped feature set includes three sets of parameters: material type number, batch variation number, average temperature, average humidity, and average airflow. These parameters characterize the operational characteristics of different batches of materials under different environmental conditions. The control system then calculates the matching degree between the grouped feature set and the initial control scheme parameter set. The matching degree is calculated by comparing the temperature, humidity, and airflow differences between the two sets of data item by item, taking the absolute value of each difference, dividing it by the baseline value of the initial control parameters, and then calculating the average percentage of the three. The system presets a matching degree threshold of 10%, which is determined through experimental verification. When the average percentage is below 10%, the system determines that the matching degree is low, indicating that the initial control scheme is inconsistent with the current material characteristics and needs to be re-optimized. The process of determining this 10% threshold is as follows: in multiple batch tests, by adjusting different thresholds and observing changes in system energy consumption and operational stability, it was found that when the threshold is set at 10%, the system response time and energy consumption reach a balance, so this threshold is fixed as the optimal threshold.

[0054] When the matching degree is detected to be below 10%, the control system executes a weighted average algorithm to adjust the temperature, humidity, and airflow parameters in the grouped feature set. The weighted average algorithm includes the following three steps: First, setting parameter weights. The weight for temperature is set to 0.4, the weight for humidity is set to 0.3, and the weight for airflow is set to 0.3. The weights are determined based on experimental results of heat transfer characteristics in the drying chamber, with temperature having the greatest impact on the drying process, accounting for 40%, followed by humidity and airflow, each accounting for 30%. Second, calculating correction values. The control system calculates the difference between each parameter in the grouped feature set and the initial control scheme, and multiplies the difference by the corresponding weight. Third, calculating the average correction value and adding the result to the original parameter values ​​to obtain the optimized temperature, humidity, and airflow parameters.

[0055] After calculation, the system generates an optimized control instruction set. This set includes three instructions: heater control instruction, humidity control instruction, and fan operation instruction. The heater control instruction adjusts the output power based on the corrected temperature parameters. When the corrected temperature parameter is 2 degrees Celsius higher than the original set value, the control system reduces the heating power by 10%; when it is below 2 degrees Celsius, the power is increased by 10%. The humidity control instruction adjusts the evaporation power based on the corrected humidity parameters. When the humidity is 3% lower than the set value, the humidity regulator power is increased by 8%, and decreased by 8% when it is higher than 3%. The fan operation instruction adjusts the fan speed based on the corrected airflow parameters. When the airflow is lower than the set value, the fan speed is increased by 10%, and decreased by 10% when it is higher than the set value.

[0056] After performing the above three adjustments, the control system re-collects external environmental variable data through the sensor network, generating an updated external environmental variable dataset. The new dataset is then matched and analyzed against the optimized control command set again. When the matching degree recovers to above 10% and the fluctuation range is less than 2% within 5 minutes, the system determines that the optimized control command is valid and enters a stable operating state; if the requirements are still not met, the system automatically repeats the above clustering and weighted adjustment process until the stability criteria are met.

[0057] In this implementation, all parameters are determined through deterministic calculations without subjective experience. The number of cluster centers is directly determined by the number of material types, and the weight values ​​are fixed through experimental statistics as follows: temperature 0.4, humidity 0.3, and airflow 0.3. The matching degree threshold is fixed at 10% through energy efficiency testing. All calculations are performed using the absolute difference and weighted average of the sampled data, avoiding complex formula derivations and ensuring the simplicity of the algorithm and the reproducibility of the results.

[0058] S3 includes acquiring real-time data on material type, batch changes, and current temperature values ​​from a sensor network, and using a data standardization method to process the data in a unified format to obtain a standardized environmental variable dataset.

[0059] If the temperature value in the standardized environmental variable dataset exceeds a preset threshold, the coupling deviation between the temperature value, humidity value, and airflow parameter is calculated through a feedback loop mechanism to obtain a deviation adjustment value.

[0060] Adjust the output power of the humidity regulator according to the deviation adjustment value, generate a new set of control instructions, and determine the updated humidity value and airflow parameters;

[0061] The sensor network collects data on material type, batch changes, and temperature values ​​after the execution of the control command set. A threshold judgment method is used to evaluate the stable state of the control command set to obtain the stable state result.

[0062] In one possible implementation, the specific implementation process of step S3 is as follows: First, the control system acquires data on material type, batch changes, and current temperature values ​​from inside the drying oven in real time through a sensor network. Material type is identified by barcode identification sensors, batch changes are recorded by batch number and loading time using a batch recording unit, and temperature values ​​are collected by temperature sensors distributed in the heating zone, air outlet, and air return vent. All sensors operate at a sampling frequency of once per second and are synchronized to the central control system via a data transmission module to form a raw real-time dataset. Before the data enters the analysis stage, the system first performs data standardization processing on material type, batch changes, and temperature values. The data standardization method involves three steps: format unification, unit correction, and time alignment. Step 1, format unification: all material numbers are encoded with Arabic numerals, batch changes are represented by loading sequence numbers, and temperatures are expressed in degrees Celsius. Step 2, unit correction: ensuring that temperature data collected by different sensors are represented within the same range, unifying to the range of 0 to 100 degrees Celsius. Step 3, time alignment: all data are corrected using a unified timestamp via the system clock to ensure that each set of data corresponds to the same operating state at the same time. After the above processing, a standardized environmental variable dataset is obtained. The control system performs threshold judgments on temperature values ​​in the standardized environmental variable dataset. The preset temperature threshold is determined based on historical system data. Specifically, during the initial operation phase, temperature changes are continuously recorded for 72 hours, and the average temperature and standard deviation are calculated. The temperature threshold is set to the average temperature plus twice the standard deviation. For example, if the average temperature is 55 degrees Celsius and the standard deviation is 3 degrees Celsius, the temperature threshold is 61 degrees Celsius. If the real-time temperature value exceeds 61 degrees Celsius, the system determines it to be in an over-limit state and enters the feedback calculation phase.

[0063] In the feedback loop mechanism, the system calculates the coupling deviation between temperature, humidity, and airflow parameters. The calculation process involves three steps: First, the system uses humidity and airflow sensors to obtain current humidity and airflow values. Humidity is expressed as a percentage, and airflow parameters are expressed in meters per second. Second, the system calculates the deviation differences between temperature, humidity, and airflow. This is done by comparing the current temperature with the average temperature over the previous 5 minutes to obtain the temperature offset; and by comparing humidity and airflow with their respective historical averages to obtain the humidity offset and airflow offset. Third, the system determines the degree of impact of temperature anomalies on humidity and airflow based on the ratios of the temperature offset to the humidity and airflow offsets. If the temperature offset is large and humidity and airflow changes increase synchronously, it indicates that the temperature rise is mainly due to insufficient air circulation or low humidity. The system sums the differences between the three offsets to obtain the total coupling deviation. This total coupling deviation is the comprehensive balance difference between temperature, humidity, and airflow, reflecting the degree of coupling imbalance among the three.

[0064] The system calculates the deviation adjustment value based on the total coupling deviation. This deviation adjustment value guides the output correction of the humidity regulator. It is determined as follows: when the total coupling deviation is greater than 0, it indicates that the temperature is too high and the humidity is too low. The system multiplies the total deviation by an adjustment coefficient of 0.5 to obtain the humidity adjustment increment. When the total coupling deviation is less than 0, it indicates that the temperature is too low and the humidity is too high. The system multiplies the absolute value of the total deviation by an adjustment coefficient of 0.3 to obtain the humidity adjustment decrement. The adjustment coefficient is determined through equipment calibration experiments, where 0.5 is used to enhance humidity compensation and 0.3 is used to suppress excessive humidity output.

[0065] The control system adjusts the output power of the humidity regulator based on the deviation adjustment value, generating a new set of control commands. This set includes two parts: a humidity adjustment command and an airflow adjustment command. The humidity adjustment command increases or decreases the humidity regulator's output power by a corresponding percentage. For example, a positive deviation adjustment value of 4 indicates that the humidity regulator's output power needs to be increased by 4%; a negative deviation adjustment value of 3 indicates that the humidity regulator's output power needs to be decreased by 3%. The airflow adjustment command automatically adjusts the fan speed based on the new humidity setpoint. When humidity increases, the fan speed increases by 5% to enhance airflow; when humidity decreases, the fan speed decreases by 5% to maintain thermal balance. After adjustment, the system records the new humidity value and airflow parameters, forming an updated environmental variable dataset.

[0066] Next, the control system uses a sensor network to collect data on material type, batch changes, and temperature values ​​again, detecting the operating status after the execution of the control commands. The system uses a threshold judgment method to evaluate the stability after control. This evaluation method includes three judgment criteria: first, within 5 minutes of continuous sampling, the temperature fluctuation range does not exceed ±1 degree Celsius; second, the humidity fluctuation range does not exceed ±2%; and third, the airflow fluctuation range does not exceed ±0.3 meters per second. If all three conditions are met, the system determines that a stable state has been reached and outputs a stable state result of "stable"; if any one of them exceeds the range, the output result is "unstable," and the system automatically repeats the above feedback adjustment process until a stable state is reached.

[0067] In this embodiment, the parameters are determined as follows: the temperature threshold is determined by historical statistical calculation; the deviation of humidity and airflow is obtained by comparing real-time data with historical average values; the deviation adjustment value is calculated by multiplying the total coupling deviation by a fixed adjustment coefficient; the adjustment coefficient is determined to be 0.5 and 0.3 through experimental calibration; the temperature, humidity, and airflow fluctuation range in the stability judgment criteria are determined by long-term operation testing of the equipment to ensure that the balance between energy consumption and stability is maintained within the normal fluctuation range.

[0068] S4 includes obtaining timestamps, device identifiers, and operating parameters from control commands, performing structured processing on the control commands using command parsing technology, generating a structured operating log dataset, and determining the integrity of the operating log dataset;

[0069] If the runtime log dataset is complete, log analysis technology is used to clean the runtime log dataset. By removing noisy data, a cleaned log dataset is generated, resulting in standardized runtime log data.

[0070] For the cleaned log dataset, a neural network is used to analyze the fluctuation patterns of the operating parameters through a multilayer perceptron algorithm to generate an abnormal pattern dataset and identify the outliers in the abnormal pattern dataset.

[0071] Based on the abnormal pattern dataset, frequent itemset mining is performed on the abnormal points using association analysis techniques to generate association rules between potential faults and abnormal patterns, thereby obtaining a preliminary association map of the potential faults.

[0072] In one possible implementation, step S4 is carried out as follows: First, the control system extracts the timestamp, equipment identifier, and operating parameters from the control commands generated in the previous step. The timestamp represents the specific moment each command was generated, recorded in a six-segment numerical format of year, month, day, hour, minute, and second, for example, 14:30:25 on October 16, 2025 is recorded as 20251016143025. The equipment identifier is a unique number for each drying room device, such as heater number 101, fan number 102, and humidity regulator number 103. The operating parameters include heating power, humidity output power, and fan speed, expressed as percentages, for example, heating power is 80%, humidity output power is 65%, and fan speed is 70%. The above information is output from the control system in the form of data frames and stored in the log cache module.

[0073] Next, the system employs command parsing technology to structure the control commands. The command parsing process includes three steps: First, the raw commands are segmented into three categories of data: timestamp, device identifier, and operating parameters. Second, each category of data undergoes type identification and unit conversion to ensure that the time is in time format, the device identifier is in numeric format, and the operating parameters are in percentage format. Third, the parsing results are arranged in chronological order and stored as a structured operating log dataset. The structured log dataset is organized in tabular form, with each row containing a timestamp, device number, and corresponding operating parameters.

[0074] After the structured log dataset is generated, the system checks its integrity. Integrity checking includes two parts: time continuity checking and data field integrity checking. Time continuity checking is performed by calculating the time difference between two adjacent log records. When the time interval exceeds 10 seconds, the system considers the log to be missing and automatically marks it as abnormal. Data field integrity checking determines whether each log record contains all three fields (timestamp, device identifier, and operating parameters). When all records meet the time continuity and field integrity requirements, the system determines the log dataset is complete and proceeds to the next step; if incomplete, a re-collection process is triggered until the data is complete.

[0075] Once the structured operational log dataset is complete, the system employs log analysis techniques to clean it. Data cleaning includes three steps: noise identification, anomaly removal, and standardization. First, noise identification: the system calculates the difference in operational parameters between three consecutive log entries. If any difference exceeds 20%, the record is marked as noisy data. Second, anomaly removal: the marked noisy records are deleted from the dataset. Third, standardization: all operational parameters are uniformly converted to a percentage range of 0 to 100, and timestamps are formatted to a uniform length. After completion, the system obtains the cleaned log dataset, where each record represents the stable operational status of a device at a specific time.

[0076] The system then inputs the cleaned log dataset into a neural network model for analysis. This neural network, constructed using a multilayer perceptron algorithm, comprises an input layer, hidden layers, and an output layer. The input layer receives three parameters from each log record: heating power, humidity output power, and fan speed. The hidden layer contains several neural nodes used to extract the patterns of change among the operating parameters. The output layer outputs fluctuation pattern feature values ​​to determine abnormal operating conditions of the system. The neural network generates an anomaly pattern dataset by analyzing parameter change trends at different time points. This anomaly pattern dataset contains feature values ​​for each time period and their corresponding anomaly scores. The anomaly scores reflect the degree of fluctuation of the operating parameters, ranging from 0 to 100. When the score is higher than 80, the system determines that there is a significant anomaly in that time period.

[0077] Next, the system performs anomaly identification and frequent itemset mining on the abnormal pattern dataset. Anomaly identification is completed by analyzing anomaly scores one by one. When the anomaly scores of three consecutive log entries are all higher than 80, the system marks that time period as an anomaly. Anomalies represent potential abnormal states in system operation, such as excessive fluctuations in fan speed or unstable humidity output. Subsequently, the system uses association analysis technology to perform frequent itemset mining on the anomalies. The mining process includes four steps: First, counting the frequency of device numbers appearing in all anomalies; second, filtering device numbers with a frequency exceeding 10% of the total records as high-frequency items; third, counting the co-occurrence frequency among high-frequency items; and fourth, establishing the association between potential faults and abnormal patterns based on the co-occurrence frequency.

[0078] The association rules between potential faults and abnormal patterns are generated based on the association analysis results. For example, if the co-occurrence frequency of fan number 102 and humidity controller number 103 at the abnormal point reaches 15%, a rule is generated: "Unstable fan operation may lead to abnormal humidity control." All association rules constitute a preliminary association map of potential faults. This map represents equipment numbers as nodes and fault associations as lines. The more nodes and the denser the lines, the higher the degree of fault association among the equipment groups.

[0079] In this implementation, the integrity judgment threshold of 10-second time interval is determined by the system sampling frequency; the 20% difference threshold for noise identification is derived from experimental data statistics to ensure the accuracy of anomaly removal; the judgment threshold of 80% for anomaly score is determined by historical operational sample analysis to balance the ratio of false positives to false negatives; and the 10% frequency threshold for high-frequency items is the optimal feature boundary point of the system under empirical statistics. All parameters are fixed during system initialization and do not rely on manual adjustment, ensuring repeatability and objectivity.

[0080] S5 includes obtaining fault record data from historical fault datasets, grouping the fault record data using a clustering algorithm, calculating the similarity of each group with the preliminary association map, and obtaining the fault record data group with the highest similarity.

[0081] For the fault record data grouping, the included device identifier and fault timestamp are extracted, and the device identifier and fault timestamp are analyzed using association rule mining tools to determine the candidate device set;

[0082] If the candidate device set contains at least one device, then the operation log data of each device in the candidate device set is obtained, and the operation log data is compared with the failure mode of the preliminary correlation map using a time series analysis tool to determine the most matching device and obtain a preliminary judgment result.

[0083] Based on the preliminary judgment results, relevant fault repair records are obtained from the historical fault dataset. A pattern matching tool is used to compare the fault repair records with the preliminary correlation map to determine the final fault root cause location.

[0084] In one possible implementation, the specific implementation process of step S5 is as follows: First, the control system obtains fault record data from the historical fault dataset. The historical fault dataset is a collection of historical operation logs automatically saved during the long-term operation of the system, containing the timestamps of each device's fault occurrence, device identifiers, fault types, and repair status. The timestamps are recorded in a time format accurate to the second, such as 20251016103045 representing 10:30:45 on October 16, 2025; device identifiers are represented by numerical codes, such as heater number 101, fan number 102, and humidity regulator number 103; fault types include categories such as over-temperature, fan blockage, and abnormal humidity.

[0085] The control system first performs clustering on the fault record data. The clustering algorithm groups historical fault records into several groups based on feature similarity. The process includes the following four steps: First, the system selects three features: fault type, occurrence time interval, and involved equipment number. Second, the system determines the number of clusters, automatically setting it to 5 groups based on the number of fault samples. Third, the system calculates the difference between any two fault records on the three features; the smaller the difference, the higher the similarity. Fourth, the records with the smallest difference are grouped into the same group, and this process is repeated until all fault records are grouped into their respective groups. After clustering, five fault record data groups are obtained.

[0086] Then, the system calculates the similarity between each group and the preliminary association map generated in the previous step. The similarity is calculated by comparing the overlap ratio of device numbers in the group with the device nodes in the preliminary association map item by item. For example, if a group contains device numbers 101, 102, and 103, while the preliminary association map contains device numbers 101, 103, and 104, then the device matching rate for that group is 2 items overlapping, accounting for 66%. Simultaneously, the system compares the temporal distribution characteristics of both, i.e., the overlap rate between the concentrated intervals of fault timestamps in the group and the intervals of anomaly occurrences in the map. If the temporal overlap ratio is 70%, then the overall similarity is taken as the average of the device matching rate and the temporal overlap rate, which is 68%. After calculating the similarity for all groups, the system selects the group with the highest similarity as the target group, i.e., the historical record most likely related to the current potential fault.

[0087] For selected fault record data groups, the system extracts the device identifiers and fault timestamps. Then, the system uses an association rule mining tool to perform co-occurrence analysis on this data to determine the candidate device set. The association rule mining process includes four steps: First, it counts the frequency of all device numbers appearing in the fault record groups; second, it calculates the number of times any two devices co-occur within the same time period (time difference less than 60 seconds); third, it filters device combinations whose co-occurrence frequency exceeds 10% of the total sample size; fourth, it groups these device numbers into a candidate device set. The candidate device set represents a group of devices with significant fault correlations in historical data. For example, if the co-occurrence frequency of fan number 102 and humidity controller number 103 is 15%, exceeding the 10% threshold, both will be included in the candidate device set.

[0088] If the candidate device set contains at least one device, the system proceeds to the next step of analysis. The system extracts the operational log data for each device in the candidate device set from the log database. The operational log data includes three items: timestamp, device output parameters, and operational status, recording the actual operation of the device with second-level time precision. The system uses time series analysis tools to compare the operational log data with the fault modes in the preliminary correlation map. The time series comparison process includes three steps: First, the system calculates the changing trend of each parameter (such as temperature, speed, and humidity output) in the log data; second, it extracts the abnormal change characteristics of the corresponding parameters in the fault modes, such as sudden increases, sudden decreases, or periodic fluctuations; third, it compares the similarity of the direction and magnitude of the changes. When the similarity is higher than 80%, the system determines that the log characteristics of that device best match the fault mode in the map. Through the above analysis, the system obtains a preliminary judgment result, that is, determines the device number most likely to cause a fault.

[0089] After obtaining the initial assessment results, the system further extracts fault repair records related to the device from the historical fault dataset. Each repair record includes the repair time, repair measures, and post-repair operation results. The system uses a pattern matching tool to compare the repair records with the preliminary correlation map. The matching process includes the following three steps: First, the system extracts the operation steps and repair objects involved in the repair records; second, it matches these operation steps with the fault types of the corresponding nodes in the preliminary correlation map item by item; third, it calculates the matching ratio, and when the matching ratio exceeds 70%, the system confirms that the device is the final root cause of the fault.

[0090] In this embodiment, the key parameters are determined as follows: the number of clusters is fixed at 5 to balance data accuracy and calculation speed; the time overlap threshold is set to 60 seconds to ensure the temporal correlation of the same fault event; the average weight in similarity calculation is 50% each, determined by experimental verification; the frequency threshold for association analysis is set to 10% to ensure that high-frequency items are representative; the similarity judgment threshold for time series comparison is set to 80%, which is the optimal discrimination point of the system; the final confirmation threshold for pattern matching is 70%, which has been verified through historical repair to effectively reduce the misjudgment rate.

[0091] S6 includes obtaining real-time data acquisition results from the operation logs of the wind turbine components, processing the data acquisition results using a time series analysis tool, extracting the operating status characteristics of the wind turbine components, and obtaining a time series dataset;

[0092] If the operating status features in the time series dataset exceed a preset threshold, an association rule mining tool is used to analyze the operating status features and the historical operating data of the heater and sensor to determine the set of association rules.

[0093] Based on the set of association rules, a chain propagation model is used to calculate the propagation path of wind turbine component failures to heaters and sensors, and the path weight distribution is obtained.

[0094] For the path weight distribution, a prediction model is used to analyze the matching degree between the path weight distribution and historical chain reaction data to determine the final impact range.

[0095] In one possible implementation, step S6 is carried out as follows: First, the control system obtains real-time data acquisition results from the wind turbine component's operation log. The operation log includes operating parameters such as timestamps, wind turbine speed, current, voltage, power, and vibration amplitude. The timestamp is recorded as a continuous sequence of years, months, days, hours, minutes, and seconds, for example, 20251016104530 represents 10:45:30 on October 16, 2025; the speed is expressed in revolutions per minute; the current is expressed in amperes; the voltage is expressed in volts; the power is expressed in watts; and the vibration amplitude is expressed in millimeters. The sensor network records the above parameters at a sampling frequency of once per second and synchronizes them to the central control system to form a real-time operation dataset of the wind turbine component.

[0096] The collected results were then processed using time series analysis tools to extract the operating status characteristics of the wind turbine components. The analysis process consisted of four steps: First, the system sorted the data according to timestamp order to ensure time continuity; second, it calculated the 5-second sliding window average for each operating parameter to smooth short-term fluctuations; third, it extracted the rate of change of each parameter, i.e., calculated the difference between adjacent sampling points and recorded the direction of change (increasing or decreasing); fourth, it identified the operating status characteristics based on the trend of the rate of change. If the rotational speed decreased by more than 5% within 10 consecutive seconds, or the vibration amplitude increased by more than 10% consecutively, the system recorded it as an "abnormal operating trend." After the analysis was completed, the system generated a time series dataset, with each data point containing a timestamp, parameter value, and operating status identifier.

[0097] Threshold judgments are performed on the operating status characteristics in the time series dataset. Preset thresholds are determined based on the equipment's rated operating parameters: the speed deviation threshold is set to ±10% of the rated speed; the current deviation threshold is set to ±15% of the rated current; and the vibration amplitude threshold is set to 20% of the upper limit of the normal value. For example, if the rated speed is 1500 rpm, the deviation range is 1350 to 1650 rpm; if the rated current is 10 amps, the allowable range is 8.5 to 11.5 amps; and if the upper limit of normal vibration is 2 mm, the threshold is 2.4 mm. If any parameter exceeds the corresponding threshold, the system determines that the fan component is abnormal.

[0098] When an anomaly is detected, the system invokes an association rule mining tool to perform correlation analysis between the fan's operating status characteristics and the historical operating data of the heaters and sensors. The correlation analysis process includes the following steps: First, the system extracts historical operating data of the heaters and sensors from the database, including temperature output, humidity output, and sampling frequency. Second, the system aligns the abnormal time periods of the fan with the timestamps of the heaters and sensors to form a multi-device synchronized dataset. Third, the system counts the number of times the heater and sensor parameters change simultaneously during the abnormal fan period. Fourth, the system calculates the co-occurrence frequency; when the co-occurrence frequency is higher than 10% of the total number of samples, a potential correlation is considered to exist. Based on these results, the system generates a set of association rules, such as: "When fan vibration increases by 10%, the heater temperature rises by more than 3 degrees Celsius" or "When fan current rises abnormally, the sensor signal delay increases by 2 seconds." The set of association rules is stored in the form of logical relationships for subsequent propagation analysis.

[0099] Based on the obtained set of association rules, a chain propagation model is used to calculate the propagation path of the fan component failure to the heater and sensor. The propagation path calculation process includes three stages: First, establishing a node structure model. The system sets the fan, heater, and sensor as nodes, and the connections between each node represent physical or data coupling relationships. Second, determining the direction of influence between nodes based on association rules. For example, when a fan malfunction causes a heater temperature increase, a directional path "fan → heater" is established. Third, calculating path weights. Path weights represent the intensity of influence and are determined based on co-occurrence frequency. If the co-occurrence rate of fan malfunction and heater temperature increase is 20%, the path weight is 0.2; if the co-occurrence rate of fan malfunction and sensor signal delay is 15%, the path weight is 0.15. The system records all path weights and forms a weight distribution table, where each path includes a starting node, a target node, and a corresponding weight value.

[0100] Subsequently, the system performs a matching degree analysis on the path weight distribution to determine the final impact range. The matching degree analysis is performed using a predictive model built upon historical cascading failure data. The predictive model comprises three main steps: First, it extracts typical propagation paths and their corresponding impact ranges from the historical cascading failure data. The impact range is expressed as a percentage; for example, a fan failure has a 30% probability of affecting the heater and a 20% probability of affecting the sensor. Second, it compares the current path weight distribution with the weight patterns in the historical cascading failure data item by item. If the similarity between the current path weight and the historical data is higher than 80%, the path is considered a major propagation path. Third, it determines the final impact range based on the sum of the weights of the major propagation paths. For example, if the fan's path weight to the heater is 0.2 and its path weight to the sensor is 0.15, with a total of 0.35, the system determines the fan failure's impact range to be 35%. This result is output as the "Predicted Impact Range," used for subsequent maintenance priority calculations.

[0101] In this embodiment, a 5-second time window is the optimal interval for system smoothing analysis; the speed, vibration, and current thresholds are determined based on equipment calibration, with deviation ratios taken from 10%, 15%, and 20% of the rated parameters, respectively; the co-occurrence frequency threshold of 10% is the empirical limit value for fault association; path weights are directly quantified by co-occurrence ratios, without subjective factors; the similarity threshold of 80% is the best matching standard obtained by the system through statistical verification; the scope of influence is represented by the sum of path weights, ensuring quantifiability and comparability.

[0102] S7 includes obtaining the chain reaction prediction range from sensor data and historical data, using a pre-established decision tree algorithm to determine the operating status of each device, and obtaining the probability value of the chain reaction occurring;

[0103] A maintenance priority sequence is generated based on the chain reaction probability value and the operating status. If the priority is higher than a preset threshold, the maintenance priority sequence is determined.

[0104] A list of high-priority devices is obtained from the maintenance priority sequence, and an alarm signal is sent to the control system. If the device priority is higher than a preset threshold, the target of the alarm signal is determined through real-time monitoring.

[0105] Based on the alarm signal and real-time monitoring data, a preset recovery path algorithm is used to generate a recovery path for the operation of the drying room, thereby obtaining the optimized parameter control scheme.

[0106] In one possible implementation, step S7 is carried out as follows: First, the control system obtains the predicted range of the chain reaction from sensor data and historical data. Sensor data includes temperature, humidity, airflow, current, and vibration parameters of the fan, heater, and humidity controller during the current operating cycle; historical data includes similar data from previous cycles and their corresponding fault records. The control system merges the two types of data to form a complete operating database. The predicted range of the chain reaction is calculated in the previous stage (step S6) and is a weighted distribution table representing the influence relationship between devices. Each record in the table includes the path direction, influence intensity, and correlation probability between the three types of devices: fan, heater, and sensor. For example, if the influence weight of the fan on the heater is 0.25 and the influence weight on the sensor is 0.15, then the predicted range of the chain reaction is 40%.

[0107] Next, a pre-established decision tree algorithm is invoked to determine the operating status of each device. The decision tree algorithm is established through supervised learning during system initialization. Input variables include five parameters: temperature, humidity, airflow, current, and vibration. The output is the device's operating status level. The algorithm's judgment process includes the following four steps: First, the system reads the real-time operating parameters of each device. Second, it judges each parameter value layer by layer based on the comparison results with thresholds. The thresholds are determined as follows: temperature deviation threshold is ±10% of the rated value, humidity deviation threshold is ±8%, airflow deviation threshold is ±12%, current deviation threshold is ±15%, and vibration deviation threshold is ±20%. Third, the system compares each parameter sequentially along the decision tree path. When any two parameters simultaneously exceed their corresponding thresholds, the system classifies the operating status as "abnormal"; if only one parameter exceeds the limit, it is classified as "slightly abnormal"; if both are within the range, it is classified as "normal". Fourth, the algorithm statistically analyzes the operating status of all devices, generating a status result table. Each record includes the device number, current status, and corresponding probability value.

[0108] Based on the output of the decision tree algorithm, the system calculates the probability of a cascading failure. This probability represents the likelihood of a cascading failure due to coupling between devices. The calculation method involves multiplying the probability of device malfunction by the weight of the cascading failure, and then summing the results along the path direction. For example, if the probability of a fan malfunction is 0.4, the propagation weight from the fan to the heater is 0.25, and the propagation weight to the sensor is 0.15, then the probability of a cascading failure is 0.4 × (0.25 + 0.15) = 0.16, or 16%. The system calculates these probabilities for all devices to obtain a cascading failure probability table, which is used for subsequent maintenance priority assessment.

[0109] The control system generates a maintenance priority sequence based on the cascading reaction probability value and equipment operating status. The maintenance priority calculation rules are as follows: First, assign scores to equipment status: 1 for normal status, 2 for minor abnormality, and 3 for abnormality; Second, the system calculates a comprehensive score for each piece of equipment: Comprehensive score = Operating status score × Cascading reaction probability × 100; Third, sort all equipment comprehensive scores in descending order to obtain the maintenance priority sequence. For example, if the heater score is 48, the fan score is 32, and the humidity controller score is 20, then the maintenance priority order is heater, fan, humidity controller.

[0110] A priority threshold is set to determine whether to proceed with the alarm handling process. This threshold was determined through multiple runs during system initialization and is fixed at 40 points. When the maintenance priority score is higher than 40 points, the system classifies the device as a high-priority device and initiates the alarm signal generation step.

[0111] The control system extracts a list of high-priority devices from the maintenance priority sequence, generates and sends alarm signals. The alarm signal includes the device number, alarm time, operating status, and cascading reaction probability value. The specific process of the system sending the signal to the control platform is as follows: First, the system selects the device with the highest score from the high-priority device list as the primary alarm target; second, it calls the real-time monitoring module to confirm the current operating status of the device. If the real-time monitoring data shows that the device is still in an abnormal state for more than 10 seconds, the system confirms the alarm signal is valid; third, the alarm signal is sent to the control system's main interface through the communication interface, simultaneously triggering the audible and visual warning module.

[0112] Based on alarm signals and real-time monitoring data, a preset recovery path algorithm is executed to generate a recovery path for the drying chamber operation. The recovery path algorithm is a fixed rule model used to plan the recovery sequence and parameter adjustment strategy under multi-device coupling conditions. The algorithm execution process includes the following three steps: First, the system starts with high-priority devices and determines the recovery sequence according to their scoring order. If the heater has the highest score, its parameter recovery operation is performed first. Second, the system calls the corresponding recovery strategy based on the device type. For heaters, the output power is reduced by 5%; for fans, the speed is reduced by 10%; for humidity regulators, the evaporation power is reduced by 8%. Third, the system monitors temperature, humidity, and airflow fluctuations within 5 minutes after the recovery operation. When the fluctuations of all three indicators are below 2% and remain stable for more than 3 minutes, the system determines that the recovery is complete. The algorithm outputs a recovery path table, which includes the device recovery sequence, recovery time, and adjusted parameter values.

[0113] Finally, the system generates an optimized parameter control scheme based on the recovery path table. This scheme integrates real-time operating data, maintenance priorities, and recovery path information, automatically adjusting the target output values ​​of each device to restore the system as a whole to a stable state. For example, when the heater temperature recovers to 58 degrees Celsius, the fan speed stabilizes at 1500 rpm, and the humidity remains at 68%, the system saves this state as an optimized parameter scheme and stores it in the database for automatic recall in subsequent cycles.

[0114] In this embodiment, all parameters are determined using deterministic numerical settings: temperature, humidity, airflow, current, and vibration thresholds are determined through equipment calibration data; the priority threshold of 40 points is determined through energy consumption and response time experiments; the abnormal duration threshold of 10 seconds is used to filter transient fluctuations; and the fluctuation stability judgment value of 2% is the optimal range obtained through empirical testing. All calculation steps follow fixed logic, the source of parameters is traceable, and the algorithm output results are stable and consistent.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing faults in a heat pump drying room, characterized in that, include: S1. Collect multivariate real-time data including temperature, humidity and airflow from the heaters, fans and humidity regulators inside the drying oven through a sensor network. Use a neural network model to process the collected multivariate real-time data to identify the fluctuation pattern of the current production environment and obtain the initial control scheme of dynamic parameters. S2. Based on the initial control scheme of the obtained dynamic parameters, obtain the external environmental variables and batch change information, and use a clustering algorithm to group and match the external environmental variables with the initial control scheme to determine the optimized control instructions that are suitable for the current material characteristics. S3. If the temperature value in the optimized control command exceeds the preset threshold, the output power of the humidity regulator is adjusted through a feedback loop mechanism to balance the multivariate coupling changes and determine the stable state of the control command. S4. Extract equipment operation log data from the control commands in a stable state, and use a neural network model to analyze the abnormal patterns in the equipment operation log data to obtain a preliminary correlation map of potential faults. S5. Based on the preliminary correlation map of potential faults, obtain the historical fault dataset, and use a clustering algorithm to compare the similarity between the historical fault dataset and the preliminary correlation map to determine the location of the root cause of the fault under the influence of system equipment. S6. If the determined root cause of the fault involves the wind turbine components, the impact path of that location on the heater and sensor is calculated by simulating the chain propagation model to obtain the predicted range of the chain reaction. S7. Based on the predicted range of the chain reaction, generate a maintenance priority sequence and send an alarm signal to the control system to determine the recovery path of the overall curing barn operation in order to achieve efficient parameter control and fault correlation analysis. S2 includes: Raw data on material type and batch variation are acquired from a sensor network outside the baking room. Data cleaning methods are used to denoise and unify the format of the raw data to obtain a standardized dataset of external environmental variables. The K-means clustering algorithm is used to group and match the standardized external environmental variable dataset with the initial control scheme parameter set. Clustering is performed based on the feature vectors of material type and batch change to obtain the group feature set corresponding to the current material characteristics. If the matching degree between the material type or batch change data in the grouped feature set and the parameter set of the initial control scheme is lower than a preset threshold, the temperature, humidity and airflow parameters in the grouped feature set are adjusted by a weighted average algorithm to obtain an optimized control instruction set. The operating parameters of heater control, humidity regulation and fan operation are adjusted according to the optimized control instruction set. The adjusted material type and batch change data are collected through the sensor network to obtain the updated external environmental variable dataset. S6 includes: Real-time data acquisition results are obtained from the operation logs of the wind turbine components. The data acquisition results are processed using time series analysis tools to extract the operating status characteristics of the wind turbine components and obtain a time series dataset. If the operating status features in the time series dataset exceed a preset threshold, an association rule mining tool is used to analyze the operating status features and the historical operating data of the heater and sensor to determine the set of association rules. Based on the set of association rules, a chain propagation model is used to calculate the propagation path of wind turbine component failures to heaters and sensors, and the path weight distribution is obtained. For the path weight distribution, a prediction model is used to analyze the matching degree between the path weight distribution and historical chain reaction data to determine the final scope of influence. S7 includes: The chain reaction prediction range is obtained from sensor data and historical data, and the operating status of each device is determined by a pre-established decision tree algorithm to obtain the probability value of the chain reaction occurring. A maintenance priority sequence is generated based on the probability value of the chain reaction and the operating status. If the priority is higher than a preset threshold, the maintenance priority sequence is determined. A list of high-priority devices is obtained from the maintenance priority sequence, and an alarm signal is sent to the control system. If the device priority is higher than a preset threshold, the target of the alarm signal is determined through real-time monitoring. Based on the alarm signal and real-time monitoring data, a preset recovery path algorithm is used to generate a recovery path for the operation of the drying room, thereby obtaining the optimized parameter control scheme.

2. The method for diagnosing faults in a heat pump drying room according to claim 1, characterized in that: S1 includes: Temperature, humidity, and airflow data are acquired from sensors in the heaters, fans, and humidity regulators inside the drying room. The data is then integrated into a unified multivariate real-time dataset using a data fusion algorithm. The multivariate real-time dataset is processed using a neural network. The temperature data, humidity data, and airflow data are extracted using a pre-established neural network model to obtain the fluctuation pattern feature set. If the temperature or humidity data in the fluctuation pattern feature set exceeds a preset threshold, then the dynamic parameter control value is calculated based on the fluctuation pattern feature set to obtain the initial control scheme parameter set. The operating parameters of heater control, humidity regulation and fan operation are adjusted according to the initial control scheme parameter set. The adjusted temperature data, humidity data and airflow data are fed back through the sensor network to obtain the updated multivariate real-time dataset.

3. The method for diagnosing faults in a heat pump drying room according to claim 1, characterized in that: S3 includes: Real-time data on material type, batch changes, and current temperature values ​​are acquired from a sensor network. The data is then processed to unify the format using a data standardization method to obtain a standardized environmental variable dataset. If the temperature value in the standardized environmental variable dataset exceeds a preset threshold, the coupling deviation between the temperature value, humidity value, and airflow parameter is calculated through a feedback loop mechanism to obtain a deviation adjustment value. Adjust the output power of the humidity regulator according to the deviation adjustment value, generate a new set of control instructions, and determine the updated humidity value and airflow parameters. The sensor network collects data on material type, batch changes, and temperature values ​​after the execution of the control command set. A threshold judgment method is used to evaluate the stable state of the control command set to obtain the stable state result.

4. The method for diagnosing faults in a heat pump drying room according to claim 1, characterized in that: S4 includes: The timestamp, device identifier, and operating parameters are obtained from the control instructions. The control instructions are then processed in a structured manner using instruction parsing technology to generate a structured operating log dataset, and the integrity of the operating log dataset is determined. If the runtime log dataset is complete, log analysis technology is used to clean the runtime log dataset. By removing noisy data, a cleaned log dataset is generated, resulting in standardized runtime log data. For the cleaned log dataset, a neural network is used to analyze the fluctuation patterns of the operating parameters through a multilayer perceptron algorithm to generate an abnormal pattern dataset and identify the outliers in the abnormal pattern dataset. Based on the abnormal pattern dataset, frequent itemset mining is performed on the abnormal points using association analysis techniques to generate association rules between potential faults and abnormal patterns, thereby obtaining a preliminary association map of the potential faults.

5. The method for diagnosing faults in a heat pump drying room according to claim 1, characterized in that: S5 includes: Fault record data is obtained from historical fault datasets. The fault record data is grouped using a clustering algorithm. The similarity between each group and the preliminary association map is calculated to obtain the fault record data group with the highest similarity. For the fault record data grouping, the included device identifier and fault timestamp are extracted, and the device identifier and fault timestamp are analyzed using association rule mining tools to determine the candidate device set; If the candidate device set contains at least one device, then the operation log data of each device in the candidate device set is obtained, and the operation log data is compared with the failure mode of the preliminary correlation map using a time series analysis tool to determine the most matching device and obtain a preliminary judgment result. Based on the preliminary judgment results, relevant fault repair records are obtained from the historical fault dataset. A pattern matching tool is used to compare the fault repair records with the preliminary correlation map to determine the final fault root cause location.

Images