Fault monitoring and remote early warning method and system for fried instant noodle production line

Abstract

The application provides a fried instant noodle production line fault monitoring and remote early warning method and system, relates to the technical field of fault monitoring, and comprises the following steps: configuring a perception layer at a process node of a fried instant noodle production line; synchronously activating the perception layer to obtain a process node data set after the production line is operated; extracting matching features in a process chain, calling an abnormality recognition library, establishing an abnormality perceiver mapped with the process node, performing process recognition on the process node data set by using the abnormality perceiver, and establishing an independent abnormality recognition result; performing cross-process influence analysis by using a state propagation function, establishing a collaborative abnormality recognition result to compensate the independent abnormality recognition result, and reporting an abnormality early warning. The application solves the technical problems that, in the prior art, due to isolated monitoring of each process, there is a lack of linkage analysis, which leads to low fault recognition accuracy and lagging early warning response, achieves the technical effects of improving fault recognition accuracy and early warning real-time performance, and realizes collaborative fault monitoring between processes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault monitoring, in particular to a fried instant noodle production line fault monitoring and remote early warning method and system. BACKGROUND

[0002] As a kind of instant food that is deeply loved by consumers, the production process of fried instant noodles highly depends on automatic assembly line operation. In order to guarantee product quality and production efficiency, the operation state monitoring and fault warning of fried instant noodle production line become the key link in production management. The existing fried instant noodle production line fault monitoring method mostly adopts the configuration of sensors on key equipment, judges the equipment operation state by collecting temperature, vibration, current and other parameters, and triggers an alarm when the set threshold is exceeded. This kind of method mostly takes single equipment or local section as the monitoring object, lacks unified modeling and collaborative analysis of the whole production line process chain, leading to fragmented fault identification, unable to accurately reflect the conduction relationship between abnormal processes, difficult to accurately locate the fault source, and the warning response is lagging, which cannot provide effective fault handling information for production personnel in time, leading to prolonged production interruption time and damaged product quality. SUMMARY

[0003] The present application provides a fried instant noodle production line fault monitoring and remote early warning method and system, which solves the technical problems of low fault identification accuracy and lagging warning response caused by isolated process monitoring and lack of inter-process linkage analysis in the prior art, and achieves the technical effects of improving fault identification accuracy and warning real-time, and realizing inter-process collaborative fault monitoring.

[0004] In view of the above problems, on the one hand, the present application provides a fried instant noodle production line fault monitoring and remote early warning method, which comprises: after reading the process chain of fried instant noodle production line, configuring a perception layer at each process node; when the fried instant noodle production line runs, synchronously activating the perception layer to execute production data perception of the process node, and establishing a process node dataset; extracting matching features from the process chain, calling an abnormality recognition library according to the matching features, and establishing an abnormality perceiver mapped with the process node by using the abnormality recognition library; using the abnormality perceiver to perform process identification on the corresponding process node dataset, and establishing an independent abnormality identification result; after extracting the process node dataset as a process state vector, using a state propagation function to perform cross-process influence analysis on the process state vector, and establishing a collaborative abnormality identification result; using the collaborative abnormality identification result to compensate the independent abnormality identification result, and reporting an abnormality warning.

[0005] Preferably, after the process node data set is extracted as a process state vector, a state propagation function is used for cross-process impact analysis of the process state vector to establish a collaborative abnormality identification result, including: regarding each process node of the instant noodle production line as a local state field source point, a dynamic state field is established according to the time evolution of the process state vector; after the state coupling path is constructed according to the dynamic state field, the state propagation function is used for linkage strength and state impact analysis between processes to construct a three-order collaborative tensor; the three-order collaborative tensor is used for collaborative abnormality tracing to establish a collaborative abnormality identification result.

[0006] Preferably, after the state coupling path is constructed according to the dynamic state field, the state propagation function is used for linkage strength and state impact analysis between processes to construct a three-order collaborative tensor, including: using the independent abnormality identification result and the state coupling path to configure a propagation focus source point; taking the propagation focus source point as a search starting point, performing a backtracking fluctuation search of the state coupling path to establish a backtracking fluctuation search result; after the search starting point is reconstructed according to the backtracking fluctuation search result, the state propagation function is used for linkage strength and state impact analysis between processes to construct a three-order collaborative tensor.

[0007] Preferably, the matching features are extracted from the process chain, the abnormality identification library is called according to the matching features, and the abnormality perception device mapped with the process node is established by using the abnormality identification library, including: a process chain sequence is established according to the process chain, and the processes in the process chain sequence include slicing, cooking, molding, frying, cooling, and packaging; after a feature tuple is extracted from each process of the process chain sequence, a global feature of the process chain sequence is extracted; the feature tuple and the global feature are used as matching features to call the abnormality identification library to establish the abnormality perception device.

[0008] Preferably, the perception layer includes a basic sensing unit, a noise filtering unit, and a multi-modal data fusion interface, the perception parameters of the basic sensing unit include temperature, humidity, pressure, vibration, sound, visual image, flow rate, and current, the noise filtering unit is used for data preprocessing of the basic sensing unit, and the multi-modal data fusion interface is used for alignment and fusion of multiple types of perception data.

[0009] Preferably, after the independent abnormality identification result is compensated by using the collaborative abnormality identification result, an abnormality early warning is issued, including: generating an abnormality state identifier of each process node according to the independent abnormality identification result; performing abnormality state identifier tracing compensation by using the collaborative abnormality identification result; and issuing an abnormality early warning according to the abnormality state identifier after the tracing compensation.

[0010] Preferably, the reporting of the abnormality early warning comprises: obtaining a warning response of the instant noodle production line, packaging the warning response and the abnormality early warning as abnormality history data; calling the abnormality history data to identify sensitive features of the instant noodle production line, and using the sensitive features to monitor and manage the instant noodles.

[0011] Preferably, the establishing of the collaborative abnormality identification result further comprises: introducing an oil product state sensor in the perception layer to establish an oil product monitoring time sequence; using the oil product monitoring time sequence to perform oil product degradation fitting to establish a time sequence evolution abnormality; constructing a time sequence risk fluctuation path through the time sequence evolution abnormality and a state coupling path; and using the time sequence risk fluctuation path to compensate the collaborative abnormality identification result.

[0012] Preferably, the reporting of the abnormality early warning comprises: creating a warning verification window according to the abnormality early warning; determining whether there is a production line maintenance response in the warning verification window, and if there is no production line maintenance response, performing shutdown processing of the instant noodle production line.

[0013] In another aspect, the application also provides a fault monitoring and remote early warning system for an instant noodle production line, which comprises: a perception configuration module configured to configure a perception layer at each process node after reading a process chain of the instant noodle production line; a production data perception module configured to activate the perception layer to perform production data perception of the process node and establish a process node data set after the instant noodle production line runs; an abnormality perception constructor module configured to extract matching features from the process chain, call an abnormality identification library according to the matching features, and establish an abnormality perception corresponding to the process node using the abnormality identification library; an independent abnormality identification module configured to use the abnormality perception to perform process identification of the process node data set, and establish an independent abnormality identification result; a collaborative abnormality identification module configured to extract the process node data set as a process state vector, use a state propagation function to perform cross-process influence analysis of the process state vector, and establish a collaborative abnormality identification result; and an abnormality early warning module configured to compensate the independent abnormality identification result using the collaborative abnormality identification result, and report an abnormality early warning.

[0014] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0015] After reading the process chain of the fried instant noodle production line, a perception layer is configured at each process node to realize structured understanding and comprehensive perception deployment of the entire production process. When the fried instant noodle production line is running, the perception layer is activated synchronously to perform production data perception of the process node, dynamically collect real-time running data of each process node, establish a process node dataset, and form a raw dataset that can be used for anomaly identification. The matching features are extracted from the process chain, the anomaly identification library is called according to the matching features, the anomaly perceiver mapped with the process node is established by using the anomaly identification library, a feature-driven identification model matching mechanism is realized, the most suitable anomaly identification algorithm is ensured for different process nodes, and the accuracy of anomaly identification is improved. The process identification of the corresponding process node dataset is performed by using the anomaly perceiver, preliminary node-level anomaly diagnosis is performed on each process, an independent anomaly identification result is established, and the first-level anomaly identification information is formed. After the process node dataset is extracted as a process state vector, cross-process influence analysis of the process state vector is performed by using a state propagation function, a state transmission relationship between processes is modeled, downstream chain effects caused by upstream anomalies are identified, multi-node collaborative diagnosis is realized, and a collaborative anomaly identification result is established. After the collaborative anomaly identification result is used to compensate the independent anomaly identification result, an anomaly warning is reported, the accuracy and robustness of anomaly judgment are enhanced by fusing the independent and collaborative identification results, and finally reliable anomaly warning output is realized.

[0016] In summary, by configuring a perception layer at each process node of the fried instant noodle production line, the present application comprehensively collects data and establishes an independent anomaly identification result, and simultaneously considers the linkage relationship between processes to perform collaborative anomaly identification, thereby effectively improving the accuracy of fault identification and the real-time performance of the warning, discovering the anomaly of a single process node in time, accurately judging the fault chain reaction in the multi-process collaborative process, greatly reducing the fault discovery time, improving the overall operation stability of the fried instant noodle production line, and reducing the risk of production interruption and product quality damage caused by faults.

[0017] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented in accordance with the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The flowchart of the fried instant noodle production line fault monitoring and remote warning method provided by the embodiment of the present application.

[0019] Figure 2 The flowchart of establishing a collaborative anomaly identification result in the fried instant noodle production line fault monitoring and remote warning method provided by the embodiment of the present application.

[0020] Figure 3 A structural schematic diagram of a fried instant noodle production line fault monitoring and remote early warning system provided by an embodiment of the present application is shown.

[0021] Reference signs: perception configuration module 10, production data perception module 20, abnormality perception constructor module 30, independent abnormality identification module 40, collaborative abnormality identification module 50, abnormality early warning module 60. DETAILED DESCRIPTION

[0022] The embodiment of the present application provides a fried instant noodle production line fault monitoring and remote early warning method and system, solves the technical problems that the existing technology is isolated in each process monitoring, lacks inter-process linkage analysis, and thus has low fault identification accuracy and slow early warning response, and achieves the technical effects of improving fault identification accuracy and early warning real-time performance and realizing inter-process collaborative fault monitoring.

[0023] Embodiment one, as shown in the figure, the embodiment of the present application provides a fried instant noodle production line fault monitoring and remote early warning method, which comprises: Figure 1

[0024] Step S100: after reading the process chain of the fried instant noodle production line, a perception layer is configured at each process node.

[0025] Further, the perception layer comprises a basic sensing unit, a noise filtering unit and a multi-modal data fusion interface, the perception parameters of the basic sensing unit comprise temperature, humidity, pressure, vibration, sound, visual image, flow rate and current, the noise filtering unit is used for data preprocessing of the basic sensing unit, and the multi-modal data fusion interface is used for alignment and fusion of multiple types of perception data.

[0026] Specifically, the process chain refers to a logical sequence set of a plurality of continuous processes in the fried instant noodle production process, typical processes include cutting, steaming, frying, cooling and packaging. The process node is a specific processing link unit in the process chain, which has relatively independent control logic and equipment resources. The perception layer is a data acquisition and preliminary processing module deployed on each process node, which is mainly used for obtaining device state, environmental parameters and operation process data.

[0027] ​Firstly, the process configuration file or real-time state interface of the instant noodle production line is read to extract the process chain information of the whole production line, including the name, position, device type and upstream and downstream process correlation of each process. Subsequently, an independent perception layer is deployed for each process node, which includes a basic sensing unit, a noise filtering unit and a multi-modal data fusion interface. The basic sensing unit is used to collect various environmental and operating parameters related to the process, including temperature, humidity, pressure, vibration, sound, visual image, flow rate, current and other parameters. For example, in the frying process node, a temperature sensor monitors the oil temperature in real time, a humidity sensor detects the water vapor content in the environment, a pressure sensor senses the force on the device, a vibration sensor monitors the stability of the device operation, a sound sensor collects the device operation noise, a camera captures the material state during frying, a flow rate sensor monitors the flow rate of the oil, and a current sensor detects the motor operating current. The raw data collected by the basic sensing unit is input into the noise filtering unit to remove invalid data caused by environmental interference, device aging and other factors. For example, low-pass filtering algorithm is used to remove high-frequency electromagnetic noise, or wavelet transform is used to remove periodic interference signals. Subsequently, multiple types of perception data are time-synchronized and semantically aligned through the multi-modal data fusion interface to form standardized and highly consistent fusion data. For example, Kalman filtering method is used to align multiple types of perception data on the time axis, establish a unified data format and sampling rate, and form a complete process node data set to provide high-quality data support for subsequent fault monitoring and early warning.

[0028] Step S200: When the instant noodle production line is running, the perception layer is activated synchronously to perform production data perception of the process node, and the process node data set is established.

[0029] Specifically, when the instant noodle production line is put into operation, the production state is automatically detected and a synchronous activation signal is sent out to trigger the perception layer deployed on each process node to enter the working state. Each perception layer synchronously collects the production data of the corresponding process according to the preset sampling frequency and mode based on the type of sensors configured, including but not limited to temperature, humidity, pressure, vibration, sound, visual image, flow rate, current and other physical quantities. After all the collected data are processed and fused, they are stored in a structured manner according to the timestamp, process identifier, parameter type and other meta-information, and finally a plurality of process node data sets covering all process nodes are formed. These process node data sets will be used as input for the subsequent anomaly perceptron to realize single-point and multi-point fault identification and risk analysis.

[0030] Step S300: Extract matching features from the process chain, call an anomaly recognition library according to the matching features, and establish an anomaly perceptron mapped to the process node based on the anomaly recognition library.

[0031] Specifically, the matching features refer to a set of features extracted from the process chain that can identify the process characteristics and key parameters, including process type, process parameter range, pre-post process correlation, and other information. The anomaly recognition library is a database that collects various known abnormal events, typical fault cases, and normal state comparison samples of instant noodle production lines. The data types include structured sensor data, images, sounds, process logs, etc., serving as the basic data source for training the anomaly perceptron. The anomaly perceptron is an anomaly detection model for a specific process node constructed based on matching features and the anomaly recognition library, which can identify specific abnormal conditions of the process node.

[0032] The acquired process chain is analyzed to extract the matching features of each process node, including the type identification, key monitoring parameters, quality standards, and correlation features between processes. With the extracted matching features as the index condition, the most relevant historical abnormal data and normal data for each process are extracted from the anomaly recognition library to construct multiple data subsets containing sensor data change patterns, abnormal feature representations, and corresponding diagnosis results, where each data subset corresponds to a process node. Further based on these data subsets, machine learning algorithms such as support vector machines, deep neural networks, or random forests are used to train anomaly detection models, build a dedicated anomaly perceptron model for each process node, and establish a mapping relationship between the anomaly perceptron and the process node, ensuring that each process node has a corresponding anomaly monitoring mechanism. After training, the anomaly perceptron model is deployed on the perception layer or edge computing node of the corresponding process node to realize anomaly recognition of the process node data set.

[0033] Step S400: Use the anomaly perceptron to identify the process node data set, and establish an independent anomaly recognition result.

[0034] Specifically, the anomaly perceptron model established in step S300 is applied to the process node data set of the corresponding process node to perform the anomaly detection process. The anomaly perceptron of each process node runs independently, continuously receives and analyzes the process node data stream from the perception layer, extracts features and identifies patterns from the data, and calculates the similarity of the current data to known abnormal patterns or the degree of deviation from the normal state. For different types of anomalies, the anomaly perceptron outputs the recognition result of the anomaly type and the corresponding confidence score. All detected abnormal events are recorded, including anomaly type, timestamp, abnormal location identification, severity, confidence, etc., forming an anomaly recognition result for each process node. This anomaly recognition result is based only on the data analysis of a single process node and does not consider the correlation between processes, so it is referred to as an independent anomaly recognition result.

[0035] Step S500: After extracting the process node dataset as a process state vector, the cross-process influence analysis of the process state vector is performed using a state propagation function to establish a collaborative abnormality identification result.

[0036] Specifically, the process node dataset of each process node within a specific time window is converted into a process state vector, which is used to represent the current running state of the process, and a state vector field of the entire instant noodle production line is constructed. The state propagation function is called to analyze the propagation path of these vectors in space (process sequence) and time, identify the possible propagation direction and coupling relationship of the abnormality. The state propagation function is a mathematical model that describes how the abnormal state propagates and affects the process chain, which is constructed based on historical data and physical models, and can predict the influence degree and delay time of the abnormal state of one process on other processes. Finally, combined with these propagation paths, it is inferred that the abnormality of which node may be caused by the influence of other nodes, thereby generating a collaborative abnormality identification result at the system level, which is used to trace the cause of the abnormality and the risk chain.

[0037] Step S600: After compensating the independent abnormality identification result using the collaborative abnormality identification result, an abnormality warning is reported.

[0038] Specifically, after obtaining the independent abnormality identification result and the collaborative abnormality identification result, the two are logically fused to correct the limitations and one-sidedness that may exist in the independent abnormality detection. First, check the abnormal source and propagation path found in the collaborative abnormality identification result, then adjust the independent abnormality identification result in multiple aspects: for the intermediate process nodes located on the abnormal propagation path, supplement the nature of their influence by the upstream abnormality according to the propagation path; for the terminal nodes whose independent abnormality is obvious but are actually only affected, reduce the root cause abnormality rating and mark them as secondary abnormality. This trace compensation mechanism can more comprehensively and accurately reflect the actual abnormality and risk of each process node. After compensation, a more accurate comprehensive abnormality identification result is formed, and a warning report is generated, including the type, time, influence chain, location, etc. of the abnormality, and is pushed to the operation and maintenance terminal through the HMI interface or industrial communication interface.

[0039] Further, step S300 includes:

[0040] Step S310: Establish a process chain sequence according to the process chain, wherein the processes in the process chain sequence include slicing, cooking, molding, frying, cooling, and packaging.

[0041] Step S320: After extracting the feature tuple of each process in the process chain sequence, extract the global features of the process chain sequence.

[0042] Step S330: calling an exception recognition library with the feature tuple and the global feature as matching features to establish an exception sensor.

[0043] Specifically, the process chain sequence refers to an ordered set of process steps in the fried instant noodle production line, embodying the flow structure of the production process. Each node process has a time sequence dependency relationship and a process coupling relationship. According to the production line control logic or the deployment order of field devices, the key process nodes of fried instant noodle manufacturing are identified and coded, and the following typical process chain sequence is constructed: cutting- steaming- molding- frying- cooling- packaging. This sequence serves as the structural basis for subsequent multi-node feature extraction, model mapping, and abnormality propagation analysis.

[0044] The feature tuple is a set of parameters that describe the characteristics of a single process, including key process parameters, equipment characteristics, quality indicators, and other multi-dimensional information. The global feature is a set of features used to describe the overall running state and process synergy of the entire process chain sequence in the fried instant noodle production line, reflecting the mutual constraints, time sequence relationships, and system-level constraints between processes. For each process, various types of raw data are read from the perception layer and typical feature tuple examples as shown in Table 1 are extracted:

[0045] Table 1—Typical Feature Tuple Examples for Processes

[0046] Process Feature Tuple Example Cutting Surface {Sword Shaft Rotation Speed, Raw Material Conveying Rate, Cutting Surface Frequency Fluctuation Value} Cooking {Steam Temperature, Humidity Fluctuation, Pressure Stability Index} Forming {Press Wheel Rotation Speed, Bread Forming Rate, Thickness Uniformity, Forming Vibration Peak Value} Frying {Oil Temperature Distribution, Image Color Difference, Foam Density, Current Characteristics} Cooling {Cooling Wind Speed, Bread Surface Temperature Gradient, Equipment Vibration Frequency, Water Vapor Content} Packaging {Packaging Film Tensile Force, Sealing Temperature, Image Positioning Offset, Barcode Recognition Accuracy}

[0047] Based on the feature tuple, the entire process chain sequence is further analyzed, and global features are extracted across nodes, including overall running tempo, average delay of material transfer between processes, and synchronization deviation between processes. The overall running tempo refers to the unit output time after synchronization of all processes, which can measure the ability of the entire production line to complete the complete processing flow in unit time. Tempo fluctuations mean multi-point coordination disorders; the average delay of material transfer between processes refers to the average interval time from process output to the next process, reflecting the smoothness of logistics between processes. Increased delay may indicate pre-accumulation or post-blocking; the synchronization deviation between processes refers to the statistical value of the time offset of process start or stop, describing whether there is an abnormal start lag or advance in the process chain.

[0048] The feature tuples of each process and the global features extracted are used as matching features to retrieve historical records from the anomaly identification library. Similarity matching queries are performed in the anomaly identification library using these matching features. The cosine similarity, Euclidean distance, and other algorithms of feature vectors are used to calculate the similarity between the current process features and the records in the library. The most similar historical data sets to the process characteristics of each process of the current production line are retrieved. Each record in these data sets is labeled with the corresponding abnormal type label. After obtaining the historical anomaly data sets, the anomaly perceptron for a specific process is trained. The training process is as follows: the historical data sets retrieved from the anomaly identification library are cleaned, standardized, and denoised to ensure data quality. According to the process characteristics and data characteristics, suitable machine learning model architectures are selected, such as support vector machines, random forests, deep neural networks, or integrated models. The historical data sets are used for model training. During the training process, the parameters of the model are constantly adjusted to optimize the performance of the model. The accuracy, recall rate, F1 score, and other indicators are used to evaluate the performance of the model on the test data set to ensure that the model meets the application requirements. The trained anomaly perceptron model is deployed to the corresponding process node. Each process node has a dedicated anomaly perceptron that can accurately identify various anomalies that may occur in the specific process.

[0049] Further, as shown in Figure 2 Step S500 includes:

[0050] Step S510: Each process node of the fried instant noodle production line is regarded as a local state field source point. The dynamic state field is established according to the time evolution of the process state vector.

[0051] Step S520: After constructing the state coupling path according to the dynamic state field, the state propagation function is used to analyze the linkage strength and state influence between processes, and a three-order collaborative tensor is constructed.

[0052] Step S530: Collaborative anomaly tracing is performed using the three-order collaborative tensor, and a collaborative anomaly identification result is established.

[0053] Specifically, each process node (including cutting, cooking, molding, frying, cooling, and packaging) of the instant noodle production line is regarded as an independent local state field source point. The process state vector of each process node is taken as the initial input, and the sliding time window is used to model the state evolution trend to generate a time-varying process state sequence. Subsequently, a time evolution function (such as an exponential decay function or a Markov state transition model) is used to numerically model these process state sequences, thereby constructing a dynamic state field around each process node with time as the dimension. This dynamic state field is a time series state mapping formed on the basis of time evolution, and can be understood as a state trajectory field formed by the change of process state over time. The dynamic state field reflects the evolution law and spatial diffusion ability of the process state value between process nodes, providing a basis for subsequent state coupling.

[0054] The time series similarity and conduction delay between each dynamic state field are analyzed to construct a set of state coupling paths, which refer to paths where there is mutual influence between processes, such as the linkage path of "frying abnormality-cooling process temperature abnormality". Then, these state coupling paths are modeled, and the state propagation function is used to quantitatively evaluate the state propagation strength between nodes. In the construction of the state propagation function, the following technical path is adopted: first, the state vector sequence of the local state field source point i and the state vector sequence of the target node j are obtained respectively, and the time series alignment is performed through the sliding window method, then the Pearson correlation coefficient Corr is calculated to measure the trend consistency of the process node state change, the value range of the correlation coefficient Corr is [-1, 1], and the larger the value is, the stronger the influence of the local state field source point i on the target node j; at the same time, mutual information or Granger causality test is introduced as a supplementary indicator of nonlinearity or causality; the state propagation function is constructed comprehensively as follows: P(i, j, Δt) = α * Corr(Si(t), Sj(t+Δt)) + β * MI(Si, Sj), where Si(t) represents the state vector of node i at time t; Sj(t+Δt) represents the state vector of node j after a delay Δt; Sj represents the state vector of node j at time t; MI represents mutual information; α and β are adjustable weight factors for controlling the fusion degree of linear and nonlinear factors, and Δt is the conduction delay time. Finally, the quantitative results of the state propagation strength between process nodes over time are organized into a three-order collaborative tensor, which has the following structure: T(i, j, t), representing the state influence strength of process node i on process node j in time period t. This three-order tensor describes the coupling relationship in multiple processes and multiple time slices in the production process, and can provide high-dimensional structural support for subsequent collaborative anomaly identification and anomaly tracing.

[0055] Based on the third-order collaborative tensor constructed in the foregoing, the state coupling path is analyzed by tensor analysis technology. Preferably, the third-order collaborative tensor can be expressed as the product of a plurality of potential factors by using CP decomposition or Tucker decomposition method, so as to identify the dominant propagation factor and the key coupling relationship in the influence path. First, a known abnormal process node Na is selected as the abnormality tracing starting point, and the linkage intensity of all propagation paths in the third-order collaborative tensor with the node as the target is analyzed. A linkage intensity threshold is set, and upstream process nodes satisfying T(i, Na, t) greater than the linkage intensity threshold are screened, which are determined as the influence source of abnormal propagation. The abnormality confidence, state change amplitude and state propagation intensity of the upstream process nodes are weighted and summed to calculate the collaborative abnormality score, and the nodes with high scores are subjected to aggregation analysis to form a collaborative abnormality node set. The collaborative abnormality node set is used to comprehensively reflect the multi-node coupling influence relationship that may jointly cause the abnormality of the current target node, and finally form the collaborative abnormality recognition result.

[0056] Further, step S520 comprises:

[0057] Step S521: configuring a propagation focus source point by using the independent abnormality recognition result and the state coupling path.

[0058] Step S522: performing a backtracking fluctuation search of the state coupling path with the propagation focus source point as the search starting point, and establishing a backtracking fluctuation search result.

[0059] Step S523: after reconstructing the search starting point according to the backtracking fluctuation search result, performing linkage intensity and state influence analysis between processes by using the state propagation function, and constructing a third-order collaborative tensor.

[0060] Specifically, the propagation focus source point refers to the process nodes determined by the abnormality sensor as currently existing abnormality. These process nodes are used as initial focus points for state propagation and linkage analysis. According to the independent abnormality recognition result, the process nodes detected as abnormal by the abnormality sensor in the process chain are marked as propagation focus source points. In combination with the state coupling path graph constructed in the foregoing, the abnormality process nodes are used as root nodes, and state tracking analysis is prepared to be performed.

[0061] With the propagation focus source point as the search starting point, backtracking analysis is performed along the state coupling path to the upstream nodes, and the dynamic state field of each preceding process node in the relevant period is subjected to fluctuation analysis. Whether there is an abnormal inducement is judged by calculating the mean square deviation change rate, frequency drift, trend mutation and other indexes of the state. If a preceding process node is not marked as an independent abnormality, but its state fluctuation is strong and the fluctuation time is earlier than the abnormality occurrence time of the current propagation focus source point, it is regarded as a new candidate propagation focus source point and recorded as a backtracking fluctuation search result.

[0062] The identified candidate propagation focus source is combined with the original propagation focus source set as a new collaborative propagation starting point set, and the state propagation modeling is re-executed. The state propagation function is used to calculate the state influence degree of each propagation focus source on its downstream nodes in different time periods, and the influences are quantified as intensity matrices, and a three-order collaborative tensor is accumulated and constructed in a three-dimensional structure, which is used to depict the state linkage strength between process nodes at different times in the entire backtracking-propagation network, providing a comprehensive data structure for collaborative anomaly identification and tracing.

[0063] Further, the step S500 of establishing the collaborative anomaly identification result further includes:

[0064] Step S540: introducing an oil product state sensor in the perception layer to establish an oil product monitoring time sequence.

[0065] Step S550: using the oil product monitoring time sequence to perform oil product degradation fitting to establish a time sequence evolution anomaly.

[0066] Step S560: constructing a time sequence risk fluctuation path through the time sequence evolution anomaly and the state coupling path.

[0067] Step S570: compensating the collaborative anomaly identification result using the time sequence risk fluctuation path.

[0068] Specifically, the oil product monitoring time sequence is a sequence formed by the change of oil product state data collected by the oil product state sensor over time, which is used to analyze the dynamic change of the oil product state. The time sequence evolution anomaly is an abnormal change of the oil product state in the time sequence found by oil product degradation fitting, such as abnormal increase of oil temperature or too fast deepening of oil color. The time sequence risk fluctuation path is a path reflecting the abnormal fluctuation of the oil product state and its propagation risk, which is constructed by combining the time sequence evolution anomaly and the state coupling path.

[0069] An oil product state sensor is integrated in the perception layer of the frying process node of the fried instant noodle production line, which can measure key oil product state parameters such as polar compound content, color index, acid value, temperature, and use time. The above-mentioned key oil product state parameters are periodically sampled, and an oil product monitoring time sequence is constructed in time sequence. The sequence is used to describe the change trend and degradation process of the frying oil during use, providing basic data for subsequent risk analysis.

[0070] The collected oil monitoring time series is modeled using curve fitting method. For example, polynomial regression, exponential decay model or time series prediction model based on long short-term memory network is used to fit the trend of oil quality change over time or cumulative processing batch. When the prediction model output deviates from the actual oil parameter beyond a certain threshold (such as more than 5%), it is determined that there is a time evolution anomaly. This time evolution anomaly reflects the potential negative impact of oil deterioration on the overall production line operation status. For example, in a continuous frying process, the dielectric constant and polar component content of the oil are recorded every 10 minutes by an oil state sensor, and a sliding window (10 steps) is used to construct the training data, and a long short-term memory network is used to predict the future 30-minute oil state trend. If the prediction trend shows that the polar component will exceed the standard in 20 minutes, an early oil pre-deterioration warning is output; if the residual error between the actual observation value and the predicted value continues to rise beyond the set threshold, it is also considered a time evolution anomaly.

[0071] In combination with the aforementioned dynamic state field and state coupling path between process nodes, the time evolution anomaly is taken as a risk trigger source and propagated to the downstream related processes. Taking the frying process as the starting point, the influence of the state coupling path on subsequent processes such as cooling and packaging is evaluated, and a time risk fluctuation path is constructed. This time risk fluctuation path reflects the chain of abnormal effects triggered by oil deterioration.

[0072] The above-mentioned time risk fluctuation path is fused with the collaborative anomaly recognition result. On the basis of collaborative tensor analysis, the adjustment weight of the oil deterioration anomaly factor is introduced to compensate for the nodes with low existing anomaly score but significantly affected by oil time fluctuation, improve the recognition coverage and early warning foresight of complex coupled anomalies.

[0073] Further, step S600 includes:

[0074] Step S610: generating an abnormal state identifier for each process node according to the independent anomaly recognition result.

[0075] Step S620: using the collaborative anomaly recognition result for anomaly state identifier trace compensation.

[0076] Step S630: issuing an abnormality warning according to the anomaly state identifier after trace compensation.

[0077] Specifically, after the independent abnormality recognition result is compensated by the collaborative abnormality recognition result, the warning output process is entered, which mainly includes the following steps: first, according to the independent abnormality recognition result of each process node, an initial abnormality state identifier is generated, which includes whether each node has an abnormality, the probability value, type and severity of the abnormality. Then, the initial abnormality state identifier is compared and fused with the collaborative abnormality recognition result. If a certain process node is not marked as abnormal in the independent abnormality recognition result, but the upstream or downstream nodes of the process node are obviously abnormal in the collaborative abnormality path, there may be a missed detection, which is compensated by the collaborative result. In addition, if a certain process node is determined to be abnormal in the independent abnormality recognition result, but the collaborative analysis believes that it is not significantly affected, the abnormality level of the process node can be appropriately adjusted downward to reduce false positives. Finally, based on the fused abnormality state identifier, the warning level is determined and a remote abnormality warning report is generated, which is pushed to the remote operation and maintenance center through the HMI interface or the industrial communication interface.

[0078] Further, after the abnormality warning is reported, it also includes:

[0079] Step S640: creating a warning verification window according to the abnormality warning.

[0080] Step S650: determining whether there is a production line maintenance response in the warning verification window, and if there is no production line maintenance response, performing a shutdown process of the instant noodle production line.

[0081] Specifically, after the abnormal state identifier reports and pushes the abnormal early warning, an early warning verification window is automatically configured for the current process node and its upstream and downstream associated nodes. The window is a time period for monitoring whether there is manual confirmation, system response or maintenance intervention, and its length can be dynamically adjusted according to the process risk level, for example, set to 3-10 minutes. Whether there is a response event is continuously sensed within the verification window. Specifically, any of the following response signals is listened to as a criterion for maintenance response: personnel confirmation operation, such as maintenance personnel confirming that the HMI or mobile terminal has been handled; production line control system response action, such as triggering local process speed reduction, parameter adjustment or oil replacement command; sensing data returning to normal, such as abnormal indicators returning to a safe range within the verification window. If none of the above maintenance response signals is detected before the end of the early warning verification window, the abnormality is considered to be unhandled, and an automatic shutdown mechanism is triggered according to the configured strategy. This mechanism can be executed according to the following strategy: limiting the process section (such as stopping only the frying and cooling modules); whole-line linkage shutdown to prevent abnormality spread; recording abnormal event logs, forming a tracking number and sending it to the remote operation and maintenance center. For example, in the frying process node, a serious deterioration of oil state abnormality is identified, an abnormal early warning is generated immediately, and a 5-minute early warning verification window is established in the upstream and downstream processes (cooking, cooling). If no manual confirmation, oil replacement or related sensing data recovery or any response action is detected within the window period, the frying module shutdown command is automatically triggered after the end of the window period, and the production line control system is linked to stop step by step to ensure the safety of the production line.

[0082] Further, after the abnormal early warning is reported, the method further comprises:

[0083] Step S710: obtaining the early warning response of the instant noodle production line, and packaging the early warning response and the abnormal early warning as abnormal historical data.

[0084] Step S720: calling the abnormal historical data to identify the sensitive features of the instant noodle production line, and using the sensitive features for monitoring and management of instant noodles.

[0085] Specifically, after the abnormal early warning is reported, a response listening channel is established, and early warning response data including but not limited to the following types are collected: feedback records of operating personnel (such as HMI panel confirmation, mobile terminal receipt); production line control system response action, process parameter change log, fault handling log. The above early warning response and the original abnormal early warning information (such as abnormal type, process node, early warning time, early warning intensity, coordination tensor number, etc.) are packaged to generate standardized abnormal historical data entries, and recorded in the local database or remote intelligent operation and maintenance platform, forming a closed-loop traceability mechanism.

[0086] Periodically or based on a data trigger mechanism, the stored abnormal history data is analyzed, and sensitive features of the production line or specific process are identified through statistical and machine learning models (such as principal component analysis, random forest, etc.). These sensitive features are key features or links that are prone to abnormalities and are identified by analyzing abnormal history data. For example: a certain type of equipment (such as a frying temperature control valve) is more prone to failure under high temperature and humidity conditions; a certain type of oil product has a sharp increase in failure probability after running for more than X hours; and excessive material accumulation between the cooking and cooling processes easily triggers cooling abnormalities. These sensitive features are dynamically updated to the perception layer configuration and abnormality perception model parameters to achieve monitoring optimization and early perception strategy adjustment for specific scenarios, thereby improving the robustness and foresight of the instant noodle production line fault monitoring.

[0087] In summary, the instant noodle production line fault monitoring and remote early warning method provided by the embodiment of the application has the following beneficial effects:

[0088] After reading the process chain of the instant noodle production line, a perception layer is configured at each process node to achieve structured understanding and comprehensive perception deployment of the entire production process. When the instant noodle production line is running, the perception layer is activated to perform process node production data perception, dynamically collect real-time running data of each process node, establish a process node dataset, and form an original dataset that can be used for abnormality identification. The matching features are extracted from the process chain, the abnormality identification library is called according to the matching features, the abnormality perception device is established with the process node according to the abnormality identification library, the feature-driven identification model matching mechanism is realized, the most suitable abnormality identification algorithm is ensured for different process nodes, and the abnormality identification accuracy is improved. The abnormality perception device is used to identify the process of the corresponding process node dataset, and the preliminary node-level abnormality diagnosis of each process is performed to establish an independent abnormality identification result and form the first level of abnormality identification information. After the process node dataset is extracted as a process state vector, the state propagation function is used to analyze the cross-process influence of the process state vector, the state transmission relationship between processes is modeled, the downstream chain effect caused by upstream abnormalities is identified, multi-node collaborative diagnosis is realized, and a collaborative abnormality identification result is established. After the collaborative abnormality identification result is used to compensate the independent abnormality identification result, an abnormality warning is reported, the accuracy and robustness of abnormality judgment are enhanced by fusing the independent and collaborative identification results, and finally a reliable abnormality warning output is realized.

[0089] Overall, the embodiments of the present application configure a perception layer at each process node of the instant noodle production line, comprehensively collect data and establish independent abnormality identification results, and simultaneously consider the linkage relationship between processes for collaborative abnormality identification, effectively improving the accuracy of fault identification and the real-time of early warning, not only can timely discover the abnormality of a single process node, but also can accurately judge the fault chain reaction in the multi-process collaborative process, greatly reducing the fault discovery time, improving the overall operation stability of the instant noodle production line, and reducing the risk of production interruption and product quality damage caused by faults.

[0090] Embodiment two, as shown in the same inventive concept as the preceding embodiment one, the embodiments of the present application provide an instant noodle production line fault monitoring and remote early warning system, the system comprises: Figure 3

[0091] A perception configuration module 10 is configured to configure a perception layer at each process node after reading the process chain of the instant noodle production line.

[0092] A production data perception module 20 is configured to activate the perception layer to perform production data perception of the process node and establish a process node dataset when the instant noodle production line is running.

[0093] An abnormality perceiver construction module 30 is configured to extract matching features from the process chain, call an abnormality identification library according to the matching features, and establish an abnormality perceiver mapped with the process node using the abnormality identification library.

[0094] An independent abnormality identification module 40 is configured to perform process identification of the process node dataset using the abnormality perceiver, and establish an independent abnormality identification result.

[0095] A collaborative abnormality identification module 50 is configured to extract the process node dataset as a process state vector, perform cross-process impact analysis of the process state vector using a state propagation function, and establish a collaborative abnormality identification result.

[0096] An abnormality early warning module 60 is configured to compensate the independent abnormality identification result using the collaborative abnormality identification result, and report an abnormality early warning.

[0097] Further, the perception layer comprises a basic sensing unit, a noise filtering unit, and a multi-modal data fusion interface, the perception parameters of the basic sensing unit include temperature, humidity, pressure, vibration, sound, visual image, flow rate, and current, the noise filtering unit is configured to preprocess the data of the basic sensing unit, and the multi-modal data fusion interface is configured to align and fuse multiple types of perception data.

[0098] Further, the abnormality perceiver construction module 30 of the embodiments of the present application is further configured to perform the following steps: ​

[0099] According to the process chain sequence, each process in the process chain sequence includes slicing, cooking, forming, frying, cooling, and packaging; after extracting a feature tuple for each process of the process chain sequence, a global feature of the process chain sequence is extracted; the feature tuple and the global feature are called as matching features to call an abnormality recognition library to establish an abnormality sensor.

[0100] Further, the embodiment of the present application cooperates the abnormality recognition module 50 to further perform the following steps:

[0101] Each process node of the fried instant noodle production line is regarded as a local state field source point, a dynamic state field is established according to time evolution of the process state vector, a state coupling path is constructed according to the dynamic state field, linkage strength and state influence between processes are analyzed by using the state propagation function, a three-order cooperation tensor is constructed, and the three-order cooperation tensor is used for cooperative abnormality tracing to establish a cooperative abnormality recognition result.

[0102] Further, the embodiment of the present application cooperates the abnormality recognition module 50 to further perform the following steps:

[0103] The independent abnormality recognition result and the state coupling path are used to configure a propagation attention source point, the propagation attention source point is taken as a search starting point, a backtracking fluctuation search of the state coupling path is performed to establish a backtracking fluctuation search result, the search starting point is reconstructed according to the backtracking fluctuation search result, linkage strength and state influence between processes are analyzed by using the state propagation function, and a three-order cooperation tensor is constructed.

[0104] Further, the embodiment of the present application cooperates the abnormality recognition module 50 to further perform the following steps:

[0105] An oil product state sensor is introduced into the perception layer to establish an oil product monitoring time sequence, the oil product monitoring time sequence is used for oil product degradation fitting to establish a time sequence evolution abnormality, a time sequence risk fluctuation path is constructed through the time sequence evolution abnormality and the state coupling path, and the time sequence risk fluctuation path is used for compensation of the cooperative abnormality recognition result.

[0106] Further, the embodiment of the present application cooperates the abnormality recognition module 60 to further perform the following steps:

[0107] An abnormality state identifier of each process node is generated according to the independent abnormality recognition result, the cooperative abnormality recognition result is used for trace compensation of the abnormality state identifier, and an abnormality early warning is reported according to the trace compensated abnormality state identifier.

[0108] Further, the embodiment of the present application cooperates the abnormality recognition module 60 to further perform the following steps:

[0109] According to the abnormal early warning, a warning verification window is created; whether there is a production line maintenance response in the warning verification window is judged, if there is no production line maintenance response, a shutdown processing of the instant noodle production line is executed.

[0110] Further, the system of the embodiment of the present application is also used to execute the following steps:

[0111] The early warning response of the instant noodle production line is acquired, the early warning response and the abnormal early warning are packaged and recorded as abnormal history data; the sensitive features of the instant noodle production line are identified by calling the abnormal history data, and the instant noodle is monitored and managed by using the sensitive features.

[0112] Through the foregoing detailed description of the instant noodle production line fault monitoring and remote early warning method, the person skilled in the art can clearly know the instant noodle production line fault monitoring and remote early warning system in the embodiment. For the system disclosed in the second embodiment, since it corresponds to the method disclosed in the first embodiment, it has corresponding functional modules and beneficial effects, and the related parts can be referred to the method part description.

[0113] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fault monitoring and remote early warning in a fried instant noodle production line, characterized in that, The method includes: After reading the process chain of the fried instant noodle production line, a perception layer is configured at each process node; Once the instant noodle production line is running, the perception layer is activated simultaneously to perceive the production data of the process nodes and establish a process node dataset. Extract matching features from the process chain, call the anomaly identification library based on the matching features, and establish an anomaly sensor that maps to the process nodes using the anomaly identification library; The anomaly sensor is used to identify the process of the corresponding process node dataset, and an independent anomaly identification result is established. After extracting the process node dataset into process state vectors, the cross-process influence analysis of the process state vectors is performed using the state propagation function to establish collaborative anomaly identification results. After compensating the independent anomaly identification results with the collaborative anomaly identification results, an anomaly warning is issued; After extracting the process node dataset into process state vectors, the cross-process impact analysis of the process state vectors is performed using the state propagation function to establish collaborative anomaly identification results, including: Each process node in the fried instant noodle production line is regarded as a local state field source point, and a dynamic state field is established by performing time evolution based on the process state vector. After constructing the state coupling path based on the dynamic state field, the state propagation function is used to analyze the linkage strength and state influence between processes, and a third-order collaborative tensor is constructed. The third-order collaborative tensor is used to trace collaborative anomalies and establish collaborative anomaly identification results. After constructing the state coupling path based on the dynamic state field, the state propagation function is used to analyze the linkage strength and state influence between processes, and a third-order cooperative tensor is constructed, including: The independent anomaly identification results and the state coupling path are used to configure the propagation of the source of interest. Starting from the source of the propagation concern, a backtracking fluctuation search of the state coupling path is performed to establish the backtracking fluctuation search results; After reconstructing the search starting point based on the backtracking fluctuation search results, the state propagation function is used to analyze the linkage strength and state influence between processes, and a third-order collaborative tensor is constructed.

2. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, The step of extracting matching features from the process chain, calling an anomaly detection library based on the matching features, and establishing an anomaly sensor mapped to the process nodes using the anomaly detection library includes: A process chain sequence is established based on the process chain, wherein the processes in the process chain sequence include cutting, steaming, shaping, frying, cooling, and packaging; After extracting feature tuples for each process in the process chain sequence, the global features of the process chain sequence are extracted. The feature tuples and global features are used as matching features to call the anomaly detection library to establish an anomaly perceiver.

3. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, The perception layer includes a basic sensing unit, a noise filtering unit, and a multimodal data fusion interface. The sensing parameters of the basic sensing unit include temperature, humidity, pressure, vibration, sound, visual image, flow rate, and current. The noise filtering unit is used for data preprocessing of the basic sensing unit, and the multimodal data fusion interface is used for aligning and fusing multiple types of sensing data.

4. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, The step of compensating the independent anomaly identification result with the collaborative anomaly identification result and then reporting an anomaly warning includes: Anomaly status identifiers for each process node are generated based on the independent anomaly identification results; The collaborative anomaly identification results are used for source tracing and compensation of anomaly status identifiers; An anomaly warning is issued based on the anomaly status identifier after source tracing and compensation.

5. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, After the abnormal warning is issued, the following is included: Obtain the early warning response from the fried instant noodle production line, and package and record the early warning response and the abnormal early warning as abnormal historical data; The abnormal historical data is used to identify sensitive features of the fried instant noodle production line, and these sensitive features are used for monitoring and management of the fried instant noodles.

6. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, The establishment of collaborative anomaly identification results also includes: Introduce oil condition sensors into the sensing layer to establish an oil monitoring time series; The oil product monitoring time series was used to fit oil product deterioration and establish time-series evolution anomalies. A temporal risk fluctuation path is constructed by the aforementioned temporal evolution anomaly and state coupling path; The time-series risk fluctuation path is used to compensate for the collaborative anomaly identification results.

7. The method for fault monitoring and remote early warning of a fried instant noodle production line as described in claim 1, characterized in that, After the abnormal warning is issued, the following is included: Create an alert verification window based on the aforementioned anomaly alert; Determine whether there is a production line maintenance response in the warning verification window. If there is no production line maintenance response, then execute the shutdown process for the instant noodle frying production line.

8. A fault monitoring and remote early warning system for a fried instant noodle production line, characterized in that, The system is used to execute the method for fault monitoring and remote early warning of the instant noodle production line according to any one of claims 1-7, including: The perception configuration module is used to configure a perception layer at each process node after reading the process chain of the fried instant noodle production line. The production data sensing module is used to synchronously activate the sensing layer to sense the production data of the process nodes after the instant noodle production line is running, and to establish a process node dataset. An anomaly sensor construction module is used to extract matching features from the process chain, call the anomaly identification library based on the matching features, and establish an anomaly sensor that maps to the process nodes using the anomaly identification library. An independent anomaly identification module is used to identify the corresponding process node dataset using the anomaly sensor and establish an independent anomaly identification result. The collaborative anomaly identification module is used to extract the process node dataset into process state vectors, and then use the state propagation function to perform cross-process influence analysis of the process state vectors to establish collaborative anomaly identification results. An anomaly warning module is used to compensate the independent anomaly recognition result with the collaborative anomaly recognition result and then report an anomaly warning.

Images