Abnormality early warning method and system used in fan manufacturing process

Abstract

The invention provides an abnormity early warning method and system used in a fan manufacturing process. The abnormity early warning method comprises the steps that a multi-source process data stream composed of part forming process data, structure assembling process data and performance testing process data in the fan manufacturing process is acquired; performing process feature analysis processing on the multi-source process data stream to obtain a time sequence evolution feature process coupling feature; performing combined anomaly identification processing on the time sequence evolution features and the process coupling features through a pre-deployed manufacturing anomaly diagnosis model to generate a process anomaly identification result; based on the front-back sequence relation of all the procedures in the fan manufacturing process and the abnormal probability value and the possibility level in the procedure abnormal identification result, the abnormal target procedure node and the associated procedure range of the abnormal influence are positioned; and generating an early warning instruction according to the target process node and the associated process range, and pushing the early warning instruction to a manufacturing execution system. Abnormity real-time monitoring and early warning in the fan manufacturing process are achieved, and the fan manufacturing quality and production efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of fan intelligent manufacturing, in particular to an abnormality early warning method and system for fan manufacturing process. BACKGROUND

[0002] In the fan manufacturing process, due to the involvement of multiple complex processes such as component forming, structure assembly and performance testing, there is a close correlation and dependence relationship between each process. The abnormality detection in the traditional fan manufacturing process mainly relies on manual inspection and post-quality detection, which is not only inefficient, but also difficult to accurately identify and locate at the early stage of abnormality, leading to untimely abnormality handling, which may cause more serious quality problems, and even affect the operation efficiency and product quality of the entire production line. With the development of intelligent manufacturing technology, although some automatic monitoring systems have been applied to the fan manufacturing process, these systems can only process data of a single process, and lack comprehensive analysis and abnormality early warning capability for multi-source process data flow, and cannot comprehensively and accurately reflect the potential abnormality risks in the fan manufacturing process. SUMMARY

[0003] The purpose of the present application is to provide an abnormality early warning method and system for fan manufacturing process, which can comprehensively and accurately reflect the potential abnormality risks in the fan manufacturing process.

[0004] To achieve the above-mentioned purpose, the first aspect of the embodiments of the present disclosure provides an abnormality early warning method for fan manufacturing process, the method comprising: acquiring a multi-source process data flow composed of continuously collected component forming process data, structure assembly process data and performance testing process data with time stamp identification in the fan manufacturing process; performing process feature analysis processing on the multi-source process data flow to obtain time sequence evolution features for characterizing the change rule of process parameters in the time dimension in each process data and process coupling features for characterizing the parameter correlation relationship between different processes; performing joint abnormality identification processing on the time sequence evolution features and the process coupling features through a pre-deployed manufacturing abnormality diagnosis model to generate process abnormality identification results including abnormal probability values of each process data and possibility levels of abnormality occurrence; locating the target process node of abnormality occurrence and the range of associated processes affected by abnormality based on the sequence relationship of each process in the fan manufacturing process, the abnormal probability values in the process abnormality identification results and the possibility levels; According to the target process node and the associated process range, a pre-warning instruction including process positioning information and an abnormal response strategy is generated, and the pre-warning instruction is pushed to a manufacturing execution system, and the pre-warning instruction is used to trigger a visualized abnormal response operation in the manufacturing execution system.

[0005] In some preferred embodiments, the target process node where the abnormality occurs and the associated process range affected by the abnormality are located based on the sequence relationship between the processes in the fan manufacturing process, the abnormal probability value in the process abnormality identification result, and the possibility level. The abnormal probability value in the process abnormality identification result is analyzed, and the time stamp with an abnormal probability value exceeding a preset threshold is selected as an abnormal time point; According to the time sequence evolution characteristics corresponding to the abnormal time point, the time sequence evolution characteristics of the component forming process, the structure assembly process, and the performance test process are compared with the difference degree of the normal state characteristic template, and the process with the largest difference degree is taken as the target process node where the abnormality occurs; The process coupling characteristics corresponding to the abnormal time point are extracted, the strength of the process thermal coupling characteristics and the process geometric coupling characteristics of the target process node and other processes are analyzed, and the process with a coupling characteristic strength exceeding a preset coupling threshold is selected as a candidate process; According to the sequence relationship between the processes in the fan manufacturing process, if the possibility level of the previous process of the candidate process exceeds a preset level threshold, the candidate process is taken as an associated process affected by the abnormality; The manufacturing process path of the target process node and the associated process is analyzed, and according to the sequence relationship between the processes in the fan manufacturing process, the process range boundary affected by the abnormality is determined; The process identification of the target process node, the process identification of the associated process, and the process range boundary are used to locate the associated process range affected by the abnormality.

[0006] In some preferred embodiments, the time sequence evolution characteristics corresponding to the abnormal time point are extracted, the time sequence evolution characteristics of the component forming process, the structure assembly process, and the performance test process are compared with the difference degree of the normal state characteristic template, and the process with the largest difference degree is determined as the target process node where the abnormality occurs. The time sequence evolution characteristics of the component forming process corresponding to the abnormal time point are obtained, and the time sequence evolution characteristics of the component forming process include temperature change rate characteristics, pressure change acceleration characteristics, and mold movement stability characteristics; The time sequence evolution characteristics of the structure assembly process corresponding to the abnormal time point are obtained, and the time sequence evolution characteristics of the structure assembly process include assembly rhythm characteristics, bolt fastening uniformity characteristics, and gap control precision characteristics; Obtaining a time sequence evolution feature of the performance test procedure corresponding to the abnormal time point, the time sequence evolution feature of the performance test procedure including a vibration complexity feature, a bearing wear feature and a motor efficiency feature; Obtaining a preset normal state feature template, the normal state feature template including a normal temperature change rate range, a normal pressure change acceleration range and a normal mold movement stability range of the component forming procedure, a normal assembly rhythm range, a normal bolt fastening uniformity range and a normal gap control precision range of the structure assembly procedure, a normal vibration complexity range, a normal bearing wear range and a normal motor efficiency range of the performance test procedure; Calculating the Euclidean distance between the time sequence evolution feature of the component forming procedure and the normal state feature template as a first difference value, calculating the Euclidean distance between the time sequence evolution feature of the structure assembly procedure and the normal state feature template as a second difference value, and calculating the Euclidean distance between the time sequence evolution feature of the performance test procedure and the normal state feature template as a third difference value; Comparing the first difference value, the second difference value and the third difference value, and selecting the procedure with the largest difference value as the target procedure node with abnormality.

[0007] In some preferred embodiments, the procedure coupling feature corresponding to the abnormal time point is extracted, the strength of the procedure thermal coupling feature and the procedure geometric coupling feature of the target procedure node and other procedures is analyzed, and the procedure with a coupling feature strength exceeding a preset coupling threshold is screened as a candidate procedure, including: Extracting a procedure thermal coupling feature value of the target procedure node and the component forming procedure from the procedure coupling feature, the procedure thermal coupling feature value representing the parameter correlation degree of the target procedure node and the component forming procedure; Extracting a procedure geometric coupling feature value of the target procedure node and the structure assembly procedure from the procedure coupling feature, the procedure geometric coupling feature value representing the parameter correlation degree of the target procedure node and the structure assembly procedure; Extracting a comprehensive coupling feature value of the target procedure node and the performance test procedure from the procedure coupling feature, the comprehensive coupling feature value being calculated by weighted average of the procedure thermal coupling feature value and the procedure geometric coupling feature value; Setting a preset coupling threshold, the preset coupling threshold being the mean value of the normal procedure coupling feature value plus twice the standard deviation; Screening the procedures with the procedure thermal coupling feature value, the procedure geometric coupling feature value and the comprehensive coupling feature value exceeding the preset coupling threshold, and taking the procedures as candidate procedures.

[0008] In some preferred embodiments, the process feature analysis processing on the multi-source process data stream obtains time sequence evolution features for characterizing variation rules of process parameters in time dimension in each process data and process coupling features for characterizing parameter correlation relationships between different processes, which includes: dividing the multi-source process data stream into analysis windows with fixed time length, each analysis window containing continuous multiple time stamp data points; performing trend extraction processing on the component forming process data in each analysis window, calculating the first order difference of the temperature fluctuation sequence as a temperature change rate feature, calculating the second order difference of the pressure change sequence as a pressure change acceleration feature, calculating the direction consistency index of the stroke trajectory sequence as a mold movement stability feature, and combining the temperature change rate feature, the pressure change acceleration feature and the mold movement stability feature as time sequence evolution features of the component forming process; performing pattern recognition processing on the structure assembly process data in each analysis window, calculating the periodic fluctuation frequency of the docking offset sequence as an assembly rhythm feature, calculating the standard deviation of the torque fluctuation sequence as a bolt fastening uniformity feature, calculating the extreme difference of the size change sequence as a gap control accuracy feature, and combining the assembly rhythm feature, the bolt fastening uniformity feature and the gap control accuracy feature as time sequence evolution features of the structure assembly process; performing statistical analysis processing on the performance test process data in each analysis window, calculating the energy distribution entropy of the dynamic balance vibration sequence as a vibration complexity feature, calculating the main frequency offset of the noise spectrum sequence as a bearing wear feature, calculating the mean shift of the power loss sequence as a motor efficiency feature, and combining the vibration complexity feature, the bearing wear feature and the motor efficiency feature as time sequence evolution features of the performance test process; performing cross-correlation analysis processing on the time sequence evolution features of different processes, calculating the covariance of the temperature change rate feature of the component forming process and the bolt fastening uniformity feature of the structure assembly process as a process thermal coupling feature, calculating the mutual information value of the gap control accuracy feature of the structure assembly process and the vibration complexity feature of the performance test process as a process geometric coupling feature, and combining the process thermal coupling feature and the process geometric coupling feature as process coupling features.

[0009] In some preferred embodiments, the joint abnormality identification processing on the time sequence evolution features and the process coupling features by the pre-deployed manufacturing abnormality diagnosis model generates process abnormality identification results including abnormal probability values and possibility levels of abnormality occurrence of each process data, which includes: inputting the time sequence evolution features and the process coupling features into the feature encoding layer of the manufacturing abnormality diagnosis model, and mapping into an encoding feature vector through a fully connected network; Performing time series modeling processing on the encoded feature vector through the time series memory module of the manufacturing anomaly diagnosis model to extract time series context features including historical feature information and current feature information; The association reasoning module of the manufacturing anomaly diagnosis model performs inter-process relationship modeling on the temporal context features to extract association features including process coupling logic. The association reasoning module adopts a graph neural network structure in which nodes represent process features and edges represent process coupling features. Inputting the temporal context features and the associated features into the abnormality discrimination layer of the manufacturing abnormality diagnosis model, and calculating the abnormality probability value of each process data through a logistic regression classifier; The abnormality probability value is subjected to sliding window smoothing processing to generate an abnormality probability value corresponding to the process data of each timestamp and a probability level of abnormality occurrence.

[0010] In some preferred embodiments, generating an early warning instruction including process location information and an abnormal response strategy based on the target process node and the associated process range includes: Acquire process identification information of the target process node, where the process identification information of the target process node includes the process name, process equipment number, and position coordinates of the process in the manufacturing flow chart; Acquire process identification information of the associated process range, wherein the process identification information of the associated process range includes the associated process name, the associated process equipment number, and the position coordinates of the associated process in the manufacturing flow chart; Combining the process identification information of the target process node and the process identification information of the associated process range into process location information; Obtaining a preset abnormal response strategy library, wherein the abnormal response strategy library includes equipment shutdown instructions, parameter adjustment instructions, and manual intervention instructions corresponding to different process nodes; Matching a corresponding exception response strategy from the exception response strategy library according to the process identification information of the target process node; The process location information is combined with the abnormal response strategy to generate the early warning instruction.

[0011] In some preferred embodiments, matching a corresponding exception response strategy from the exception response strategy library according to the process identification information of the target process node includes: Parsing the process name in the process identification information of the target process node to determine whether the process name belongs to a component molding process, a structure assembly process, or a performance testing process; If the process name belongs to the component forming process, an abnormal response strategy containing mold temperature adjustment instructions, pressure parameter correction instructions, and mold stroke calibration instructions is extracted from the abnormal response strategy library; If the process name belongs to the structure assembly process, an abnormal response strategy containing assembly position correction instructions, bolt torque compensation instructions, and gap size adjustment instructions is extracted from the abnormal response strategy library; If the process name belongs to the performance test process, an abnormal response strategy containing vibration suppression control instructions, bearing lubrication supplement instructions, and motor power optimization instructions is extracted from the abnormal response strategy library; Obtain the process equipment number of the target process node, extract the special maintenance instructions corresponding to the process equipment number from the abnormal response strategy library, and the special maintenance instructions contain equipment-specific troubleshooting steps; Combine the abnormal response strategy corresponding to the process name with the special maintenance instructions to generate the abnormal response strategy.

[0012] In some preferred embodiments, the acquisition of the multi-source process data stream composed of continuously collected component forming process data, structure assembly process data, and performance test process data with time stamp identification in the fan manufacturing process includes: In the component forming process, the temperature fluctuation sequence of the forming mold is collected by a temperature sensor, the pressure change sequence of the material extrusion process is collected by a pressure sensor, and the stroke trajectory sequence of the mold opening and closing is collected by a displacement sensor. The temperature fluctuation sequence, the pressure change sequence, and the stroke trajectory sequence are combined into component forming process data; In the structure assembly process, the butt joint offset sequence of the blade and the hub is collected by a visual recognition device, the torque fluctuation sequence of the bolt fastening is collected by a torque sensor, and the size change sequence of the assembly gap is collected by a laser range finder. The butt joint offset sequence, the torque fluctuation sequence, and the size change sequence are combined into structure assembly process data; In the performance test process, the dynamic balance vibration sequence of the fan rotor is collected by a vibration sensor, the noise frequency spectrum sequence of the bearing operation is collected by an acoustic sensor, and the power loss sequence of the motor loading is collected by a power analyzer. The dynamic balance vibration sequence, the noise frequency spectrum sequence, and the power loss sequence are combined into performance test process data; Add process time stamps synchronized with the global clock of the manufacturing execution system to the component forming process data, the structure assembly process data, and the performance test process data; Perform data alignment processing on the three types of process data after adding time stamps to generate multi-source process data streams with continuous time sequence relationships with time dimension consistency of different process data at the same manufacturing time point.

[0013] In a second aspect, the present disclosure provides an abnormality early warning system for a fan manufacturing process, comprising: a processor and a machine readable storage medium, the machine readable storage medium being connected with the processor; the machine readable storage medium, configured to store programs, instructions or codes; the processor, configured to execute the programs, instructions or codes in the machine readable storage medium to implement the abnormality early warning method for the fan manufacturing process according to any one of the first aspect.

[0014] The present disclosure provides an abnormality early warning method and system for a fan manufacturing process. Compared with the prior art, the present disclosure has the following beneficial effects: By comprehensively obtaining and analyzing multi-source process data streams, the time sequence evolution characteristics and process coupling characteristics of each process data are deeply mined, and the fan manufacturing process is monitored in all directions and multiple dimensions. The pre-deployed manufacturing abnormality diagnosis model is used for joint abnormality identification and processing of the time sequence evolution characteristics and the process coupling characteristics, which can accurately and timely identify the abnormality and generate a process abnormality identification result containing abnormality confidence, thereby providing strong data support for abnormality processing. Based on the process abnormality identification result, the target process node where the abnormality occurs and the associated process range of the abnormality impact can be accurately located, thereby not only comprehensively and accurately reflecting the potential abnormality risk in the fan manufacturing process, but also generating early warning instructions according to the target process node and the associated process range, which can be quickly pushed to the manufacturing execution system to trigger abnormality response operations, thereby realizing the timeliness and effectiveness of abnormality processing. Therefore, by comprehensively using multi-source process data stream analysis, process feature analysis and abnormality diagnosis model and other technical means, the abnormality early warning capability in the fan manufacturing process is significantly improved.

[0015] Other features and advantages of the present disclosure will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings are included to provide a further understanding of the present disclosure, and constitute a part of the specification, and are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation on the present disclosure. In the drawings: Figure 1 FIG. 1 is a flowchart of an abnormality early warning method for a fan manufacturing process according to an embodiment of the present disclosure.

[0017] Figure 2 FIG. 2 is a block diagram of an abnormality early warning system for a fan manufacturing process according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0019] The specific embodiments of the present disclosure are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present disclosure, and are not used to limit the present disclosure.

[0020] The present disclosure provides an abnormality early warning method for a fan manufacturing process, Figure 1 is a flow chart of an abnormality early warning method for a fan manufacturing process according to an embodiment. Specifically, the method comprises the following steps.

[0021] In step S11, multi-source process data streams composed of continuously collected component forming process data, structure assembly process data and performance test process data with time stamp identification in the fan manufacturing process are acquired; The component forming process data is data generated in the process of forming a component into a specific shape and size by various processing methods (such as casting, forging, stamping, etc.) in fan manufacturing. The data includes temperature, pressure, forming time, etc. The structure assembly process data is the process of assembling each component according to the design requirements to form the overall structure of the fan. The relevant data includes assembly sequence, assembly torque, assembly gap, etc. The performance test process data is the detection of various performance indicators of the manufactured fan, such as power, efficiency, noise, etc. The multi-source process data stream is a data stream composed of data from different processes (component forming, structure assembly, performance test) with time sequence, reflecting the dynamic information of the whole fan manufacturing process.

[0022] In the embodiments of the present disclosure, in the fan manufacturing process, sensors and data acquisition devices are installed at key links of each process, and data is collected uninterruptedly at a preset time interval (such as every second or every minute). The collected data is marked with a time stamp to record the specific time when the data is generated. These data from different processes are gathered in chronological order to form a multi-source process data stream. For example, in the component forming process, a temperature sensor collects mold temperature data every 1 second and records the collection time as a time stamp; in the structure assembly process, a torque sensor records the assembly torque and the corresponding time every time the bolt is tightened; in the performance test process, a power tester collects power data at a certain time interval and marks the time during the fan operation. All these data are combined in chronological order to form a multi-source process data stream of the fan manufacturing process.

[0023] In step S12, the multi-source process data stream is subjected to process feature analysis processing to obtain time sequence evolution features for representing the variation law of the process parameters in the time dimension in the process data and process coupling features for representing the parameter correlation relationship between different processes. The process feature analysis processing is to analyze and mine the multi-source process data stream by using specific algorithms and models to extract information that can reflect the essential features of the process. The time sequence evolution features are used to describe the variation law of the process parameters in the time dimension, such as the upward, downward, periodic fluctuation trend of a certain parameter over time. The process coupling features are used to reflect the correlation relationship between parameters in different processes, that is, how the parameter change of one process affects the parameters of other processes.

[0024] In the embodiments of the present disclosure, the multi-source process data stream is subjected to process feature analysis processing, which usually adopts time series analysis method and correlation analysis method. For the time sequence evolution features, the parameter data of each process is analyzed by using a time series analysis algorithm (such as moving average method, exponential smoothing method, ARIMA model, etc.) to mine the variation trend and law of the parameter over time.

[0025] For example, for the temperature parameter in the component forming process, the ARIMA model can be used to predict the variation trend of the temperature at future time and analyze the fluctuation of the temperature. For the process coupling features, a correlation analysis algorithm (such as Apriori algorithm, FP-growth algorithm, etc.) is used to analyze the correlation relationship between different process parameters. For example, whether there is a certain correlation between the assembly torque in the structure assembly process and the power in the performance test process is analyzed, and how the power changes when the assembly torque changes within a certain range.

[0026] In step S13, the time evolution features and the process coupling features are jointly subjected to abnormality identification processing by a pre-deployed manufacturing abnormality diagnosis model, to generate process abnormality identification results including abnormality probability values and abnormality possibility levels of each process data; In the joint abnormality identification processing, the time evolution features and the process coupling features are comprehensively considered, and the manufacturing abnormality diagnosis model is used for abnormality judgment, rather than separately analyzing a certain type of features.

[0027] In the embodiments of the present disclosure, the pre-deployed manufacturing abnormality diagnosis model can be constructed based on a machine learning algorithm (such as a support vector machine, a neural network, etc.) or a deep learning algorithm (such as a convolutional neural network, a recurrent neural network, etc.). When performing the joint abnormality identification processing, the extracted time evolution features and process coupling features are input into the manufacturing abnormality diagnosis model as input data. The model calculates and analyzes the input data according to the normal and abnormal data patterns learned in advance, and outputs the abnormality probability values of each process data. Then, according to a preset threshold or rule, the abnormality probability values are converted into possibility levels. For example, when the abnormality probability value is less than 0.3, it is determined as a low possibility level; when it is between 0.3 and 0.7, it is determined as a medium possibility level; and when it is greater than 0.7, it is determined as a high possibility level.

[0028] In step S14, based on the sequence relationship between the processes in the fan manufacturing process, the abnormality probability values and the possibility levels in the process abnormality identification results, a target process node where the abnormality occurs and a range of associated processes affected by the abnormality are located; The sequence relationship between the processes is that the processes in the fan manufacturing process are performed in a certain logical order, and the subsequent process can only be performed after the previous process is completed. The range of associated processes affected by the abnormality is a set of processes that may be affected by the abnormality of the target process node.

[0029] In the embodiments of the present disclosure, based on the sequence relationship between the processes in the fan manufacturing process, and in combination with the obtained process abnormality identification results (abnormality probability values and possibility levels), a process with a high abnormality probability value and a high or medium possibility level is first determined as a target process node. Then, according to the association relationship between the processes and the process flow, it is analyzed which subsequent processes or processes associated with parameters may be affected by the abnormality of the target process node, so as to determine the range of associated processes affected by the abnormality. For example, if the blade forming process is abnormal, it may affect the subsequent hub assembly process (because the size or shape of the blade does not meet the requirements, which will affect the assembly) and the whole machine performance test process (abnormal performance of the blade will cause the performance of the whole machine to decrease).

[0030] In step S15, according to the target process node and the associated process range, a pre-warning instruction including process positioning information and abnormal response strategy is generated, and the pre-warning instruction is pushed to a manufacturing execution system, and the pre-warning instruction is used to execute a visualized abnormal response operation triggered by the manufacturing execution system.

[0031] The process positioning information is used to clearly indicate the information of the target process node where the abnormality occurs and the associated process range affected by the abnormality. The abnormal response strategy is a specific response measure and processing method formulated for different abnormal situations. The manufacturing execution system (MES) is a computer system used for managing, monitoring and controlling the manufacturing process, and can coordinate multiple links such as production planning, scheduling, quality control, etc.

[0032] In the embodiments of the present disclosure, according to the determined target process node and associated process range, a pre-warning instruction containing process positioning information (specifically indicating which process is abnormal and the affected process) and abnormal response strategy (such as suspending the target process, checking the associated process parameters, adjusting the production plan, etc.) is generated. Then the pre-warning instruction is pushed to the manufacturing execution system, and after receiving the instruction, the manufacturing execution system triggers a visualized abnormal response operation according to the instruction content, such as displaying abnormal information on the operation interface, popping up a prompt box to remind the operator, automatically adjusting the production plan, etc.

[0033] The above technical solution comprehensively acquires and analyzes multi-source process data flow, deeply excavates the time sequence evolution characteristics and process coupling characteristics of each process data, and realizes omnidirectional and multidimensional monitoring of the fan manufacturing process. The pre-deployed manufacturing abnormality diagnosis model is used for joint abnormality identification and processing of the time sequence evolution characteristics and the process coupling characteristics, which can accurately and timely identify the abnormality and generate a process abnormality identification result containing abnormal confidence, thereby providing strong data support for abnormality processing. Based on the process abnormality identification result, the method can accurately locate the target process node where the abnormality occurs and the associated process range affected by the abnormality, thereby not only comprehensively and accurately reflecting the potential abnormal risk in the fan manufacturing process. Moreover, the pre-warning instruction generated according to the target process node and the associated process range can be quickly pushed to the manufacturing execution system to trigger the abnormal response operation, thereby realizing the timeliness and effectiveness of the abnormality processing. Therefore, by comprehensively using multi-source process data flow analysis, process feature analysis and abnormality diagnosis model and other technical means, the abnormality early warning capability in the fan manufacturing process is significantly improved.

[0034] In some preferred embodiments, in step S14, the target process node where the abnormality occurs and the associated process range affected by the abnormality are located based on the sequence relationship between each process in the fan manufacturing process, the abnormal probability value in the process abnormality identification result, and the possibility level, including: In step S141, the abnormal probability value in the process abnormality identification result is analyzed, and the time stamp with an abnormal probability value exceeding a preset threshold value is screened as an abnormal time point; In the embodiments of the present disclosure, the abnormal probability value corresponding to each time point is extracted from the obtained process abnormality identification result. Then, each abnormal probability value is compared with a preset threshold value. The setting of the preset threshold value needs to comprehensively consider the normal fluctuation range in the fan manufacturing process and the sensitivity requirement of abnormality detection. If the abnormal probability value of a certain time point is greater than the preset threshold value, the time point is marked as an abnormal time point.

[0035] For example, it is assumed that in the fan manufacturing process, abnormality detection is performed on a certain process (such as the temperature parameter of the blade forming process), and the preset threshold value is set to 0.7. After the temperature data collected in a period of time is analyzed, the obtained abnormal probability value sequence is [0.2, 0.3, 0.8, 0.1, 0.6, 0.9]. By comparing with the preset threshold value 0.7, it is found that the third time point (the abnormal probability value is 0.8) and the sixth time point (the abnormal probability value is 0.9) exceed the preset threshold value, and therefore the two time points are determined as abnormal time points.

[0036] In step S142, according to the time sequence evolution characteristics corresponding to the abnormal time points, the difference degrees of the time sequence evolution characteristics of the component forming process, the structure assembly process and the performance test process from the normal state characteristic template are compared, and the process with the largest difference degree is determined as the target process node with abnormality; In the embodiments of the present disclosure, for each abnormal time point, the time sequence evolution characteristics of the component forming process, the structure assembly process and the performance test process corresponding to the abnormal time point are obtained. Then, the time sequence evolution characteristics are compared with the normal state characteristic template established in advance, respectively. The comparison method can adopt various distance measurement methods, such as Euclidean distance, Mahalanobis distance, etc. The distance value calculated is the difference degree. Finally, the difference degrees of the three processes are compared, and the process with the largest difference degree is determined as the target process node with abnormality.

[0037] For example, it is assumed that at a certain abnormal time point, the time sequence evolution characteristics of the blade forming process, the hub assembly process and the whole machine performance test process are obtained, respectively. By comparison with the normal state characteristic template, it is calculated that the difference degree of the blade forming process is 0.5, the difference degree of the hub assembly process is 0.3, and the difference degree of the whole machine performance test process is 0.2. Since the difference degree of the blade forming process is the largest, the blade forming process is determined as the target process node with abnormality.

[0038] In step S143, the process coupling features corresponding to the abnormal time point are extracted, and the intensity of the process thermal coupling features and the process geometric coupling features of the target process node and other processes is analyzed, and a process with a coupling feature intensity exceeding a preset coupling threshold is screened as a candidate process; In the embodiments of the present disclosure, the process coupling features corresponding to the abnormal time point are extracted, including process thermal coupling features and process geometric coupling features. For each process coupling feature, the intensity value thereof is calculated. The calculation method can be determined according to specific coupling relationships and data analysis methods, for example, the correlation coefficient, mutual information and other methods can be used to measure the intensity of the coupling feature. Then, the coupling feature intensity of each process is compared with the preset coupling threshold. If the coupling feature intensity of a certain process exceeds the preset coupling threshold, the process is taken as a candidate process.

[0039] For example, it is assumed that in the fan manufacturing, the process thermal coupling features and the process geometric coupling features between the blade forming process and the hub assembly process are analyzed. Through calculation, the process thermal coupling feature intensity of the hub assembly process is 0.6, and the process geometric coupling feature intensity is 0.7. The preset coupling threshold is set to 0.5. Since the intensities of the two coupling features of the hub assembly process exceed the preset coupling threshold, the hub assembly process is taken as a candidate process.

[0040] In step S144, according to the sequence relationship of each process in the fan manufacturing process, if the possibility level of the previous process of the candidate process exceeds a preset level threshold, the candidate process is taken as a related process affected by the abnormality. In the embodiments of the present disclosure, according to the sequence relationship of each process in the fan manufacturing process, the previous process of each candidate process is determined. Then, the possibility level of the previous process in the process abnormality identification result is viewed. The possibility level is compared with the preset level threshold. If the possibility level of the previous process exceeds the preset level threshold, it means that the previous process has a greater possibility of abnormality and may affect the current candidate process, so the candidate process is taken as a related process affected by the abnormality.

[0041] For example, it is assumed that the fan manufacturing process flow is: blade forming → hub assembly → whole machine performance test. In step S143, the hub assembly process is determined as a candidate process. The possibility level of the blade forming process (the previous process of the hub assembly process) is viewed, and the preset level threshold is set to “medium”. If the possibility level of the blade forming process is “high”, which exceeds the preset level threshold, the hub assembly process is taken as a related process affected by the abnormality.

[0042] In step S145, manufacturing process path analysis is performed on the target process node and the associated processes, and according to the sequence relationship between the processes in the fan manufacturing process, the process range boundary of abnormal influence is determined. In the embodiments of the present disclosure, manufacturing process path analysis is performed on the target process node and the associated processes determined. According to the sequence relationship between the processes in the fan manufacturing process, the analysis is performed in the forward direction and the reverse direction along the production process from the target process node. The forward analysis considers the influence of the abnormality of the target process node on the subsequent processes; the reverse analysis considers whether the abnormality of the previous process leads to the abnormality of the target process node. Through the analysis, the propagation range of the abnormal influence in the process flow is determined, and the process range boundary is determined.

[0043] For example, it is assumed that the fan manufacturing process flow is: blade forming → hub assembly → whole machine performance test → packaging and delivery. The target process node is the blade forming process, and the associated process is the hub assembly process. Manufacturing process path analysis is performed, and the forward analysis finds that the abnormality of the blade forming process may cause the hub assembly process and the whole machine performance test process to be affected; the reverse analysis does not find that other previous processes have a direct influence on the abnormality of the blade forming process. Therefore, the process range boundary of abnormal influence is determined to be from the blade forming process to the whole machine performance test process.

[0044] In step S146, the process identifier of the target process node, the process identifier of the associated process, and the process range boundary are used to locate the associated process range of the abnormal influence.

[0045] In the embodiments of the present disclosure, the process identifier of the target process node determined, the process identifier of the associated process determined, and the process range boundary information determined are integrated. In this way, the target process in which the abnormality occurs, the associated process affected by the abnormality, and the range of the abnormal influence are clearly indicated, and the location of the associated process range of the abnormal influence is completed.

[0046] For example, it is assumed that the target process node is the blade forming process (process identifier “A”), the associated process is the hub assembly process (process identifier “B”) and the whole machine performance test process (process identifier “C”), and the process range boundary is from the blade forming process to the whole machine performance test process. The positioning result can be expressed as: the target process node is “A”, the associated process is “B” and “C”, and the abnormal influence range is the process flow from “A” to “C”.

[0047] In some preferred embodiments, in step S142, the time sequence evolution feature corresponding to the abnormal time point is extracted, the time sequence evolution features of the component forming process, the structure assembly process and the performance test process are compared with the difference degree of the normal state feature template, and the process with the largest difference degree is determined as the target process node with abnormality, comprising: In step S1421, the time sequence evolution feature of the component forming process corresponding to the abnormal time point is obtained, and the time sequence evolution feature of the component forming process includes the temperature change rate feature, the pressure change acceleration feature and the mold movement stability feature. In the embodiments of the present disclosure, at the abnormal time point, the time sequence evolution feature of the component forming process is extracted. The feature includes the temperature change rate feature, the pressure change acceleration feature and the mold movement stability feature. The temperature change rate feature reflects the change speed of the mold temperature at the abnormal time point, the pressure change acceleration feature reflects the change acceleration of the material extrusion pressure at the abnormal time point, and the mold movement stability feature describes the movement stability of the mold at the abnormal time point.

[0048] In step S1422, the time sequence evolution feature of the structure assembly process corresponding to the abnormal time point is obtained, and the time sequence evolution feature of the structure assembly process includes the assembly rhythm feature, the bolt fastening uniformity feature and the gap control accuracy feature. In the embodiments of the present disclosure, at the abnormal time point, the time sequence evolution feature of the structure assembly process is extracted. The feature includes the assembly rhythm feature, the bolt fastening uniformity feature and the gap control accuracy feature. The assembly rhythm feature reflects the rhythm law of the blade and the hub butt joint at the abnormal time point, the bolt fastening uniformity feature reflects the uniformity of the bolt fastening at the abnormal time point, and the gap control accuracy feature describes the control accuracy of the assembly gap at the abnormal time point.

[0049] In step S1423, the time sequence evolution feature of the performance test process corresponding to the abnormal time point is obtained, and the time sequence evolution feature of the performance test process includes the vibration complexity feature, the bearing wear feature and the motor efficiency feature. In the embodiments of the present disclosure, at the abnormal time point, the time sequence evolution feature of the performance test process is extracted. The feature includes the vibration complexity feature, the bearing wear feature and the motor efficiency feature. The vibration complexity feature reflects the vibration complexity of the fan rotor at the abnormal time point, the bearing wear feature reflects the wear condition of the bearing at the abnormal time point, and the motor efficiency feature describes the efficiency change of the motor at the abnormal time point.

[0050] In step S1424, a preset normal state feature template is obtained, wherein the normal state feature template includes a normal temperature change rate range, a normal pressure change acceleration range, and a normal mold motion stability range for the component molding process; a normal assembly rhythm range, a normal bolt tightening uniformity range, and a normal gap control accuracy range for the structure assembly process; and a normal vibration complexity range, a normal bearing wear range, and a normal motor efficiency range for the performance testing process. In the disclosed embodiments, the preset normal state feature templates are the feature ranges collected and analyzed during normal process operation. For component molding processes, these include the normal temperature change rate range, the normal pressure change acceleration range, and the normal mold motion stability range; for structural assembly processes, these include the normal assembly rhythm range, the normal bolt tightening uniformity range, and the normal gap control accuracy range; and for performance testing processes, these include the normal vibration complexity range, the normal bearing wear range, and the normal motor efficiency range.

[0051] In step S1425, the Euclidean distance between the temporal evolution characteristics of the component molding process and the normal state feature template is calculated as a first difference value, the Euclidean distance between the temporal evolution characteristics of the structure assembly process and the normal state feature template is calculated as a second difference value, and the Euclidean distance between the temporal evolution characteristics of the performance testing process and the normal state feature template is calculated as a third difference value; In the disclosed embodiment, the Euclidean distance is used to measure the degree of difference between two feature vectors. For the component forming process, the time-series evolution feature vector at the abnormal time point is compared with the corresponding feature vector in the normal state feature template, and the Euclidean distance between them is calculated to obtain a first difference value. Similarly, for the structural assembly process and the performance testing process, the Euclidean distance between their time-series evolution features and the normal state feature template is calculated respectively to obtain a second difference value and a third difference value. The larger the Euclidean distance, the greater the difference between the characteristics of the process at the abnormal time point and the characteristics of the normal state.

[0052] In step S1426 , the first difference value, the second difference value, and the third difference value are compared, and the process with the largest difference value is selected as the target process node where the abnormality occurs.

[0053] In this disclosed embodiment, the first, second, and third difference values ​​are compared, and the process with the largest difference value is considered the target process node where the anomaly occurred. This is because the larger the difference value, the greater the degree of deviation between the characteristics of the process at the time of the anomaly and the characteristics of the normal state, and the more likely it is to be the source of the anomaly.

[0054] In some preferred embodiments, in step S143, the process coupling features corresponding to the abnormal time point are extracted, the strength of the process thermal coupling feature and the process geometric coupling feature of the target process node and other processes are analyzed, and the processes whose coupling feature strength exceeds a preset coupling threshold are screened as candidate processes, including: In step S1431, a process thermal coupling feature value of the target process node and the component forming process is extracted from the process coupling features, and the process thermal coupling feature value represents the parameter correlation degree between the target process node and the component forming process. In the embodiments of the present disclosure, the process thermal coupling feature value of the target process node and the component forming process is extracted from the process coupling features. The process thermal coupling feature value reflects the parameter correlation degree between the target process node and the component forming process. For example, the temperature change of the component forming process may have an impact on other processes, and the strength of the impact can be measured by the process thermal coupling feature value.

[0055] In step S1432, a process geometric coupling feature value of the target process node and the structure assembly process is extracted from the process coupling features, and the process geometric coupling feature value represents the parameter correlation degree between the target process node and the structure assembly process. In the embodiments of the present disclosure, the process geometric coupling feature value of the target process node and the structure assembly process is extracted from the process coupling features. The feature value reflects the parameter correlation degree between the target process node and the structure assembly process. For example, the gap control of the structure assembly process may have an impact on other processes, and the strength of the impact can be reflected by the process geometric coupling feature value.

[0056] In step S1433, a comprehensive coupling feature value of the target process node and the performance test process is extracted from the process coupling features, and the comprehensive coupling feature value is calculated by weighted average of the process thermal coupling feature value and the process geometric coupling feature value. In the embodiments of the present disclosure, for the target process node and the performance test process, a comprehensive coupling feature value is extracted. The comprehensive coupling feature value is calculated by weighted average of the process thermal coupling feature value and the process geometric coupling feature value. Through weighted average, the coupling relationship of thermal and geometric aspects can be considered comprehensively, and the correlation degree between the target process node and the performance test process can be measured more comprehensively.

[0057] In step S1434, a preset coupling threshold is set, and the preset coupling threshold is the mean value of the normal process coupling feature value plus twice the standard deviation. In the embodiments of the present disclosure, the preset coupling threshold is a criterion for judging whether the coupling relationship between processes is strong. Generally, the preset coupling threshold is set to the mean value of the normal process coupling characteristic value plus twice the standard deviation. In this way, the fluctuation range of the coupling characteristic value under normal circumstances can be considered to some extent. When the coupling characteristic value exceeds the threshold, it indicates that the correlation degree between processes is strong and may be affected by an anomaly.

[0058] In step S1435, processes whose process thermal coupling characteristic value, process geometric coupling characteristic value and comprehensive coupling characteristic value exceed the preset coupling threshold are screened out as candidate processes.

[0059] In the embodiments of the present disclosure, the process thermal coupling characteristic value, the process geometric coupling characteristic value and the comprehensive coupling characteristic value are compared with the preset coupling threshold, and processes exceeding the preset coupling threshold are screened out. These processes are considered to be related processes that may be affected by an anomaly because their coupling relationship with the target process node where the anomaly occurs is strong, and the anomaly may be propagated to these processes through this coupling relationship.

[0060] In some preferred embodiments, in step S12, the process feature analysis processing is performed on the multi-source process data stream to obtain time sequence evolution features for representing the variation of process parameters in the time dimension and process coupling features for representing the parameter correlation relationship between different processes, including: In step S121, the multi-source process data stream is divided into analysis windows of fixed time length, and each analysis window contains continuous multiple time stamp data points; In the embodiments of the present disclosure, in order to facilitate the analysis of the multi-source process data stream, it is first divided into analysis windows of fixed time length. The length of the analysis window can be set according to the actual situation, for example, a suitable time period can be selected so that each window contains enough time stamp data points to capture the variation characteristics of the process parameters.

[0061] When dividing the analysis window, the data is selected in time sequence, and the start time and end time of each window are continuous. For example, starting from the start time of the multi-source process data stream, a time period of fixed length is selected as the first analysis window, and then the subsequent time periods are selected as other analysis windows by moving backward one by one. Each analysis window contains multiple time stamp data points, which correspond to the parameter values of different processes in the time period.

[0062] In step S122, trend extraction processing is performed on the part forming process data in each analysis window, a first-order difference of the temperature fluctuation sequence is calculated as a temperature change rate feature, a second-order difference of the pressure change sequence is calculated as a pressure change acceleration feature, a direction consistency index of the stroke trajectory sequence is calculated as a mold movement stability feature, and the temperature change rate feature, the pressure change acceleration feature, and the mold movement stability feature are combined as a time sequence evolution feature of the part forming process; In the embodiments of the present disclosure, in each analysis window, trend extraction processing is performed on the part forming process data to obtain a time sequence evolution feature thereof. For the temperature fluctuation sequence, a first-order difference thereof is calculated. The first-order difference can reflect the change rate between adjacent time points. Specifically, for each temperature value in the temperature fluctuation sequence, the difference between the previous temperature value and the next temperature value is obtained as the first-order difference. The first-order differences are arranged in time sequence to obtain the temperature change rate feature. The temperature change rate feature reflects the change speed of the mold temperature in the analysis window.

[0063] For the pressure change sequence, a second-order difference thereof is calculated. The second-order difference can reflect the change of the change rate of the sequence, that is, the acceleration. The first-order difference of the pressure change sequence is calculated first, and then the first-order difference sequence is calculated to obtain the second-order difference. The second-order differences are arranged in time sequence to obtain the pressure change acceleration feature. The pressure change acceleration feature reflects the change acceleration of the material extrusion pressure in the analysis window.

[0064] For the stroke trajectory sequence, a direction consistency index thereof is calculated. The direction consistency index can measure the stability of the movement direction of the mold during opening and closing. The specific calculation method can be determined according to the direction change of adjacent positions in the stroke trajectory sequence. For example, the angle change between adjacent positions can be calculated, and these angle changes are converted into an index value by a certain algorithm as the direction consistency index. The direction consistency index is arranged in time sequence to obtain the mold movement stability feature. The mold movement stability feature reflects the movement stability of the mold in the analysis window.

[0065] Finally, the temperature change rate feature, the pressure change acceleration feature, and the mold movement stability feature are combined to obtain the time sequence evolution feature of the part forming process. The time sequence evolution feature set describes the time variation law of the part forming process parameters in the analysis window.

[0066] In step S123, the mode recognition processing is performed on the structure assembly process data in each analysis window, the periodic fluctuation frequency of the butt joint offset sequence is calculated as the assembly rhythm feature, the standard deviation of the torque fluctuation sequence is calculated as the bolt fastening uniformity feature, and the extreme difference of the size change sequence is calculated as the gap control precision feature, and the assembly rhythm feature, the bolt fastening uniformity feature and the gap control precision feature are combined as the time sequence evolution feature of the structure assembly process. In the embodiment of the present disclosure, in each analysis window, the mode recognition processing is performed on the structure assembly process data to obtain the time sequence evolution feature thereof. For the butt joint offset sequence, the periodic fluctuation frequency thereof is calculated. The periodic fluctuation frequency can reflect the rhythm law in the butt joint process of the blade and the hub. The main frequency component in the butt joint offset sequence can be found out by spectrum analysis or the like, and the main frequency component is taken as the periodic fluctuation frequency. The periodic fluctuation frequency embodies the rhythm feature in the assembly process.

[0067] For the torque fluctuation sequence, the standard deviation thereof is calculated. The standard deviation can measure the dispersion degree of data. In the bolt fastening process, the fluctuation of the torque reflects the uniformity of the bolt fastening. The standard deviation of the torque fluctuation sequence is calculated to obtain the bolt fastening uniformity feature. The bolt fastening uniformity feature embodies the uniformity degree of the bolt fastening in the analysis window.

[0068] For the size change sequence, the extreme difference thereof is calculated. The extreme difference is the maximum value minus the minimum value in the sequence, which can reflect the fluctuation range of the sequence. In the case of the size change of the assembly gap, the extreme difference embodies the precision of the gap control. The maximum value minus the minimum value in the size change sequence is the gap control precision feature. The gap control precision feature embodies the control precision of the assembly gap in the analysis window.

[0069] Finally, the assembly rhythm feature, the bolt fastening uniformity feature and the gap control precision feature are combined to obtain the time sequence evolution feature of the structure assembly process. The time sequence evolution feature set describes the time variation law of the structure assembly process parameters in the analysis window.

[0070] In step S124, the statistical analysis processing is performed on the performance test process data in each analysis window, the energy distribution entropy of the dynamic balance vibration sequence is calculated as the vibration complexity feature, the main frequency offset of the noise frequency spectrum sequence is calculated as the bearing wear feature, and the mean value drift of the power loss sequence is calculated as the motor efficiency feature, and the vibration complexity feature, the bearing wear feature and the motor efficiency feature are combined as the time sequence evolution feature of the performance test process. In the embodiments of the present disclosure, statistical analysis is performed on the performance test procedure data in each analysis window to obtain the time sequence evolution characteristics thereof. For the dynamic balance vibration sequence, the energy distribution entropy thereof is calculated. The energy distribution entropy can measure the complexity of the vibration signal. The energy distribution entropy is obtained by converting the energy distribution into an entropy value through energy analysis on the vibration sequence. The energy distribution entropy reflects the complexity of the vibration of the fan rotor in the analysis window and serves as a vibration complexity characteristic.

[0071] For the noise spectrum sequence, the main frequency offset thereof is calculated. The main frequency offset can reflect the wear condition of the bearing during operation. With the wear of the bearing, the main frequency of the noise spectrum can be offset. By comparing the main frequencies of the noise spectrum at different time points, the offset is calculated and serves as a bearing wear characteristic. The bearing wear characteristic reflects the wear degree of the bearing in the analysis window.

[0072] For the power loss sequence, the mean value drift thereof is calculated. The mean value drift can reflect the change of the mean value of the power loss of the motor at different time points. The mean value of the power loss sequence in the analysis window is calculated, and then compared with the mean value of a reference time period to calculate the drift of the mean value, which serves as a motor efficiency characteristic. The motor efficiency characteristic reflects the efficiency change of the motor in the analysis window.

[0073] Finally, the vibration complexity characteristic, the bearing wear characteristic and the motor efficiency characteristic are combined to obtain the time sequence evolution characteristics of the performance test procedure. The time sequence evolution characteristics set describes the time variation law of the performance test procedure parameters in the analysis window.

[0074] In step S125, cross-correlation analysis is performed on the time sequence evolution characteristics of different procedures, the covariance of the temperature change rate characteristic of the component forming procedure and the bolt fastening uniformity characteristic of the structure assembly procedure is calculated as a procedure thermal coupling characteristic, the mutual information value of the gap control precision characteristic of the structure assembly procedure and the vibration complexity characteristic of the performance test procedure is calculated as a procedure geometric coupling characteristic, and the procedure thermal coupling characteristic and the procedure geometric coupling characteristic are combined as a procedure coupling characteristic.

[0075] In the embodiments of the present disclosure, after obtaining the time sequence evolution characteristics of each process, cross-correlation analysis and processing are required to be performed on the time sequence evolution characteristics of different processes to obtain process coupling characteristics. For the temperature change rate characteristics of the component forming process and the bolt fastening uniformity characteristics of the structure assembly process, the covariance thereof is calculated. The covariance can measure the degree of linear correlation between two variables. By calculating the temperature change rate characteristic sequence and the bolt fastening uniformity characteristic sequence, the covariance value obtained is the process thermal coupling characteristic. The process thermal coupling characteristic reflects the parameter correlation between the temperature change of the component forming process and the bolt fastening of the structure assembly process.

[0076] For the gap control precision characteristics of the structure assembly process and the vibration complexity characteristics of the performance test process, the mutual information value thereof is calculated. The mutual information value can measure the degree of information correlation between two variables. By calculating the gap control precision characteristic sequence and the vibration complexity characteristic sequence, the mutual information value obtained is the process geometric coupling characteristic. The process geometric coupling characteristic reflects the parameter correlation between the gap control of the structure assembly process and the vibration of the performance test process.

[0077] Finally, the process thermal coupling characteristic and the process geometric coupling characteristic are combined to obtain the process coupling characteristic. The process coupling characteristic set describes the parameter correlation between different processes.

[0078] In some preferred embodiments, in step S13, the time sequence evolution characteristics and the process coupling characteristics are jointly subjected to abnormality identification processing by the pre-deployed manufacturing abnormality diagnosis model to generate a process abnormality identification result including abnormality probability values and abnormality occurrence possibility levels of each process data, which includes the following steps. In step S131, the time sequence evolution characteristics and the process coupling characteristics are input into the feature encoding layer of the manufacturing abnormality diagnosis model and mapped into an encoded feature vector through a fully connected network. In the embodiments of the present disclosure, the feature encoding layer of the manufacturing abnormality diagnosis model is the first processing module of the model. The time sequence evolution characteristics and the process coupling characteristics are input into the feature encoding layer. The feature encoding layer adopts a fully connected network structure. The fully connected network can perform nonlinear transformation on the input features and map them into an encoded feature vector. Each neuron in the fully connected network is connected to each element of the input features. Through a series of weight and bias parameters, the input features are weighted and summed and processed by a nonlinear activation function to finally obtain the encoded feature vector. The encoded feature vector is an abstract representation of the input features and contains more feature information.

[0079] In step S132, the time series modeling processing is performed on the encoded feature vector by the time series memory module of the manufacturing abnormality diagnosis model, and time series context features including historical feature information and current feature information are extracted. In the embodiment of the present disclosure, the time series memory module of the manufacturing abnormality diagnosis model adopts a long short-term memory network (LSTM) structure. The LSTM network can effectively process time series data and can capture long-term dependencies in the data. The encoded feature vector is input into the LSTM network, and the LSTM network updates the hidden state at the current time according to the input at the current time and the hidden state at the previous time through a series of gating mechanisms.

[0080] The gating mechanisms include an input gate, a forget gate, and an output gate, which can control the inflow, retention, and output of information. By continuously updating the hidden state, the LSTM network can learn the historical feature information and the current feature information in the time series data. Finally, the hidden state of the LSTM network at each time is taken as the time series context feature. The time series context feature contains the context information of the encoded feature vector in the time dimension, which plays an important role in abnormality identification.

[0081] In step S133, the inter-process relationship modeling processing is performed on the time series context feature by the association reasoning module of the manufacturing abnormality diagnosis model, and the association features including process coupling logic are extracted. The association reasoning module adopts a graph neural network structure, in which the nodes represent process features and the edges represent process coupling features. In the embodiment of the present disclosure, the association reasoning module of the manufacturing abnormality diagnosis model adopts a graph neural network (GNN) structure. The graph neural network can model data in a graph structure. In this embodiment, the process features are taken as nodes and the process coupling features are taken as edges to construct a graph structure. The time series context features are input into the graph neural network, and the graph neural network updates and propagates the node features according to the edge information between the nodes. Through multiple iterations of updating and propagating, the graph neural network can learn the coupling logic relationship between the processes. Finally, the association features including the process coupling logic are extracted from the output of the graph neural network. The association features reflect the mutual influence and association relationship between different processes.

[0082] In step S134, the time series context features and the association features are input into the abnormality discrimination layer of the manufacturing abnormality diagnosis model, and the abnormality probability values of the process data are calculated by a logistic regression classifier. In the embodiments of the present disclosure, the abnormality discrimination layer of the manufacturing abnormality diagnosis model adopts a logistic regression classifier. The logistic regression classifier is a commonly used binary classification model, which can calculate the probability of each sample belonging to an abnormal class according to an input feature vector. The time sequence context features and the correlation features are spliced together to form a new feature vector, and then the feature vector is input into the logistic regression classifier. The logistic regression classifier will perform linear combination on the input feature vector through a series of weight and bias parameters, and then map the result of the linear combination to a probability value between 0 and 1 through a logistic function. The probability value represents the size of the possibility that the process data belongs to the abnormal class, that is, the abnormal probability value. For the data of each process, a corresponding abnormal probability value is calculated, which can reflect the possibility of whether each process exists abnormality at the current time.

[0083] In step S135, the abnormal probability value is subjected to sliding window smoothing processing to generate an abnormal probability value corresponding to the process data of each timestamp and a possibility level of abnormality.

[0084] In the embodiments of the present disclosure, in order to more accurately determine whether the process exists abnormality, the abnormal probability value needs to be subjected to sliding window smoothing processing. The sliding window smoothing processing is a commonly used time sequence smoothing method, which can reduce the fluctuation of the abnormal probability value and make the result more stable. Specifically, a sliding window of a suitable size is selected to slide on the sequence of abnormal probability values. For the abnormal probability values in the window, the average value or other statistical quantities thereof are calculated as the smoothed abnormal probability value corresponding to the center timestamp of the window. With the movement of the window, the smoothed abnormal probability value is continuously calculated to form a new sequence of abnormal probability values.

[0085] Further, according to the smoothed abnormal probability value, the possibility level of abnormality corresponding to each timestamp can be determined. For example, several probability thresholds can be set to divide the abnormal probability value into different intervals, and each interval corresponds to a possibility level of abnormality, such as low possibility, medium possibility, high possibility, etc. The abnormal probability value and the possibility level of abnormality corresponding to each timestamp are combined together to generate a process abnormality identification result. The process abnormality identification result reflects the change trend of the abnormal confidence with time.

[0086] In some preferred embodiments, in step S15, the pre-warning instruction including process positioning information and abnormal response strategy is generated according to the target process node and the associated process range, including: In step S151, the process identification information of the target process node is acquired, and the process identification information of the target process node includes the process name, the process equipment number, and the position coordinates of the process in the manufacturing flowchart. In the embodiments of the present disclosure, the process identification information of the target process node with an abnormality is obtained. The process identification information includes a process name, a process equipment number, and a position coordinate of the process in a manufacturing flowchart. The process name can specify the specific process with an abnormality, the process equipment number can be used to locate the specific equipment, and the position coordinate of the process in the manufacturing flowchart can intuitively display the position of the process in the entire manufacturing flow.

[0087] In step S152, the process identification information of the associated process range is obtained, and the process identification information of the associated process range includes an associated process name, an associated process equipment number, and a position coordinate of the associated process in a manufacturing flowchart. In the embodiments of the present disclosure, the process identification information of the associated process range affected by the abnormality is obtained. The associated process identification information includes an associated process name, an associated process equipment number, and a position coordinate of the associated process in a manufacturing flowchart. These information can help determine which processes may be affected by the abnormality and the specific positions of these processes in the manufacturing flow.

[0088] In step S153, the process identification information of the target process node and the process identification information of the associated process range are combined into process positioning information. In the embodiments of the present disclosure, the process identification information of the target process node and the process identification information of the associated process range are combined to form process positioning information. The process positioning information can comprehensively describe the position of the abnormality and the range of possible influence.

[0089] In step S154, a preset abnormality response strategy library is obtained, and the abnormality response strategy library includes device shutdown instructions, parameter adjustment instructions, and manual intervention instructions corresponding to different process nodes. In the embodiments of the present disclosure, the preset abnormality response strategy library is a pre-set strategy set, which includes processing strategies corresponding to different process nodes when an abnormality occurs. These strategies include device shutdown instructions, parameter adjustment instructions, and manual intervention instructions. For example, for a certain process, if an abnormality occurs, the device running of the process may need to be immediately stopped, or some parameters of the process may need to be adjusted, or manual inspection and repair may be arranged.

[0090] In step S155, an abnormality response strategy corresponding to the target process node is matched from the abnormality response strategy library according to the process identification information of the target process node. The process identification information of the target process node is used to uniquely determine a series of identification symbols or codes of the target process node, which can include the name, number, position code in the manufacturing process, etc. of the process. These identification information can accurately point to a specific process, so as to accurately match in the abnormal response strategy library. For example, in the fan manufacturing, the blade forming process can be identified as "FML-001", wherein "FML" represents fan manufacturing leaf, and "001" is the number of the process in the overall process.

[0091] The abnormal response strategy library can be a pre-constructed database, which stores the response strategies formulated for various possible abnormal situations in the fan manufacturing process. These strategies cover various response measures from simple parameter adjustment to complex process flow change, and are stored according to different processes, abnormal types, abnormal degrees, etc. For example, for the case of abnormal temperature rise in the blade forming process, the library can store response strategies such as reducing heating power and increasing cooling time.

[0092] In the embodiments of the present disclosure, first, the process identification information of the target process node obtained in the previous step is read out. The system will analyze these identification information to determine the specific process it represents and other related information it may contain (such as process subcategory, key parameters, etc.). For example, if the process identification information is "FML-001-H", it can be known after analysis that it is the blade forming process (FML), numbered 001, and "H" may indicate that the process is in the high-temperature forming stage. According to the analyzed process identification information, the system will construct the corresponding query conditions. These query conditions will be as accurate as possible to match the records in the abnormal response strategy library, to ensure that the most appropriate response strategy is found. For example, the query conditions can include process name = "blade forming", process number = "001", process stage = "high-temperature forming", etc.

[0093] Further, the constructed query conditions are searched and matched in the abnormal response strategy library. The database will filter out all records that meet the requirements according to these conditions, which contain response strategies for various abnormal situations that may occur in the target process node. Then, the system will further sort and filter these filtered strategies according to additional information such as abnormal type, abnormal degree, etc. (if any) to find the response strategy that best fits the current actual situation. For example, if the current target process node has an abnormal temperature rise that exceeds a certain threshold, the system will preferentially filter out the response strategy for this serious temperature abnormality.

[0094] Further, after screening, the system will return the most suitable abnormal response strategy to the relevant operator or control system. The operator can confirm the returned strategy to ensure that it meets the actual production situation and needs. If the operator considers that the strategy is not suitable, other strategies can be manually selected or the returned strategy can be modified.

[0095] For example, assuming that in the fan manufacturing process, the target process node is determined to be the blade forming process through the previous steps, and the process identification information is "FML-001". After the system reads and analyzes the identification information, it is clear that it is the blade forming process and the number is 001.

[0096] Then, according to these information, a query condition is constructed to query in the abnormal response strategy library. In the strategy library, multiple abnormal response strategies about the blade forming process are stored, for example: Strategy 1: When the temperature of the blade forming process is slightly increased between 80-100℃, reduce the heating power by 10%.

[0097] Strategy 2: When the temperature of the blade forming process exceeds 100℃, stop heating immediately and increase the cooling time by 30 minutes.

[0098] In this way, according to the actual situation of the current target process node (assuming that the temperature exceeds 100℃), strategy 2 is selected as the most suitable abnormal response strategy and returned to the operator. After the operator confirms the strategy, the control system stops heating immediately and increases the cooling time by 30 minutes according to the strategy requirements to deal with the temperature abnormality of the blade forming process.

[0099] In step S156, the process positioning information is combined with the abnormal response strategy to generate the warning instruction.

[0100] In the embodiments of the present disclosure, the process positioning information and the final abnormal response strategy are combined to generate a warning instruction. The warning instruction contains the location information of the abnormality and the corresponding processing strategy, which can provide clear guidance for the manufacturing execution system. Then, the warning instruction is sent to the control terminal of the abnormal process and the associated process. After receiving the warning instruction, the control terminal executes the corresponding abnormal response operation.

[0101] The generated warning instruction is sent to the control terminal of the abnormal process and the associated process. After receiving the warning instruction, the control terminal will execute the corresponding operation according to the abnormal response strategy in the instruction, such as device shutdown, parameter adjustment, manual intervention, etc., to handle the abnormal situation as soon as possible and reduce the impact of the abnormality on the fan manufacturing process.

[0102] In some preferred embodiments, in step S155, the corresponding abnormal response strategy is matched from the abnormal response strategy library according to the process identification information of the target process node, comprising: In step S1551, the process name in the process identification information of the target process node is parsed to determine whether the process name belongs to a component forming process, a structure assembly process or a performance test process; In the embodiments of the present disclosure, the process name in the process identification information of the target process node is parsed to determine whether the process belongs to a component forming process, a structure assembly process or a performance test process. Different process types may correspond to different abnormal response strategies, so the type of the process needs to be determined first.

[0103] In step S1552, if the process name belongs to a component forming process, the abnormal response strategy containing the mold temperature adjustment instruction, the pressure parameter correction instruction and the mold stroke calibration instruction is extracted from the abnormal response strategy library; In the embodiments of the present disclosure, if the process name belongs to a component forming process, the corresponding abnormal response strategy is extracted from the abnormal response strategy library. The strategy may contain mold temperature adjustment instructions, pressure parameter correction instructions and mold stroke calibration instructions, etc. For example, if the temperature of the component forming process is abnormal, the temperature of the mold may need to be adjusted; if the pressure parameter deviates from the normal range, the pressure parameter needs to be corrected; if the mold stroke deviates, the mold stroke needs to be calibrated.

[0104] In step S1553, if the process name belongs to a structure assembly process, the abnormal response strategy containing the assembly position correction instruction, the bolt torque compensation instruction and the gap size adjustment instruction is extracted from the abnormal response strategy library; In the embodiments of the present disclosure, if the process name belongs to a structure assembly process, the corresponding abnormal response strategy is extracted from the abnormal response strategy library. The strategy may contain assembly position correction instructions, bolt torque compensation instructions and gap size adjustment instructions, etc. For example, if the docking position of the blade and the hub deviates, the assembly position needs to be corrected; if the torque of the bolt fastening is uneven, the bolt torque needs to be compensated; if the assembly gap size does not meet the requirements, the gap size needs to be adjusted.

[0105] In step S1554, if the process name belongs to a performance test process, the abnormal response strategy containing the vibration suppression control instruction, the bearing lubrication supplement instruction and the motor power optimization instruction is extracted from the abnormal response strategy library; In the embodiments of the present disclosure, if the process name belongs to a performance test process, a corresponding abnormal response strategy is extracted from the abnormal response strategy library. The strategy can include vibration suppression control instructions, bearing lubrication supplement instructions, motor power optimization instructions, etc. For example, if the vibration of the fan rotor is too large, vibration suppression control is needed; if the bearing is worn, the bearing lubrication needs to be supplemented; if the motor power loss is abnormal, the motor power needs to be optimized.

[0106] In step S1555, the process equipment number of the target process node is obtained, and a special maintenance instruction corresponding to the process equipment number is extracted from the abnormal response strategy library. The special maintenance instruction includes equipment-specific troubleshooting steps. In the embodiments of the present disclosure, the process equipment number of the target process node is obtained, and a special maintenance instruction corresponding to the equipment number is extracted from the abnormal response strategy library. The special maintenance instruction includes equipment-specific troubleshooting steps, because different equipment can have different fault types and troubleshooting methods. Through the special maintenance instruction, the equipment can be more accurately troubleshooted and repaired.

[0107] In step S1556, the abnormal response strategy corresponding to the process name is combined with the special maintenance instruction to generate the abnormal response strategy.

[0108] In the embodiments of the present disclosure, the abnormal response strategy extracted according to the process name is combined with the special maintenance instruction to form the final abnormal response strategy. The abnormal response strategy comprehensively considers the process type and equipment characteristics, and can more effectively handle abnormal situations.

[0109] In some preferred embodiments, in step S11, the multi-source process data stream composed of continuously collected component forming process data, structure assembly process data and performance test process data with time stamp identification in the fan manufacturing process includes: In step S111, in the component forming process, a temperature fluctuation sequence of a forming mold is collected by a temperature sensor, a pressure change sequence of a material extrusion process is collected by a pressure sensor, and a stroke trajectory sequence of mold opening and closing is collected by a displacement sensor. The temperature fluctuation sequence, the pressure change sequence and the stroke trajectory sequence are combined into component forming process data. In the component forming process, the temperature of the forming mold, the pressure of the material extrusion, and the stroke trajectory of the mold opening and closing are all key process parameters, and their changes have an important influence on the forming quality of the component. The temperature sensor measures the temperature of the forming mold at certain time intervals. For example, a small time interval is set so that the sensor can capture the subtle changes in temperature. Over time, a series of temperature values is obtained, which are arranged in chronological order of measurement to form a temperature fluctuation sequence. The temperature fluctuation sequence reflects the changes in mold temperature over time.

[0110] The pressure sensor also measures the pressure during the material extrusion process at fixed time intervals. The pressure values obtained each time are recorded, and as the number of measurements increases, a pressure change sequence is formed. This sequence reflects the dynamic changes in pressure during the extrusion of the material.

[0111] The displacement sensor is used to monitor the stroke trajectory of the mold opening and closing. It continuously measures the position changes of the mold during opening and closing and records these position information in chronological order to form a stroke trajectory sequence. The stroke trajectory sequence describes the position state of the mold at different time points.

[0112] Finally, the three sequences are combined to obtain the component forming process data. The component forming process data set contains the change information of multiple key parameters during the component forming process.

[0113] In step S112, in the structure assembly process, the abutment offset sequence of the blade and the hub is collected by the visual recognition device, the torque fluctuation sequence of the bolt fastening is collected by the torque sensor, and the size change sequence of the assembly gap is collected by the laser range finder. The abutment offset sequence, the torque fluctuation sequence, and the size change sequence are combined into structure assembly process data. In the structure assembly process, the abutment of the blade and the hub, the torque of the bolt fastening, and the size of the assembly gap are all important factors affecting the assembly quality. The visual recognition device continuously observes the abutment process of the blade and the hub, and measures the offset amount at each abutment using its image recognition function. As the abutment operation proceeds, the offset amount measured each time is arranged in chronological order to form an abutment offset sequence. The abutment offset sequence reflects the time variation of the offset during the abutment of the blade and the hub.

[0114] The torque sensor monitors the torque size in real time during the bolt fastening process. It measures the torque value at fixed time intervals and records these values to form a torque fluctuation sequence. This sequence reflects the dynamic changes in torque during the fastening of the bolt, which is important for determining whether the bolt fastening meets the requirements.

[0115] A laser distance meter is used to measure the size of the assembly gap. It measures the size of the gap continuously during the assembly process and records the measurement results in chronological order to form a size change sequence. The size change sequence describes the size state of the assembly gap at different time points, which helps to evaluate the precision of the assembly.

[0116] Combining the three sequences together forms the structure assembly process data. The structure assembly process data set contains the change information of multiple key parameters in the structure assembly process.

[0117] In step S113, in the performance test process, the dynamic balance vibration sequence of the fan rotor is collected by the vibration sensor, the noise spectrum sequence of the bearing operation is collected by the acoustic sensor, and the power loss sequence of the motor load is collected by the power analyzer. The dynamic balance vibration sequence, the noise spectrum sequence and the power loss sequence are combined into performance test process data; In the performance test process, the dynamic balance vibration of the fan rotor, the noise of the bearing operation and the power loss of the motor load are important indicators for measuring the performance of the fan. The vibration sensor is installed at a suitable position of the fan rotor to monitor the vibration of the rotor during operation in real time. It measures the amplitude and frequency of the vibration at fixed time intervals and records these measurement values in chronological order to form a dynamic balance vibration sequence. The dynamic balance vibration sequence reflects the vibration state of the rotor at different time points.

[0118] The acoustic sensor monitors the noise generated by the bearing during operation. It can convert the noise signal into an electrical signal and analyze its frequency spectrum characteristics. With the operation of the bearing, the frequency spectrum information of the noise is recorded continuously to form a noise spectrum sequence. The sequence reflects the noise characteristics of the bearing at different time points, which helps to find out whether the bearing has abnormal wear and other problems.

[0119] The power analyzer is used to measure the power loss of the motor during the loading process. It monitors the input power and output power of the motor in real time and calculates the power loss value. The power loss values are recorded at certain time intervals to form a power loss sequence. The power loss sequence reflects the power consumption of the motor at different time points.

[0120] Combining the three sequences together, the performance test process data is obtained. The performance test process data set contains the change information of multiple key parameters in the performance test process.

[0121] In step S114, process time stamps synchronized with the global clock of the manufacturing execution system are added to the component forming process data, the structure assembly process data and the performance test process data; In the embodiments of the present disclosure, in order to accurately analyze the relationship of each process data in the time dimension, corresponding process timestamps need to be added to the component forming process data, the structure assembly process data and the performance test process data. The timestamp is an identifier representing the data collection time, which is synchronized with the global clock of the manufacturing execution system. The purpose of this is to ensure that the data of different processes is consistent and comparable in time.

[0122] When collecting each process data, the time of data collection will be recorded at the same time, and the time information will be added as a timestamp to the corresponding data sequence. For example, for the temperature fluctuation sequence in the component forming process data, each temperature value will correspond to a timestamp indicating when the temperature value is collected. Similarly, each sequence in the structure assembly process data and the performance test process data will also add corresponding timestamps.

[0123] In step S115, the three types of process data after adding the timestamp are subjected to data alignment processing to generate a multi-source process data stream with continuous time sequence relationship of time dimension consistency of different process data at the same manufacturing time point.

[0124] In the embodiments of the present disclosure, since the data collection time of different processes may differ, in order to be able to analyze these data in the same time dimension, the three types of process data after adding the timestamp need to be subjected to data alignment processing. The purpose of data alignment is to ensure that the data of different processes has the same time identifier at the same manufacturing time point, so that effective correlation analysis can be performed.

[0125] The specific alignment method can use interpolation or sampling, etc. For example, for the component forming process data and the structure assembly process data, if at a certain time point, the component forming process has data collection, while the structure assembly process does not collect data, the data value at this time point can be estimated according to the data of the adjacent time points before and after the structure assembly process by using the interpolation method. Through such processing, the data of different processes has consistency in the time dimension, and finally a multi-source process data stream with continuous time sequence relationship is generated. The multi-source process data stream contains the key parameter information of multiple processes at different time points in the fan manufacturing process.

[0126] The embodiments of the present disclosure also provide an abnormality early warning system for a fan manufacturing process, the system comprising: a processor and a machine readable storage medium, the machine readable storage medium being connected with the processor; the machine readable storage medium is used for storing programs, instructions or codes; The processor is configured to execute the program, the instruction or the code in the machine readable storage medium to implement the method for early warning of abnormality in a fan manufacturing process.

[0127] Figure 2 The illustrated system for early warning of abnormality in a fan manufacturing process 100 includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the system for early warning of abnormality in a fan manufacturing process 100 can also include a communication component 1004, which can be used for data interaction, such as data sending and / or data receiving, between the device 100 and other devices. It should be noted that the communication component 1004 is not limited to one in actual scheduling, and the structure of the system for early warning of abnormality in a fan manufacturing process 100 does not constitute a limitation on the embodiments of the present application.

[0128] The processor 1001 can be a CPU (Central Processing Unit, central processor), a general-purpose processor, a DSP (Digital Signal Processor, data signal processor), an ASIC (Application Specific Integrated Circuit, application specific integrated circuit), an FPGA (Field Programmable Gate Array, field programmable gate array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can realize or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure content of the present application. The processor 1001 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0129] The bus 1002 can include a path for transmitting information between the above-mentioned components. The bus 1002 can be a PCI (Peripheral Component Interconnect, peripheral component interconnect) bus or an EISA (Extended Industry Standard Architecture, extended industry standard architecture) bus, etc. The bus 1002 can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience, Figure 2 In the figure, only one thick line is used to represent, but it does not mean that there is only one bus or only one type of bus.

[0130] The memory 1003 can be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, a magnetic disk storage or other magnetic storage devices, or any other medium capable of storing program code, which can be read by a computer, without limitation.

[0131] The memory 1003 is configured to store program codes for implementing the embodiments of the present disclosure, and the processor 1001 is configured to control the execution of the program codes stored in the memory 1003. The processor 1001 is configured to execute the program codes stored in the memory 1003 to implement the steps of the foregoing method for early warning of abnormality in a fan manufacturing process.

[0132] The embodiments of the present disclosure also provide a computer readable storage medium, which stores program codes. The program codes are executed by a processor to implement the steps of the foregoing method for early warning of abnormality in a fan manufacturing process and the corresponding contents.

[0133] The preferred embodiments of the present disclosure are described in detail above with reference to the accompanying drawings, but the present disclosure is not limited to the specific details of the above-described embodiments. Within the technical concept of the present disclosure, various changes, modifications, replacements and variations can be made to the embodiments, and all these changes, modifications, replacements and variations shall be considered as the disclosed content of the present disclosure. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations. The technical scope of the present application should be determined by the scope of the claims.

[0134] In addition, it should be noted that each specific technical feature described in the above specific embodiments can be combined in any appropriate manner without contradiction, and should be considered as the disclosed content of the present disclosure. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations. The technical scope of the present application should be determined by the scope of the claims.

Claims

1. A method for anomaly early warning in a fan manufacturing process, characterized in that, The method comprises: acquiring a multi-source process data stream composed of continuously collected component forming process data, structure assembly process data and performance test process data with time stamp identification in the fan manufacturing process; performing process feature analysis processing on the multi-source process data stream to obtain time sequence evolution features for representing the change rule of process parameters in the time dimension in each process data and process coupling features for representing the parameter correlation relationship between different processes; performing joint abnormality identification processing on the time sequence evolution features and the process coupling features through a pre-deployed manufacturing abnormality diagnosis model to generate process abnormality identification results including abnormal probability values and possibility levels of abnormality of each process data; locating a target process node with abnormality and a range of associated processes affected by abnormality based on the sequence relationship of each process in the fan manufacturing process, the abnormal probability values and the possibility levels in the process abnormality identification results; generating an early warning instruction including process positioning information and abnormal response strategies according to the target process node and the range of associated processes, and pushing the early warning instruction to a manufacturing execution system, wherein the early warning instruction is used to trigger visual abnormality response operations in the manufacturing execution system.

2. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 1, characterized in that, The locating of the target process node with abnormality and the range of associated processes affected by abnormality based on the sequence relationship of each process in the fan manufacturing process, the abnormal probability values and the possibility levels in the process abnormality identification results comprises: analyzing the abnormal probability values in the process abnormality identification results, and screening time stamps with abnormal probability values exceeding a preset threshold as abnormal time points; comparing the time sequence evolution features of the component forming process, the structure assembly process and the performance test process with the difference degrees of the normal state feature templates according to the time sequence evolution features corresponding to the abnormal time points, and taking the process with the largest difference degree as the target process node with abnormality; extracting the process coupling features corresponding to the abnormal time points, analyzing the strength of the process thermal coupling features and the process geometric coupling features of the target process node and other processes, and screening processes with coupling feature strength exceeding a preset coupling threshold as candidate processes; according to the sequence relationship of each process in the fan manufacturing process, if the possibility level of the previous process of the candidate process exceeds a preset level threshold, the candidate process is taken as the associated process affected by abnormality; performing manufacturing process path analysis on the target process node and the associated processes, determining the range boundary of processes affected by abnormality according to the sequence relationship of each process in the fan manufacturing process; locating the range of associated processes affected by abnormality by the process identification of the target process node, the process identification of the associated processes and the range boundary.

3. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 2, characterized in that, The extraction of the time sequence evolution features corresponding to the abnormal time points, the comparison of the time sequence evolution features of the component forming process, the structure assembly process and the performance test process with the difference degrees of the normal state feature templates, and the determination of the process with the largest difference degree as the target process node with abnormality comprise: Obtain the time sequence evolution characteristics of the component forming process corresponding to the abnormal time point, which include temperature change rate characteristics, pressure change acceleration characteristics and mold movement stability characteristics; Obtain the time sequence evolution characteristics of the structure assembly process corresponding to the abnormal time point, which include assembly rhythm characteristics, bolt fastening uniformity characteristics and gap control accuracy characteristics; Obtain the time sequence evolution characteristics of the performance test process corresponding to the abnormal time point, which include vibration complexity characteristics, bearing wear characteristics and motor efficiency characteristics; Obtain a preset normal state characteristic template, which includes normal temperature change rate range, normal pressure change acceleration range and normal mold movement stability range of the component forming process, normal assembly rhythm range, normal bolt fastening uniformity range and normal gap control accuracy range of the structure assembly process, and normal vibration complexity range, normal bearing wear range and normal motor efficiency range of the performance test process; Calculate the Euclidean distance between the time sequence evolution characteristics of the component forming process and the normal state characteristic template as the first difference value, calculate the Euclidean distance between the time sequence evolution characteristics of the structure assembly process and the normal state characteristic template as the second difference value, and calculate the Euclidean distance between the time sequence evolution characteristics of the performance test process and the normal state characteristic template as the third difference value; Compare the first difference value, the second difference value and the third difference value, and select the process with the largest difference value as the target process node with abnormality.

4. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 2, characterized in that, The extraction of the process coupling characteristics corresponding to the abnormal time point, the analysis of the strength of the process thermal coupling characteristics and the process geometric coupling characteristics of the target process node and other processes, and the screening of processes with coupling characteristic strength exceeding a preset coupling threshold as candidate processes, include: Extracting the process thermal coupling characteristic value of the target process node and the component forming process from the process coupling characteristics, which represents the parameter correlation degree of the target process node and the component forming process; Extracting the process geometric coupling characteristic value of the target process node and the structure assembly process from the process coupling characteristics, which represents the parameter correlation degree of the target process node and the structure assembly process; Extracting the comprehensive coupling characteristic value of the target process node and the performance test process from the process coupling characteristics, which is calculated by weighted average of the process thermal coupling characteristic value and the process geometric coupling characteristic value; Setting a preset coupling threshold, which is the mean value of the normal process coupling characteristic value plus twice the standard deviation; Screening the processes whose process thermal coupling characteristic value, process geometric coupling characteristic value and comprehensive coupling characteristic value exceed the preset coupling threshold, and regarding the processes as candidate processes.

5. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 1, characterized in that, The process feature analysis processing of the multi-source process data stream, obtaining the time sequence evolution characteristics of the process parameters in the time dimension and the process coupling characteristics of the parameter correlation between different processes, includes: dividing the multi-source process data stream into fixed-time-length analysis windows, each analysis window containing continuous multiple timestamp data points; performing trend extraction processing on the part forming process data in each analysis window, calculating the first-order difference of the temperature fluctuation sequence as a temperature change rate feature, calculating the second-order difference of the pressure change sequence as a pressure change acceleration feature, calculating the direction consistency index of the stroke trajectory sequence as a mold motion stability feature, and combining the temperature change rate feature, the pressure change acceleration feature, and the mold motion stability feature as a time sequence evolution feature of the part forming process; performing pattern recognition processing on the structure assembly process data in each analysis window, calculating the periodic fluctuation frequency of the docking offset sequence as an assembly rhythm feature, calculating the standard deviation of the torque fluctuation sequence as a bolt fastening uniformity feature, calculating the extreme difference of the size change sequence as a gap control accuracy feature, and combining the assembly rhythm feature, the bolt fastening uniformity feature, and the gap control accuracy feature as a time sequence evolution feature of the structure assembly process; performing statistical analysis processing on the performance test process data in each analysis window, calculating the energy distribution entropy of the dynamic balance vibration sequence as a vibration complexity feature, calculating the main frequency offset of the noise spectrum sequence as a bearing wear feature, calculating the mean shift of the power loss sequence as a motor efficiency feature, and combining the vibration complexity feature, the bearing wear feature, and the motor efficiency feature as a time sequence evolution feature of the performance test process; performing cross-correlation analysis processing on the time sequence evolution features of different processes, calculating the covariance of the temperature change rate feature of the part forming process and the bolt fastening uniformity feature of the structure assembly process as a process thermal coupling feature, calculating the mutual information value of the gap control accuracy feature of the structure assembly process and the vibration complexity feature of the performance test process as a process geometric coupling feature, and combining the process thermal coupling feature and the process geometric coupling feature as a process coupling feature.

6. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 1, characterized in that, The joint abnormality recognition processing of the time sequence evolution features and the process coupling features through the pre-deployed manufacturing abnormality diagnosis model generates process abnormality identification results including abnormal probability values and abnormality occurrence probability levels of each process data, including: inputting the time sequence evolution features and the process coupling features into the feature encoding layer of the manufacturing abnormality diagnosis model, and mapping them into encoded feature vectors through a fully connected network; performing time series modeling processing on the encoded feature vectors through the time sequence memory module of the manufacturing abnormality diagnosis model, and extracting time sequence context features including historical feature information and current feature information; performing inter-process relationship modeling processing on the time sequence context features through the association reasoning module of the manufacturing abnormality diagnosis model, and extracting association features including process coupling logic, the association reasoning module adopts a graph neural network structure, in which nodes represent process features and edges represent process coupling features; inputting the timing context feature and the correlation feature into an anomaly discrimination layer of the manufacturing anomaly diagnosis model, and calculating an anomaly probability value of each process data by a logistic regression classifier; performing sliding window smoothing processing on the anomaly probability value to generate an anomaly probability value corresponding to each timestamp of the process data and a possibility level of occurrence of anomaly.

7. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 1, characterized in that, The pre-warning instruction including process positioning information and anomaly response strategy is generated according to the target process node and the correlation process range, and includes: obtaining process identification information of the target process node, wherein the process identification information of the target process node includes a process name, a process equipment number and a position coordinate of the process in a manufacturing flowchart; obtaining process identification information of the correlation process range, wherein the process identification information of the correlation process range includes a correlation process name, a correlation process equipment number and a position coordinate of the correlation process in the manufacturing flowchart; combining the process identification information of the target process node and the process identification information of the correlation process range into process positioning information; obtaining a preset anomaly response strategy library, wherein the anomaly response strategy library includes equipment shutdown instructions, parameter adjustment instructions and manual intervention instructions corresponding to different process nodes; matching a corresponding anomaly response strategy from the anomaly response strategy library according to the process identification information of the target process node; combining the process positioning information and the anomaly response strategy to generate the pre-warning instruction.

8. The method for early warning of abnormalities in the manufacturing process of a fan according to claim 7, characterized in that, The matching of the corresponding anomaly response strategy from the anomaly response strategy library according to the process identification information of the target process node includes: analyzing the process name in the process identification information of the target process node to determine whether the process name belongs to a component forming process, a structure assembly process or a performance test process; if the process name belongs to the component forming process, extracting an anomaly response strategy including a mold temperature adjustment instruction, a pressure parameter correction instruction and a mold stroke calibration instruction from the anomaly response strategy library; if the process name belongs to the structure assembly process, extracting an anomaly response strategy including an assembly position correction instruction, a bolt torque compensation instruction and a gap size adjustment instruction from the anomaly response strategy library; if the process name belongs to the performance test process, extracting an anomaly response strategy including a vibration suppression control instruction, a bearing lubrication supplement instruction and a motor power optimization instruction from the anomaly response strategy library; obtaining the process equipment number of the target process node, and extracting a special maintenance instruction corresponding to the process equipment number from the anomaly response strategy library, wherein the special maintenance instruction includes equipment-specific troubleshooting steps; combining the anomaly response strategy corresponding to the process name and the special maintenance instruction to generate the anomaly response strategy.

9. The method for early warning of abnormalities in a manufacturing process of a fan according to any one of claims 1 to 8, characterized in that, The multi-source process data stream composed of continuously collected component forming process data, structure assembly process data and performance test process data with timestamp identification in the fan manufacturing process includes: In the component forming process, the temperature fluctuation sequence of the forming mold is collected by the temperature sensor, the pressure change sequence of the material extrusion process is collected by the pressure sensor, and the stroke trajectory sequence of the mold opening and closing is collected by the displacement sensor. The temperature fluctuation sequence, the pressure change sequence and the stroke trajectory sequence are combined into component forming process data; In the structure assembly process, the butt joint offset sequence of the blade and the hub is collected by the visual recognition device, the torque fluctuation sequence of the bolt fastening is collected by the torque sensor, and the size change sequence of the assembly gap is collected by the laser range finder. The butt joint offset sequence, the torque fluctuation sequence and the size change sequence are combined into structure assembly process data; In the performance test process, the dynamic balance vibration sequence of the fan rotor is collected by the vibration sensor, the noise frequency spectrum sequence of the bearing operation is collected by the acoustic sensor, and the power loss sequence of the motor loading is collected by the power analyzer. The dynamic balance vibration sequence, the noise frequency spectrum sequence and the power loss sequence are combined into performance test process data; The component forming process data, the structure assembly process data and the performance test process data are added with process time stamps synchronized with the global clock of the manufacturing execution system; The three types of process data after adding time stamps are subjected to data alignment processing to generate multi-source process data flow with continuous time sequence relationship of time dimension consistency of different process data at the same manufacturing time point.

10. An anomaly early warning system for use in a wind turbine manufacturing process, characterized by, The system comprises: a processor and a machine readable storage medium, the machine readable storage medium and the processor are connected; the machine readable storage medium is used for storing programs, instructions or codes; the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the abnormal early warning method for fan manufacturing process in any one of claims 1-9.

Images