Intelligent monitoring method and system for milk powder production line based on artificial intelligence

Abstract

The present application provides an intelligent monitoring method and system for a milk powder production line based on artificial intelligence, which obtains the target production line equipment status data from the milk powder production line, identifies the target abnormal node reflecting the target abnormal monitoring project, and generates a corresponding first state path vector. Then, based on the production line equipment status data of the current and next equipment, the abnormal monitoring project status data reflecting the abnormal monitoring project is obtained, and the corresponding second state path vector is generated. After being fused with the first state path vector and output as a target state path vector, the abnormal indicator of the abnormal monitoring project is determined, thereby obtaining the linkage characteristic parameters of the abnormal indicator of the abnormal monitoring project in the next equipment status data, so as to determine the abnormal monitoring project node reflecting the abnormal state partition of the target abnormal monitoring project in the next equipment status data, and then predict the possible abnormalities.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based intelligent monitoring method and system for a milk powder production line. Background Art

[0002] In modern milk powder production lines, the status monitoring of milk powder production equipment is crucial. Once the production equipment is abnormal, it may not only affect production efficiency, but also reduce product quality and even cause safety accidents. Therefore, how to accurately monitor the status of equipment in real time and predict possible abnormalities has become an important issue in production management.

[0003] Traditional equipment status monitoring methods mainly rely on manual inspection and regular maintenance, which is not only labor-intensive but also often unable to detect equipment anomalies in real time. Although some modern production lines have begun to use various sensors to automatically monitor equipment status, these methods can usually only monitor a single or a few equipment parameters, making it difficult to fully and accurately reflect the working status of the equipment. Summary of the invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the present application is to provide an intelligent monitoring method and system for a milk powder production line based on artificial intelligence.

[0005] In a first aspect, the present application provides an intelligent monitoring method for a milk powder production line based on artificial intelligence, which is applied to an intelligent monitoring system of a production line, and the method comprises:

[0006] Obtaining a first state path vector corresponding to a target abnormal node in the target production line equipment status data in the milk powder production line; the target production line equipment status data belongs to a target production line monitoring event, the target production line monitoring event includes a target abnormal monitoring item, and the target abnormal node reflects the abnormal state partition of the target abnormal monitoring item in the target production line equipment status data;

[0007] Based on the first abnormal monitoring item node corresponding to the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, the abnormal monitoring item status data is obtained in the x+1-th production line equipment status data of the target production line monitoring event, and the second state path vector corresponding to the abnormal monitoring item status data is obtained; the state generation timing position of the x-th production line equipment status data in the target production line monitoring event is not earlier than the state generation timing position of the target production line equipment status data in the target production line monitoring event, and the state generation timing position of the x-th production line equipment status data in the target production line monitoring event is prior to the state generation timing position of the x+1-th production line equipment status data in the target production line monitoring event, and x is a positive integer;

[0008] The second state path vector and the first state path vector are merged to output a corresponding target state path vector, an abnormal indicator of the abnormal monitoring item corresponding to the abnormal monitoring item state data is determined based on the target state path vector, and a linkage characteristic parameter of the first abnormal monitoring item node in the equipment state data of the x+1th production line is obtained based on the target state path vector;

[0009] Based on the abnormal indicator of the abnormal monitoring item, the first abnormal monitoring item node and the linkage characteristic parameter, a second abnormal monitoring item node reflecting the abnormal status partition of the target abnormal monitoring item is determined in the x+1th production line equipment status data.

[0010] In a possible implementation of the first aspect, the acquiring a first state path vector corresponding to a target abnormal node in the target production line equipment state data in the milk powder production line includes:

[0011] Acquire the target production line monitoring event, split the target production line monitoring event, generate a target production line equipment status data set, and acquire the first production line equipment status data in the target production line equipment status data set;

[0012] If it is parsed that the first production line equipment status data contains the target abnormal monitoring item, the first production line equipment status data is output as the target production line equipment status data in the target production line monitoring event, and the abnormal status partition of the target abnormal monitoring item in the target production line equipment status data is output as the target abnormal node;

[0013] The state data sequence associated with the target abnormal node in the target production line equipment state data is generated as a target abnormal node state sequence, and a first state path vector corresponding to the target abnormal node state sequence is obtained.

[0014] In a possible implementation of the first aspect, the method further includes:

[0015] Obtain K candidate abnormal nodes from the first production line equipment status data, obtain abnormal node knowledge vectors corresponding to each of the K candidate abnormal nodes, estimate the abnormal node knowledge vectors, and generate estimated category information corresponding to each of the K candidate abnormal nodes; K is a positive integer;

[0016] When the estimated category information corresponding to each of the K candidate abnormal nodes contains category information associated with the target abnormal monitoring item, it is determined that the first production line equipment status data contains the target abnormal monitoring item.

[0017] In a possible implementation manner of the first aspect, outputting the abnormal state partition of the target abnormal monitoring item in the target production line equipment state data as a target abnormal node includes:

[0018] Among the K candidate abnormal nodes, outputting the candidate abnormal nodes corresponding to the category information associated with the target abnormal monitoring item as a temporary abnormal node sequence;

[0019] Based on the Jaccard similarity coefficients between the candidate abnormal nodes in the temporary abnormal node sequence, the candidate abnormal nodes in the temporary abnormal node sequence are integrated to generate abnormal nodes to be updated;

[0020] The abnormal node to be updated is optimized in terms of its boundary, and a target abnormal node including the target abnormal monitoring item is obtained in the target production line equipment status data.

[0021] In a possible implementation of the first aspect, the first abnormal monitoring item node corresponding to the target abnormal monitoring item in the xth production line equipment status data of the target production line monitoring event, obtaining abnormal monitoring item status data in the x+1th production line equipment status data of the target production line monitoring event, and obtaining a second state path vector corresponding to the abnormal monitoring item status data include:

[0022] Obtaining a first abnormal monitoring item node of the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, and determining a first selected abnormal node in the x+1-th production line equipment status data of the target production line monitoring event based on a priori attention feature vector corresponding to the first abnormal monitoring item node;

[0023] According to the first selection of abnormal nodes, performing abnormal node tracking in the x+1th production line equipment status data, generating at least one tracked abnormal node, and obtaining an abnormal feedback value corresponding to each of the at least one tracked abnormal node based on an abnormal feedback extraction model;

[0024] Outputting the tracking abnormal node corresponding to the maximum abnormal feedback value as the second selected abnormal node, and generating the state data sequence associated with the second selected abnormal node in the x+1th production line equipment state data as the abnormal monitoring item state data;

[0025] The abnormal monitoring item status data is loaded into a recursive neural network in a target abnormal tracking network model, and a second state path vector corresponding to the abnormal monitoring item status data is output according to the recursive neural network in the target abnormal tracking network model.

[0026] In a possible implementation manner of the first aspect, fusing the second state path vector and the first state path vector to output a corresponding target state path vector includes:

[0027] Loading the second state path vector to the first encoder, and outputting first encoding feature data according to the first encoder; the encoding feature dimension of the first encoding feature data is consistent with the encoding feature dimension of the second state path vector;

[0028] Loading the first state path vector to a second encoder, and outputting second encoding feature data according to the second encoder; the encoding feature dimension of the second encoding feature data is different from the encoding feature dimension of the first state path vector;

[0029] The first encoded feature data and the second encoded feature data are blended to generate the target state path vector.

[0030] In a possible implementation of the first aspect, determining the abnormal monitoring item abnormality indicator corresponding to the abnormal monitoring item state data based on the target state path vector, and acquiring the linkage characteristic parameter of the first abnormal monitoring item node in the x+1th production line equipment state data based on the target state path vector, includes:

[0031] The target state path vector is estimated to generate an abnormal monitoring item abnormality indicator of the abnormal monitoring item state data for the target abnormal monitoring item;

[0032] Based on the target state path vector, the reference abnormal node determined by the first abnormal monitoring project node in the x+1th production line equipment status data is analyzed to generate linkage characteristic parameters of the first abnormal monitoring project node in the x+1th production line equipment status data.

[0033] In a possible implementation of the first aspect, the linkage characteristic parameter includes a first linkage characteristic value and a second linkage characteristic value, and the first linkage characteristic value and the second linkage characteristic value reflect the abnormal state partition linkage state of the link node on the same production line abnormal link of the first abnormal monitoring item node in the x+1th production line equipment state data;

[0034] The determining, based on the abnormal indicator of the abnormal monitoring item, the first abnormal monitoring item node and the linkage characteristic parameter, in the x+1th production line equipment status data, a second abnormal monitoring item node reflecting the abnormal status partition of the target abnormal monitoring item comprises:

[0035] If the abnormality indicator of the abnormal monitoring item is greater than a set value, determining that the abnormal monitoring item status data includes the target abnormal monitoring item;

[0036] Based on the prior attention feature vector corresponding to the first abnormal monitoring item node, the first linkage feature value, and the second linkage feature value, abnormal node location data of the target abnormal monitoring item in the x+1th production line equipment status data is generated, and according to the abnormal node location data, the second abnormal monitoring item node is determined in the x+1th production line equipment status data;

[0037] If the abnormal indicator of the abnormal monitoring item is not greater than the set value, it is determined that the target abnormal monitoring item is not included in the abnormal monitoring item status data, and the x+2th production line equipment status data in the target production line monitoring event is obtained; the status generation timing position of the x+2th production line equipment status data in the target production line monitoring event is later than the status generation timing position of the x+1th production line equipment status data in the target production line monitoring event;

[0038] If it is parsed that the target abnormal monitoring item is included in the x+2th production line equipment status data, the target production line equipment status data in the target production line monitoring event is adjusted to the x+2th production line equipment status data, and the target abnormal node in the target production line equipment status data is adjusted to the abnormal status partition of the target abnormal monitoring item in the x+2th production line equipment status data.

[0039] In a possible implementation of the first aspect, the method further includes:

[0040] Obtaining an example production line monitoring event including an example abnormal monitoring item, obtaining example reference production line equipment status data and an example production line equipment status data sequence in the example production line monitoring event; the state generation time sequence position of the example reference production line equipment status data in the example production line monitoring event precedes the state generation time sequence position of each example production line equipment status data in the example production line equipment status data sequence in the example production line monitoring event;

[0041] Based on the training annotation data associated with the example reference production line equipment status data, determine the example abnormal node corresponding to the example abnormal monitoring item in the example reference production line equipment status data, and obtain the first example state path vector of the example abnormal node in the example reference production line equipment status data and the second example state path vector of the example abnormal node in each example production line equipment status data through a recursive neural network in a basic abnormality tracking network model;

[0042] The second example state path vector and the first example state path vector are merged to output a corresponding target example state path vector, and an abnormality location learning result corresponding to each example production line equipment state data is generated according to the abnormality location unit in the basic abnormality tracking network model;

[0043] Based on the training annotation data associated with the example reference production line equipment status data, the training annotation data respectively associated with each example production line equipment status data, and the abnormality localization learning result, the weight information of the basic abnormality tracking network model is updated, and the basic abnormality tracking network model containing the updated weight information is output as a target abnormality tracking network model; the target abnormality tracking network model is used to determine the abnormal state partition of the target abnormal monitoring item in the production line equipment status data contained in the target production line monitoring event;

[0044] The abnormality location unit includes a classification layer and a prediction layer, the abnormality location learning result includes a classification probability distribution generated by the classification layer and a linkage feature estimation value generated by the prediction layer, and the linkage feature estimation value reflects the abnormal state partition linkage state of the link node on the same production line abnormal link of the example abnormal node in the equipment state data of each example production line;

[0045] The method of updating the weight information of the basic abnormality tracking network model based on the training annotation data associated with the example reference production line equipment status data, the training annotation data respectively associated with each example production line equipment status data, and the abnormality location learning result, and outputting the basic abnormality tracking network model including the updated weight information as the target abnormality tracking network model, including:

[0046] Determine the target linkage feature value of the example abnormal node in each example production line equipment status data based on the annotated abnormal state partition data in the training annotated data associated with the example reference production line equipment status data and the annotated abnormal state partition data in the training annotated data respectively associated with each example production line equipment status data;

[0047] Determining a prediction cost value corresponding to the prediction layer according to a difference between the target linkage feature value and the linkage feature estimation value;

[0048] Determine the classification cost value corresponding to the classification layer according to the classification probability distribution and the labeled category information in the training labeled data respectively associated with each example production line equipment status data;

[0049] Based on the prediction cost value and the classification cost value, the weight information of the basic anomaly tracing network model is updated, and the basic anomaly tracing network model including the updated weight information is output as a target anomaly tracing network model.

[0050] In a second aspect, an embodiment of the present application also provides an intelligent monitoring system for a production line, the intelligent monitoring system for the production line comprising a processor and a machine-readable storage medium, the machine-readable storage medium storing a computer program, the computer program being loaded and executed in conjunction with the processor to implement the artificial intelligence-based intelligent monitoring method for a milk powder production line of the first aspect above.

[0051] By adopting the technical scheme of any of the above aspects, the target production line equipment status data is obtained from the milk powder production line, and the target abnormal node reflecting the target abnormal monitoring item is identified to generate the corresponding first state path vector. Then, based on the production line equipment status data of the current and next equipment (the xth and x+1th equipment), the abnormal monitoring item status data reflecting the abnormal monitoring item is obtained to generate the corresponding second state path vector. Next, the first and second state path vectors are fused and output as the target state path vector, and the abnormal indicators of the abnormal monitoring items corresponding to the abnormal monitoring item status data are determined based on the target state path vector, and the linkage characteristic parameters of the abnormality in the next equipment state data are obtained. Finally, according to the abnormal indicators of the abnormal monitoring items, the abnormal monitoring item nodes and the linkage characteristic parameters, the abnormal monitoring item nodes reflecting the abnormal state partition of the target abnormal monitoring item are determined in the next equipment state data. Thus, by monitoring the target production line equipment status data in the milk powder production line and predicting possible abnormalities, the problem can be handled in advance before it affects the production, avoiding the production stagnation caused by equipment failure, and improving the operation efficiency of the production line and the production quality of milk powder products. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained by combining these drawings without paying creative work.

[0053] Figure 1 A schematic diagram of a process flow of an intelligent monitoring method for a milk powder production line based on artificial intelligence provided in an embodiment of the present application;

[0054] Figure 2 A block diagram showing the functional structure of an intelligent monitoring system for a production line for implementing the above-mentioned artificial intelligence-based intelligent monitoring method for a milk powder production line provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The following description is intended to enable one of ordinary skill in the art to implement and combine the present application, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present application may be applied to other embodiments and application scenarios without departing from the principles and scope of the present application. Therefore, the present application is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0056] See also Figure 1 As shown, the present application provides an intelligent monitoring method for a milk powder production line based on artificial intelligence, comprising the following steps.

[0057] Step S110, obtaining a first state path vector corresponding to a target abnormal node in the target production line equipment state data in the milk powder production line.

[0058] In this embodiment, the target production line equipment status data belongs to a target production line monitoring event, the target production line monitoring event includes a target abnormal monitoring item, and the target abnormal node reflects the abnormal status partition of the target abnormal monitoring item in the target production line equipment status data.

[0059] For example, suppose that a milk powder production line is currently running, including steps such as collecting raw materials, processing, filling, and packaging. Each step has corresponding production line equipment, and they are all monitored to ensure the smooth progress of production. Now, a certain abnormality in the processing link (target production line equipment) is noticed, such as too high temperature. This high temperature is an abnormal node, which is displayed as an abnormal state partition in the equipment status data of the entire production line. On this basis, the state path vector corresponding to this abnormal node can be obtained. This state path vector may contain the equipment status information before and after the abnormality occurs. For example, the state path vector may include various parameters of the equipment, such as temperature, pressure, speed, etc., and reflect the changes of these parameters over time.

[0060] For example, in a milk powder production line, a processing equipment is being focused on. The main task of this equipment is to heat the raw materials to a certain temperature for sterilization. Therefore, it is necessary to monitor the temperature change of this process to ensure that it is always within the appropriate range. If the temperature of this processing equipment is regarded as a state, then the state path is the temperature change of this equipment over a period of time. For example, at 8 o'clock in the morning, the equipment starts working and the temperature gradually rises; at 9 o'clock, it reaches the set sterilization temperature; then it maintains this temperature until 12 o'clock noon, when the equipment stops working and the temperature begins to drop.

[0061] Therefore, the state path of this device during this period can be expressed as: [8 o'clock, temperature gradually increases] -> [9 o'clock, temperature reaches the set value] -> [9 o'clock to 12 o'clock, temperature is stable] -> [after 12 o'clock, temperature drops]. This state path can be converted into a vector, for example, [temperature gradually increases, temperature reaches the set value, temperature is stable, temperature drops], which is the first state path vector.

[0062] For example, the target production line equipment status data refers to all relevant data of a certain equipment (such as a filling machine) in the milk powder production line. These data may include the running speed, temperature, pressure, etc. of the equipment. These equipment status data are regarded as part of the target production line monitoring events because they are all events or data recorded when monitoring the production line.

[0063] The target abnormal monitoring items are the items that are paid special attention to during the monitoring process. For example, you may pay special attention to whether the running speed of the filling machine is normal. This is an abnormal monitoring item.

[0064] For example, when it is found that the running speed of the filling machine exceeds the normal range, a target abnormal node is formed. This target abnormal node reflects the abnormal situation of the abnormal monitoring item of concern (i.e., the running speed of the filling machine) in the entire equipment status data. This abnormal situation can also be regarded as an abnormal state partition of the equipment status data, because it represents a part of the abnormal area in the equipment status data.

[0065] For example, if the normal operating speed of the filling machine is 50 bottles per minute, but it is found that at a certain moment, the operating speed of the filling machine reaches 70 bottles per minute, this is an abnormal node. In this example, the operating speed of the filling machine is the target abnormal monitoring item, and the state of 70 bottles per minute is an abnormal state partition in the target production line equipment status data.

[0066] Step S120, based on the first abnormal monitoring item node corresponding to the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, obtain the abnormal monitoring item status data in the x+1-th production line equipment status data of the target production line monitoring event, and obtain the second state path vector corresponding to the abnormal monitoring item status data.

[0067] In this embodiment, the status generation timing position of the x-th production line equipment status data in the target production line monitoring event is not earlier than the status generation timing position of the target production line equipment status data in the target production line monitoring event, and the status generation timing position of the x-th production line equipment status data in the target production line monitoring event is earlier than the status generation timing position of the x+1-th production line equipment status data in the target production line monitoring event, and x is a positive integer.

[0068] For example, the xth and x+1th production line equipment may refer to filling and packaging equipment. Assuming that an abnormal node also appears on the xth device (filling equipment), such as the filling speed is too fast, this abnormal node will be recorded, and the abnormal monitoring item status data will be obtained from the status data of the next x+1th device (packaging equipment), that is, the status data of the packaging equipment will be recorded and the abnormal situation will be found.

[0069] That is, let’s continue with the above-mentioned milk powder production line as an example.

[0070] In the first step, the processing equipment (the xth equipment) has been focused on and its state path vector has been obtained. Assuming that the target abnormal monitoring item is the temperature of the equipment, and during the monitoring process, it is found that the temperature of the equipment is too high, this is the first abnormal monitoring item node.

[0071] Then, turn your attention to the next device, the filling device (device x+1). Start collecting and recording the operating status data of this device, including its operating speed, pressure, etc., and pay attention to whether there are abnormal conditions caused by the excessive temperature of the upstream equipment (processing equipment). For example, if the high temperature of the processing equipment causes a change in product quality, thereby affecting the normal operation of the filling equipment, such as a slowdown in the filling speed, then this slowdown in the filling speed is an abnormal monitoring item.

[0072] At the same time, the state changes of the filling equipment during this period will be recorded to form a state path, such as [normal operation, slow speed, speed returns to normal]. This state path can be converted into a vector, for example, [normal operation, slow speed, speed returns to normal], which is the second state path vector.

[0073] In this process, the equipment status data is acquired and analyzed on the xth and x+1th equipment based on the target production line monitoring events (i.e. the operation status of the milk powder production line). In this way, problems that may affect production efficiency and product quality can be discovered and handled in a timely manner.

[0074] Step S130, fuse the second state path vector and the first state path vector to output the corresponding target state path vector, determine the abnormal indicator of the abnormal monitoring item corresponding to the abnormal monitoring item status data based on the target state path vector, and obtain the linkage characteristic parameters of the first abnormal monitoring item node in the equipment status data of the x+1th production line based on the target state path vector.

[0075] For example, the abnormal information of the processing equipment (first state path vector) and the filling equipment (second state path vector) can be fused to output a new target state path vector. Based on this new target state path vector, the abnormal indicators corresponding to the abnormal monitoring item status data can be determined, such as the duration of excessive temperature, the frequency of excessive filling speed, etc.

[0076] That is, continuing with the above example of the milk powder production line, there are now two state path vectors: the first state path vector (temperature change of the processing equipment) and the second state path vector (operating speed change of the filling equipment).

[0077] Fusing these two vectors means combining them to generate a new target state path vector. This vector contains all the important state information of the two devices of interest over a period of time. How to fuse them depends on the actual application, which may be a simple splicing or may involve more complex mathematical operations.

[0078] Assuming the two vectors are simply concatenated, the target state path vector might look like this: [the temperature of the processing equipment gradually increases, the temperature reaches the set value, the temperature stabilizes, the temperature drops, the filling equipment operates normally, the speed slows down, and the speed returns to normal].

[0079] Based on this target state path vector, the abnormal monitoring item abnormality indicator corresponding to the abnormal monitoring item state data can be determined. For example, it may be found that whenever the temperature of the processing equipment exceeds a certain threshold, the operation speed of the filling equipment will slow down. Therefore, "the temperature of the processing equipment exceeds the threshold" is an abnormal monitoring item abnormality indicator.

[0080] At the same time, based on the target state path vector, the linkage characteristic parameters of the first abnormal monitoring project node (the temperature of the processing equipment is too high) in the x+1 production line equipment status data (the running speed of the filling equipment) can also be obtained. This may involve parameters including but not limited to: the time point when the equipment temperature is too high, the duration, the degree of exceeding the threshold, etc., and how these parameters affect the running speed of the filling equipment. These linkage characteristic parameters help to understand and analyze the mutual influence relationship between equipment in order to better optimize the production process.

[0081] Step S140, determining a second abnormal monitoring project node reflecting the abnormal status partition of the target abnormal monitoring project in the x+1th production line equipment status data based on the abnormal indicator of the abnormal monitoring project, the first abnormal monitoring project node and the linkage characteristic parameter.

[0082] For example, based on the obtained abnormal indicators, the abnormal state partition reflecting the target abnormality (the temperature of the processing equipment is too high) can be found in the status data of the x+1th device (packaging equipment), that is, the abnormal situation on the packaging equipment that may be caused by the high temperature can be found. This is the second abnormal monitoring project node. For example, the product quality may be reduced due to the high temperature, so that the packaging equipment cannot correctly identify the product shape for packaging.

[0083] That is, in the milk powder production line example, the abnormal indicator of the abnormal monitoring item may be the severity or possibility of the quantitative target abnormal monitoring item (such as "equipment temperature is too high") under the current production line status, and the first abnormal monitoring item node is that the temperature of the processing equipment is too high. The linkage characteristic parameters may include the time point when the temperature is too high, the duration of time, the degree of exceeding the threshold, etc.

[0084] Now it is necessary to determine the second abnormal monitoring item node of the abnormal status partition reflecting the target abnormal monitoring item (the temperature of the processing equipment is too high) in the status data of the x+1th production line equipment (filling equipment).

[0085] Based on the above information, we start to look for abnormal conditions of filling equipment related to excessive temperature of processing equipment. Through observation and analysis, we find that whenever the temperature of processing equipment exceeds the threshold, the operation speed of filling equipment will slow down. Therefore, "slowing down of operation speed of filling equipment" can be defined as the second abnormal monitoring item node, which reflects the abnormal conditions on filling equipment caused by excessive temperature of processing equipment.

[0086] In this way, an abnormal state partition is identified in the filling equipment status data, that is, those status data associated with the overtemperature of the processing equipment, which causes the filling equipment to slow down. This abnormal state partition can help further understand and control the production process and avoid similar problems from happening again.

[0087] In general, this process is to continuously monitor the status of each device on the entire production line, identify abnormal situations, record and analyze them, and finally determine the cause of the abnormality and the equipment that may be affected, so as to make timely adjustments and optimizations.

[0088] For example, after identifying the second abnormal monitoring item node, a series of subsequent operations can be performed according to its status. For example, in the milk powder production line example, if the second abnormal monitoring item node is "filling equipment running speed slows down", then possible subsequent applications include:

[0089] 1. Fault warning: Operators can be immediately warned that the filling equipment may slow down due to excessive temperatures in the processing equipment.

[0090] 2. Fault diagnosis: The problem can be further analyzed to determine, for example, whether the reduced material flow is caused by the high temperature of the processing equipment or some other reason, thereby helping the operator find a solution to the problem.

[0091] 3. Preventive maintenance: If this problem occurs frequently, operators can be advised to regularly inspect and maintain the processing equipment to prevent the equipment from overheating.

[0092] 4. Production optimization: The production plan can be adjusted based on this problem, such as changing the processing order of materials, or adjusting the working mode of the equipment to avoid the problem of overheating of the equipment.

[0093] The above are just some possible follow-up applications, and the specific applications will be adjusted according to the actual situation. In general, after identifying the second abnormal monitoring project node, a series of measures can be taken to deal with and prevent this problem, thereby improving the operating efficiency and product quality of the production line.

[0094] Based on the above steps, by acquiring the target production line equipment status data from the milk powder production line and identifying the target abnormal node reflecting the target abnormal monitoring item, the corresponding first state path vector is generated. Then, based on the production line equipment status data of the current and next equipment (the xth and x+1th equipment), the abnormal monitoring item status data reflecting the abnormal monitoring item is acquired, and the corresponding second state path vector is generated. Next, the first and second state path vectors are fused and output as the target state path vector, and the abnormal indicators of the abnormal monitoring items corresponding to the abnormal monitoring item status data are determined based on the target state path vector, and the linkage characteristic parameters of the abnormality in the next equipment state data are obtained. Finally, according to the abnormal indicators of the abnormal monitoring items, the abnormal monitoring item nodes and the linkage characteristic parameters, the abnormal monitoring item nodes reflecting the abnormal state partition of the target abnormal monitoring item are determined in the next equipment state data. Therefore, by monitoring the target production line equipment status data in the milk powder production line and predicting possible abnormalities, the problem can be handled in advance before it affects production, avoiding production stagnation caused by equipment failure, and improving the operation efficiency of the production line and the production quality of milk powder products.

[0095] In a possible implementation, step S110 may include:

[0096] Step S111, obtaining the target production line monitoring event, splitting the target production line monitoring event, generating a target production line equipment status data set, and obtaining the first production line equipment status data in the target production line equipment status data set.

[0097] For example, the target production line monitoring events of the milk powder production line within one hour are obtained. These target production line monitoring events may contain the operating status information of each device. Therefore, these target production line monitoring events can be split according to the device to generate a target production line device status data set. This target production line device status data set may include temperature change data of the processing equipment, speed change data of the filling equipment, etc.

[0098] Step S112, if it is parsed that the first production line equipment status data contains the target abnormal monitoring item, the first production line equipment status data is output as the target production line equipment status data in the target production line monitoring event, and the abnormal status partition of the target abnormal monitoring item in the target production line equipment status data is output as the target abnormal node.

[0099] For example, assuming that the first device is a processing device, the first production line device status data obtained is the status data of the processing device. Assume that the target abnormal monitoring item of interest is "the temperature of the processing device is too high". If the temperature of the device is found to exceed the normal range in the status data of the processing device, then this status data is output as the target production line device status data in the target production line monitoring event, and at the same time, the abnormal status partition of the temperature being too high is output as the target abnormal node.

[0100] Step S113: Generate a state data sequence associated with the target abnormal node in the target production line equipment state data as a target abnormal node state sequence, and obtain a first state path vector corresponding to the target abnormal node state sequence.

[0101] For example, all the associated state data of the processing equipment when the abnormal node of overtemperature occurs can be recorded, such as the running speed and pressure of the equipment. These data form a state sequence, which is the state sequence of the target abnormal node. Then, this state sequence is converted into a vector, which is the first state path vector to be found.

[0102] In a possible implementation, the above method may further include:

[0103] Step S114, obtaining K candidate abnormal nodes from the first production line equipment status data, obtaining abnormal node knowledge vectors corresponding to the K candidate abnormal nodes, estimating the abnormal node knowledge vectors, and generating estimated category information corresponding to the K candidate abnormal nodes. K is a positive integer.

[0104] For example, continuing with the example of the milk powder production line above, we will focus on the processing equipment (the first production line equipment). In the process of monitoring the processing equipment, in addition to the target abnormal monitoring item of overtemperature, some other possible abnormal conditions may be noticed. For example, it may be found that the running speed of the equipment occasionally becomes very slow, or the pressure of the equipment sometimes suddenly increases. These possible abnormal conditions are taken as candidate abnormal nodes. Assume that 3 (K=3) such candidate abnormal nodes are found. Then, for the candidate abnormal node "slow equipment running speed", the temperature, pressure, running time and other information of the equipment when this state occurs may be recorded to form an abnormal node knowledge vector. Then, these abnormal node knowledge vectors are estimated, such as using a machine learning model to predict the abnormal type that each abnormal node may correspond to. For example, after estimation, the abnormal type that each candidate abnormal node may correspond to is obtained, which is the estimated category information. For example, "slow equipment running speed" may be estimated as an abnormality of the "equipment aging" type, "increased equipment pressure" may be estimated as an abnormality of the "equipment failure" type, and so on.

[0105] Step S115 : when the estimated category information corresponding to each of the K candidate abnormal nodes contains category information associated with the target abnormal monitoring item, it is determined that the first production line equipment status data contains the target abnormal monitoring item.

[0106] For example, if the estimated category information of the abnormal node "Slow equipment operation speed" is found to be "Aging equipment", and it is known that aging equipment may lead to the target abnormal monitoring item of concern ("Processing equipment temperature is too high"), then it can be determined that the target abnormal monitoring item is included in the status data of the first production line equipment (processing equipment).

[0107] In a possible implementation, in step S112, outputting the abnormal state partition of the target abnormal monitoring item in the target production line equipment state data as a target abnormal node includes:

[0108] Step S1121: output, among the K candidate abnormal nodes, the candidate abnormal nodes corresponding to the category information associated with the target abnormal monitoring item as a temporary abnormal node sequence.

[0109] For example, continuing with the example of the milk powder production line, assume that in the status data of the processing equipment (the first production line equipment), three (K=3) candidate abnormal nodes are found: slow equipment speed, increased equipment pressure, and high equipment temperature.

[0110] Assuming that the target abnormal monitoring item is "device temperature is too high", then the candidate abnormal node is output as a temporary abnormal node sequence. This temporary abnormal node sequence contains one element: "device temperature is too high".

[0111] Step S1122: Based on the Jaccard similarity coefficients between the candidate abnormal nodes in the temporary abnormal node sequence, the candidate abnormal nodes in the temporary abnormal node sequence are integrated to generate abnormal nodes to be updated.

[0112] The Jaccard similarity coefficient is an indicator used to measure the similarity between two sets. Here, since the temporary abnormal node sequence has only one element, no calculation is required. The node with "device temperature too high" is the abnormal node to be updated.

[0113] Step S1123, optimizing the abnormal node boundary of the abnormal node to be updated, and obtaining the target abnormal node including the target abnormal monitoring item in the target production line equipment status data.

[0114] For example, abnormal node boundary optimization is a technique used to more accurately determine the scope of abnormal nodes. For example, the specific situation of the device temperature being too high may be observed, such as whether it rises suddenly or gradually, how much the increase is, how long it lasts, etc., in order to more accurately determine the boundary of this abnormal node.

[0115] In this process, it may be found that when the device temperature exceeds a certain threshold, the device's operating speed will decrease. Therefore, this "device temperature is too high and causes the device's operating speed to decrease" state can be defined as the target abnormal node. In this way, the target abnormal node containing the target abnormal monitoring item is found in the target production line equipment status data.

[0116] In a possible implementation, step S120 may include:

[0117] Step S121, obtain the first abnormal monitoring item node of the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, and determine the first selected abnormal node in the x+1-th production line equipment status data of the target production line monitoring event based on the prior attention feature vector corresponding to the first abnormal monitoring item node.

[0118] For example, continuing with the example of the milk powder production line, the abnormal monitoring item node ("Equipment Temperature Too High") of the processing equipment (the xth equipment) has been obtained, and the next focus will be on the filling equipment (the x+1th equipment).

[0119] First, we get the abnormal monitoring project node of the processing equipment - the equipment temperature is too high. Then, based on the prior attention feature vector corresponding to this node (for example, it may contain information such as the frequency and duration of the node), we look for possible abnormal nodes in the status data of the filling equipment. For example, if we find that when the temperature of the processing equipment is too high, the running speed of the filling equipment will slow down, then "the running speed of the filling equipment slows down" is the first selected abnormal node.

[0120] Step S122: According to the first selected abnormal node, abnormal node tracking is performed in the x+1th production line equipment status data to generate at least one tracked abnormal node, and the abnormal feedback value corresponding to each of the at least one tracked abnormal node is obtained based on the abnormal feedback extraction model.

[0121] For example, in the status data of the filling equipment, track the abnormal node of "filling equipment running slows down" to see if it persists or other related abnormal conditions occur. These tracked abnormal nodes are all tracking abnormal nodes. Then, based on an abnormal feedback extraction model (which may be a machine learning model), obtain the abnormal feedback value corresponding to each of these tracking abnormal nodes. This value may reflect the severity or impact range of the abnormality.

[0122] Step S123: output the tracking abnormal node corresponding to the maximum abnormal feedback value as the second selected abnormal node, and generate the state data sequence associated with the second selected abnormal node in the x+1th production line equipment state data as the abnormal monitoring item state data.

[0123] For example, select the tracking abnormal node corresponding to the largest abnormal feedback value as the second selected abnormal node. Then, record the state data sequence associated with this second selected abnormal node in the filling equipment state data, such as the start time, end time, and magnitude of the equipment running speed slowdown. This state data sequence is the abnormal monitoring item state data to be found.

[0124] Step S124, loading the abnormal monitoring item status data into the recursive neural network in the target abnormal tracking network model, and outputting a second state path vector corresponding to the abnormal monitoring item status data according to the recursive neural network in the target abnormal tracking network model.

[0125] For example, the abnormal monitoring project status data is input into a recursive neural network in a target abnormal tracking network model. A recursive neural network is a deep learning model that can process sequence data and is very suitable for processing such time-varying status data. Through this network, the second state path vector corresponding to the abnormal monitoring project status data can be obtained. This vector represents the change of the abnormal state of the filling equipment over time, providing a basis for further analysis and handling of abnormal problems.

[0126] In a possible implementation, step S130 may include:

[0127] Step S131, loading the second state path vector into a first encoder, and outputting first encoding feature data according to the first encoder. The encoding feature dimension of the first encoding feature data is consistent with the encoding feature dimension of the second state path vector.

[0128] For example, the second state path vector describing the change in the running speed of the filling equipment can be input into the first encoder (for example, a neural network model), and the first encoder will convert this vector into a new form of expression, namely the so-called first encoded feature data. The encoding feature dimension of this new feature data is consistent with the encoding feature dimension of the original second state path vector.

[0129] Step S132, loading the first state path vector into a second encoder, and outputting second encoding feature data according to the second encoder. The encoding feature dimension of the second encoding feature data is different from the encoding feature dimension of the first state path vector.

[0130] For example, a first state path vector describing the temperature change of a processing device can be input into another second encoder to obtain second encoded feature data. This second encoder may be different from the first encoder, so the encoding feature dimension of the second encoded feature data may be different from the encoding feature dimension of the first state path vector. Among them, the encoding feature dimension, that is, the length or number of elements of the feature vector, determines the amount of information that the feature vector can contain. If the encoding feature dimension is large, then the feature vector can contain more information; conversely, if the encoding feature dimension is small, then the feature vector can contain less information.

[0131] In this example, the first state path vector and the second encoded feature data have different encoding feature dimensions, which may be because the information they represent or the way they are processed are different. For example, the first state path vector may be the original feature directly extracted from the device state data, while the second encoded feature data may be a new feature obtained after some transformation or dimensionality reduction processing.

[0132] Specifically, assume that the first state path vector is a 10-dimensional vector, representing the temperature of the processing equipment at 10 time points. Then this 10-dimensional vector is input to the second encoder (such as a neural network model), which processes the input 10-dimensional vector and outputs a 5-dimensional new vector, which is the second encoded feature data. Although the dimension of the new vector is reduced, due to the role of the encoder, it may still contain the main information of the original 10-dimensional vector, and may even contain some patterns or relationships that cannot be directly seen in the original vector.

[0133] Step S133: blending the first encoding feature data and the second encoding feature data to generate the target state path vector.

[0134] For example, the two sets of coded feature data are blended. There may be many specific blending methods, such as simple splicing or more complex mathematical operations. In the end, the new vector obtained is the target state path vector, which contains all important state information of the processing equipment and the filling equipment over a period of time.

[0135] In this way, the first coded feature data and the second coded feature data are successfully fused together to form a target state path vector that fully reflects the state of the production line. This target state path vector can be used for subsequent tasks such as anomaly detection and prediction to help better manage and optimize the production process.

[0136] In a possible implementation, in step S130, determining the abnormal indicator of the abnormal monitoring item corresponding to the abnormal monitoring item state data based on the target state path vector, and obtaining the linkage characteristic parameter of the first abnormal monitoring item node in the x+1th production line equipment state data based on the target state path vector, includes:

[0137] Step S134 , estimating the target state path vector, and generating an abnormality indicator of an abnormality monitoring item of the abnormality monitoring item state data for the target abnormality monitoring item.

[0138] For example, the target state path vector is a fusion of the state information of the processing equipment and the filling equipment. This target state path vector can be estimated by some statistical methods or machine learning models to obtain an abnormal indicator of the abnormal monitoring item. This abnormal indicator may be a number that quantifies the severity or probability of the target abnormal monitoring item (such as "equipment temperature is too high") under the current production line state.

[0139] For example, there may be a model that receives as input a target state path vector and outputs a value between 0 and 1, indicating the probability that the device temperature is too high. If this value is larger, the probability that the device temperature is too high is greater, and vice versa.

[0140] Step S135, based on the target state path vector, analyze the reference abnormal node determined by the first abnormal monitoring project node in the x+1th production line equipment status data, and generate the linkage characteristic parameters of the first abnormal monitoring project node in the x+1th production line equipment status data.

[0141] For example, when the abnormal node "equipment temperature is too high" appears in the processing equipment, the filling equipment will have the problem of "slowing down the speed". This "slowing down the speed" problem is the reference abnormal node. The part of the target state path vector related to this reference abnormal node can be analyzed to see if they have some specific patterns or rules. For example, it may be found that whenever the temperature of the processing equipment exceeds a certain threshold, the running speed of the filling equipment will start to decrease after 10 minutes. This "10 minutes" is a linkage characteristic parameter, which describes the linkage relationship between the abnormal states of the two devices.

[0142] In this way, not only can abnormal conditions on the production line be monitored, but the interrelationships between these abnormalities can also be further understood, providing valuable information for optimizing the production process and preventing future abnormalities.

[0143] In a possible implementation, the linkage characteristic parameters include a first linkage characteristic value and a second linkage characteristic value, and the first linkage characteristic value and the second linkage characteristic value reflect the abnormal state partition linkage state of the link node on the same production line abnormal link of the first abnormal monitoring project node in the equipment status data of the x+1th production line.

[0144] Continuing with the example of the milk powder production line, assume that the target abnormal monitoring item node ("Equipment temperature is too high") is found in the status data of the processing equipment (the xth equipment), and the abnormal status partition ("Operation speed slows down") reflecting the target abnormal monitoring item is found in the status data of the filling equipment (the x+1th equipment). Next, we will focus on the packaging equipment (the x+2th equipment). In the above scenario, the first linkage feature value may indicate the time required from "Equipment temperature is too high" to "Operation speed slows down", while the second linkage feature value may indicate the duration of "Operation speed slows down".

[0145] Step S140 may include:

[0146] Step S141: if the abnormality indicator of the abnormal monitoring item is greater than a set value, it is determined that the abnormal monitoring item status data includes the target abnormal monitoring item.

[0147] For example, if the set threshold is 0.5, and the abnormality indicator of the abnormal monitoring item is 0.6, then it can be determined that the current production line status data contains the target abnormal monitoring item, that is, there is a problem of "equipment temperature is too high".

[0148] Step S142, based on the prior attention feature vector corresponding to the first abnormal monitoring project node, the first linkage feature value and the second linkage feature value, generate the abnormal node location data of the target abnormal monitoring project in the equipment status data of the x+1th production line, and determine the second abnormal monitoring project node in the equipment status data of the x+1th production line according to the abnormal node location data.

[0149] For example, based on the prior attention feature vector corresponding to the node "Equipment temperature is too high" (which may include information such as the frequency and duration of the node) and the two linkage feature values ​​mentioned above, the abnormal node location data of the target abnormal monitoring item in the filling equipment status data can be generated. This data can help more accurately find the abnormal part in the filling equipment status data, that is, the second abnormal monitoring item node.

[0150] Step S143: If the abnormality index of the abnormal monitoring item is not greater than the set value, it is determined that the target abnormal monitoring item is not included in the abnormal monitoring item status data, and the x+2th production line equipment status data in the target production line monitoring event is obtained. The status generation time sequence position of the x+2th production line equipment status data in the target production line monitoring event is later than the status generation time sequence position of the x+1th production line equipment status data in the target production line monitoring event.

[0151] For example, if the abnormality index of the abnormal monitoring item is 0.4, which is lower than the set threshold of 0.5, then it is considered that the current production line status data does not contain the target abnormal monitoring item. Then, the status data of the next device (i.e. packaging equipment) will be obtained to observe whether the target abnormal monitoring item exists.

[0152] Step S144, if it is parsed that the target abnormal monitoring item is included in the x+2th production line equipment status data, the target production line equipment status data in the target production line monitoring event is adjusted to the x+2th production line equipment status data, and the target abnormal node in the target production line equipment status data is adjusted to the abnormal status partition of the target abnormal monitoring item in the x+2th production line equipment status data.

[0153] Assume that the target abnormal monitoring item is found in the status data of the packaging equipment, for example, the problem of "equipment downtime" is found. Then the target production line equipment status data will be adjusted to the status data of the packaging equipment, and the target abnormal node will be adjusted to "equipment downtime". Then the subsequent abnormal monitoring and processing operations can be continued.

[0154] In this way, continuous abnormal monitoring can be carried out throughout the production line, and possible problems can be discovered and dealt with in a timely manner, thus ensuring the smooth progress of the production process.

[0155] In a possible implementation, the following further introduces a training example of an anomaly tracking network model for implementing the above method, which may include:

[0156] Step S101 , obtaining a sample production line monitoring event including a sample abnormal monitoring item, and obtaining sample reference production line equipment status data and a sample production line equipment status data sequence in the sample production line monitoring event.

[0157] In this embodiment, the state generation timing position of the example reference production line equipment status data in the example production line monitoring event precedes the state generation timing position of each example production line equipment status data in the example production line equipment status data sequence in the example production line monitoring event.

[0158] For example, suppose there is a sample production line monitoring event that has been annotated by an expert. This sample production line monitoring event contains a series of equipment status data and corresponding abnormal information. For example, the expert may have annotated that after the temperature of the processing equipment is too high, the running speed of the filling equipment will start to slow down after 10 minutes.

[0159] Step S102, based on the training annotation data associated with the example reference production line equipment status data, determine the example abnormal node corresponding to the example abnormal monitoring item in the example reference production line equipment status data, and obtain the first example state path vector of the example abnormal node in the example reference production line equipment status data and the second example state path vector of the example abnormal node in each example production line equipment status data through the recursive neural network in the basic abnormal tracking network model.

[0160] For example, firstly, based on the example abnormal node ("equipment temperature is too high") in the example reference production line equipment status data (i.e., the status data of the processing equipment), the first example state path vector is obtained through the recursive neural network in the basic abnormal tracking network model. Then, the same operation is performed on each example production line equipment status data (i.e., the status data of the filling equipment) to obtain the second example state path vector.

[0161] Step S103, fusing the second example state path vector and the first example state path vector to output a corresponding target example state path vector, and generating an abnormality location learning result corresponding to each example production line equipment state data according to the abnormality location unit in the basic abnormality tracking network model.

[0162] For example, the first example state path vector and the second example state path vector are fused to output the corresponding target example state path vector. Then, based on this target example state path vector, the abnormality location unit (including the classification layer and the prediction layer) in the network model is used to generate the abnormality location learning result corresponding to each example production line equipment state data. This result includes the classification probability distribution (reflecting the possibility of the abnormal type) and the linkage feature estimation value (reflecting the abnormal linkage state).

[0163] Step S104, based on the training annotation data associated with the example reference production line equipment status data, the training annotation data associated with each example production line equipment status data, and the abnormality location learning result, the weight information of the basic abnormality tracking network model is updated, and the basic abnormality tracking network model containing the updated weight information is output as a target abnormality tracking network model. The target abnormality tracking network model is used to determine the abnormal state partition of the target abnormal monitoring item in the production line equipment status data contained in the target production line monitoring event.

[0164] For example, first, the target linkage feature value is determined based on the equipment status data of the example reference production line and the training annotation data associated with each example production line equipment status data. Then, the prediction cost value (i.e., the difference between the target linkage feature value and the linkage feature estimate) and the classification cost value (i.e., the difference between the classification probability distribution and the annotated category information) are calculated. Finally, the weight information of the network model is updated based on these two cost values ​​to obtain the updated target anomaly tracking network model.

[0165] The abnormality location unit includes a classification layer and a prediction layer. The abnormality location learning result includes a classification probability distribution generated by the classification layer and a linkage feature estimation value generated by the prediction layer. The linkage feature estimation value reflects the linkage state of the abnormal state partition of the link node on the same production line abnormal link of the example abnormal node in the equipment status data of each example production line.

[0166] For example, in the anomaly tracking network model, the anomaly location unit consists of a classification layer and a prediction layer. The classification layer generates a classification probability distribution based on the input target example state path vector, indicating the probability of occurrence of each possible anomaly category (such as "equipment temperature is too high", "operation speed is slowed down", etc.). The prediction layer generates a linkage feature estimate that reflects the abnormal state partition linkage state of the link nodes (such as processing equipment and filling equipment) on the same production line abnormal link.

[0167] Step S104 may include:

[0168] Step S1041, based on the labeled abnormal state partition data in the training labeled data associated with the equipment status data of the example reference production line, and the labeled abnormal state partition data in the training labeled data associated with each of the equipment status data of the example production line, determine the target linkage feature value of the example abnormal node in each of the equipment status data of the example production line.

[0169] For example, first, based on the labeled abnormal state partition data in the training labeled data associated with the equipment state data of the example reference production line (i.e., the state data of the processing equipment) and each example production line equipment state data (i.e., the state data of the filling equipment), the target linkage feature value of the example abnormal node ("equipment temperature is too high") in each example production line equipment state data is determined. This value may represent the time required from "equipment temperature is too high" to "operation speed slows down".

[0170] Step S1042: determining a prediction cost value corresponding to the prediction layer according to a difference between the target linkage feature value and the linkage feature estimation value.

[0171] Step S1043 , determining the classification cost corresponding to the classification layer according to the classification probability distribution and the labeled category information in the training labeled data respectively associated with each example production line equipment status data.

[0172] For example, the prediction cost is determined by the difference between the target linkage feature value and the linkage feature estimate, which indicates the gap between the prediction result of the prediction layer and the actual situation. The classification cost is determined by the difference between the classification probability distribution and the labeled category information in the training labeled data associated with each example production line equipment status data, which indicates the gap between the classification result of the classification layer and the actual category.

[0173] Step S1044: based on the prediction cost value and the classification cost value, the weight information of the basic anomaly tracing network model is updated, and the basic anomaly tracing network model including the updated weight information is output as a target anomaly tracing network model.

[0174] For example, the weight information of the network model is updated based on the prediction cost value and the classification cost value. This process is usually completed through an algorithm called back propagation. Specifically, the weight of the network model is adjusted accordingly based on the size of the prediction cost value and the classification cost value, so that the cost value can be minimized in future predictions. Then, the network model containing the updated weight information is output as the target anomaly tracking network model.

[0175] Figure 2 The diagram schematically shows a production line intelligent monitoring system 100 that can be used to implement various embodiments described in the present application.

[0176] For one embodiment, Figure 2 A production line intelligent monitoring system 100 is shown, which has multiple processors 102, a control module (chip set) 104 coupled to one or more of the (multiple) processors 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM) / storage device 108 coupled to the control module 104, multiple input / output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 104.

[0177] The processor 102 may include multiple single-core or multi-core processors, and the processor 102 may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). In some alternative implementations, the production line intelligent monitoring system 100 can be used as a server device such as a gateway described in the embodiments of the present application.

[0178] In some alternative embodiments, the production line intelligent monitoring system 100 may include multiple computer-readable media (e.g., memory 106 or NVM / storage device 108) having instructions 114 and multiple processors 102 combined with the multiple computer-readable media and configured to execute the instructions 114 to implement modules to perform the actions described in the present disclosure.

[0179] For one embodiment, the control module 104 may include any suitable interface controller to provide any suitable interface to one or more of the processor(s) 102 and / or any suitable device or component in communication with the control module 104 .

[0180] The control module 104 may include a memory controller module to provide an interface to the memory 106. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0181] The memory 106 may be used, for example, to load and store data and / or instructions 114 for the production line intelligent monitoring system 100. For one embodiment, the memory 106 may include any suitable volatile memory, for example, a suitable DRAM. In some alternative implementations, the memory 106 may include a double data rate type four synchronous dynamic random access memory.

[0182] For one embodiment, the control module 104 may include a plurality of input / output controllers to provide interfaces to the NVM / storage device 108 and the input / output device(s) 110 .

[0183] For example, NVM / storage device 108 may be used to store data and / or instructions 114. NVM / storage device 108 may include any suitable non-volatile memory (eg, flash memory) and / or may include any suitable non-volatile storage device(s).

[0184] The NVM / storage device 108 may include storage resources that are physically part of the device on which the production line intelligent monitoring system 100 is installed, or it may be accessible to the device without being part of the device. For example, the NVM / storage device 108 may be accessed via (multiple) input / output devices 110 in conjunction with a network.

[0185] The (multiple) input / output devices 110 may provide an interface for the production line intelligent monitoring system 100 to communicate with any other appropriate device, and the input / output device 110 may include a communication component, a pinyin component, a sensor component, etc. The network interface 112 may provide an interface for the production line intelligent monitoring system 100 to communicate based on multiple networks, and the production line intelligent monitoring system 100 may wirelessly communicate with multiple components of a wireless network based on any of a plurality of wireless network standards and / or protocols, for example, access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0186] For one embodiment, one or more of the processor(s) 102 may be packaged together with the logic of multiple controllers (e.g., a memory controller module) of the control module 104. For one embodiment, one or more of the processor(s) 102 may be packaged together with the logic of multiple controllers of the control module 104 to form a system-in-package. For one embodiment, one or more of the processor(s) 102 may be integrated on the same module with the logic of multiple controllers of the control module 104. For one embodiment, one or more of the processor(s) 102 may be integrated on the same die with the logic of multiple controllers of the control module 104 to form a system-on-chip.

[0187] In various embodiments, the production line intelligent monitoring system 100 may be, but is not limited to, a terminal device such as a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the production line intelligent monitoring system 100 may have more or fewer components and / or different architectures. For example, in some alternative implementations, the production line intelligent monitoring system 100 includes multiple cameras, a keyboard, a liquid crystal display screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, a dedicated integrated circuit, and a speaker.

[0188] The above is a detailed introduction to the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An intelligent monitoring method for a milk powder production line based on artificial intelligence, characterized in that: Applied to the intelligent monitoring system of a production line, the method comprises: Obtaining a first state path vector corresponding to a target abnormal node in the target production line equipment state data in the milk powder production line; the target production line equipment state data belongs to a target production line monitoring event, the target production line monitoring event includes a target abnormal monitoring item, and the target abnormal node reflects the abnormal state partition of the target abnormal monitoring item in the target production line equipment state data; Based on the first abnormal monitoring item node corresponding to the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, the abnormal monitoring item status data is obtained in the x+1-th production line equipment status data of the target production line monitoring event, and the second state path vector corresponding to the abnormal monitoring item status data is obtained; the state generation timing position of the x-th production line equipment status data in the target production line monitoring event is not earlier than the state generation timing position of the target production line equipment status data in the target production line monitoring event, and the state generation timing position of the x-th production line equipment status data in the target production line monitoring event is prior to the state generation timing position of the x+1-th production line equipment status data in the target production line monitoring event, and x is a positive integer; The second state path vector and the first state path vector are merged to output a corresponding target state path vector, an abnormal indicator of the abnormal monitoring item corresponding to the abnormal monitoring item state data is determined based on the target state path vector, and a linkage characteristic parameter of the first abnormal monitoring item node in the equipment state data of the x+1th production line is obtained based on the target state path vector; Determine, in the x+1th production line equipment status data, a second abnormal monitoring item node reflecting the abnormal status partition of the target abnormal monitoring item according to the abnormal indicator of the abnormal monitoring item, the first abnormal monitoring item node and the linkage characteristic parameter; The step of obtaining a first state path vector corresponding to a target abnormal node in the target production line equipment state data in the milk powder production line includes: Acquire the target production line monitoring event, split the target production line monitoring event, generate a target production line equipment status data set, and acquire the first production line equipment status data in the target production line equipment status data set; If it is parsed that the first production line equipment status data contains the target abnormal monitoring item, the first production line equipment status data is output as the target production line equipment status data in the target production line monitoring event, and the abnormal status partition of the target abnormal monitoring item in the target production line equipment status data is output as the target abnormal node; The state data sequence associated with the target abnormal node in the target production line equipment state data is generated as a target abnormal node state sequence, and a first state path vector corresponding to the target abnormal node state sequence is obtained.

2. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The method further comprises: Obtain K candidate abnormal nodes from the first production line equipment status data, obtain abnormal node knowledge vectors corresponding to each of the K candidate abnormal nodes, estimate the abnormal node knowledge vectors, and generate estimated category information corresponding to each of the K candidate abnormal nodes; K is a positive integer; When the estimated category information corresponding to each of the K candidate abnormal nodes contains category information associated with the target abnormal monitoring item, it is determined that the first production line equipment status data contains the target abnormal monitoring item.

3. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 2, characterized in that: The step of outputting the abnormal state partition of the target abnormal monitoring item in the target production line equipment state data as a target abnormal node includes: Among the K candidate abnormal nodes, outputting the candidate abnormal nodes corresponding to the category information associated with the target abnormal monitoring item as a temporary abnormal node sequence; Based on the Jaccard similarity coefficients between the candidate abnormal nodes in the temporary abnormal node sequence, the candidate abnormal nodes in the temporary abnormal node sequence are integrated to generate abnormal nodes to be updated; The abnormal node to be updated is optimized in terms of its boundary, and a target abnormal node including the target abnormal monitoring item is obtained in the target production line equipment status data.

4. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The first abnormal monitoring item node corresponding to the target abnormal monitoring item in the xth production line equipment status data of the target production line monitoring event, obtaining abnormal monitoring item status data in the x+1th production line equipment status data of the target production line monitoring event, and obtaining a second state path vector corresponding to the abnormal monitoring item status data include: Obtaining a first abnormal monitoring item node of the target abnormal monitoring item in the x-th production line equipment status data of the target production line monitoring event, and determining a first selected abnormal node in the x+1-th production line equipment status data of the target production line monitoring event based on a priori attention feature vector corresponding to the first abnormal monitoring item node; According to the first selection of abnormal nodes, abnormal node tracking is performed in the x+1th production line equipment status data to generate at least one tracked abnormal node, and abnormal feedback values ​​corresponding to each of the at least one tracked abnormal node are obtained based on an abnormal feedback extraction model; Outputting the tracking abnormal node corresponding to the maximum abnormal feedback value as the second selected abnormal node, and generating the state data sequence associated with the second selected abnormal node in the x+1th production line equipment state data as the abnormal monitoring item state data; The abnormal monitoring item status data is loaded into a recursive neural network in a target abnormal tracking network model, and a second state path vector corresponding to the abnormal monitoring item status data is output according to the recursive neural network in the target abnormal tracking network model.

5. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The step of fusing the second state path vector and the first state path vector to output a corresponding target state path vector includes: Loading the second state path vector to the first encoder, and outputting first encoding feature data according to the first encoder; the encoding feature dimension of the first encoding feature data is consistent with the encoding feature dimension of the second state path vector; Loading the first state path vector to a second encoder, and outputting second encoding feature data according to the second encoder; the encoding feature dimension of the second encoding feature data is different from the encoding feature dimension of the first state path vector; The first encoded feature data and the second encoded feature data are blended to generate the target state path vector.

6. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The determining, based on the target state path vector, the abnormal indicator of the abnormal monitoring item corresponding to the abnormal monitoring item state data, and acquiring, based on the target state path vector, the linkage characteristic parameter of the first abnormal monitoring item node in the x+1th production line equipment state data, comprises: The target state path vector is estimated to generate an abnormal monitoring item abnormality indicator of the abnormal monitoring item state data for the target abnormal monitoring item; Based on the target state path vector, the reference abnormal node determined by the first abnormal monitoring project node in the x+1th production line equipment status data is analyzed to generate linkage characteristic parameters of the first abnormal monitoring project node in the x+1th production line equipment status data.

7. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The linkage characteristic parameter includes a first linkage characteristic value and a second linkage characteristic value, wherein the first linkage characteristic value and the second linkage characteristic value reflect the abnormal state partition linkage state of the link node on the same production line abnormal link of the first abnormal monitoring item node in the equipment state data of the x+1th production line; The determining, based on the abnormal indicator of the abnormal monitoring item, the first abnormal monitoring item node and the linkage characteristic parameter, in the x+1th production line equipment status data, a second abnormal monitoring item node reflecting the abnormal status partition of the target abnormal monitoring item comprises: If the abnormality indicator of the abnormal monitoring item is greater than a set value, determining that the abnormal monitoring item status data includes the target abnormal monitoring item; Based on the prior attention feature vector corresponding to the first abnormal monitoring item node, the first linkage feature value, and the second linkage feature value, abnormal node location data of the target abnormal monitoring item in the x+1th production line equipment status data is generated, and according to the abnormal node location data, the second abnormal monitoring item node is determined in the x+1th production line equipment status data; If the abnormal indicator of the abnormal monitoring item is not greater than the set value, it is determined that the target abnormal monitoring item is not included in the abnormal monitoring item status data, and the x+2th production line equipment status data in the target production line monitoring event is obtained; the status generation timing position of the x+2th production line equipment status data in the target production line monitoring event is later than the status generation timing position of the x+1th production line equipment status data in the target production line monitoring event; If it is parsed that the target abnormal monitoring item is included in the x+2th production line equipment status data, the target production line equipment status data in the target production line monitoring event is adjusted to the x+2th production line equipment status data, and the target abnormal node in the target production line equipment status data is adjusted to the abnormal status partition of the target abnormal monitoring item in the x+2th production line equipment status data.

8. The method for intelligent monitoring of a milk powder production line based on artificial intelligence according to claim 1, characterized in that: The method further comprises: Obtaining an example production line monitoring event including an example abnormal monitoring item, obtaining example reference production line equipment status data and an example production line equipment status data sequence in the example production line monitoring event; the state generation time sequence position of the example reference production line equipment status data in the example production line monitoring event precedes the state generation time sequence position of each example production line equipment status data in the example production line equipment status data sequence in the example production line monitoring event; Based on the training annotation data associated with the example reference production line equipment status data, determine the example abnormal node corresponding to the example abnormal monitoring item in the example reference production line equipment status data, and obtain the first example state path vector of the example abnormal node in the example reference production line equipment status data and the second example state path vector of the example abnormal node in each example production line equipment status data through a recursive neural network in a basic abnormality tracking network model; The second example state path vector and the first example state path vector are merged to output a corresponding target example state path vector, and an abnormality location learning result corresponding to each example production line equipment state data is generated according to the abnormality location unit in the basic abnormality tracking network model; Based on the training annotation data associated with the example reference production line equipment status data, the training annotation data respectively associated with each example production line equipment status data, and the abnormality localization learning result, the weight information of the basic abnormality tracking network model is updated, and the basic abnormality tracking network model containing the updated weight information is output as a target abnormality tracking network model; the target abnormality tracking network model is used to determine the abnormal state partition of the target abnormal monitoring item in the production line equipment status data contained in the target production line monitoring event; The abnormality location unit includes a classification layer and a prediction layer, the abnormality location learning result includes a classification probability distribution generated by the classification layer and a linkage feature estimation value generated by the prediction layer, and the linkage feature estimation value reflects the abnormal state partition linkage state of the link node on the same production line abnormal link of the example abnormal node in the equipment state data of each example production line; The method of updating the weight information of the basic abnormality tracking network model based on the training annotation data associated with the equipment status data of the example reference production line, the training annotation data respectively associated with the equipment status data of each example production line, and the abnormality location learning result, and outputting the basic abnormality tracking network model including the updated weight information as the target abnormality tracking network model, includes: Determine the target linkage feature value of the example abnormal node in each example production line equipment status data based on the annotated abnormal state partition data in the training annotated data associated with the example reference production line equipment status data and the annotated abnormal state partition data in the training annotated data respectively associated with each example production line equipment status data; Determining a prediction cost value corresponding to the prediction layer according to a difference between the target linkage feature value and the linkage feature estimation value; Determine the classification cost value corresponding to the classification layer according to the classification probability distribution and the labeled category information in the training labeled data respectively associated with each example production line equipment status data; Based on the prediction cost value and the classification cost value, the weight information of the basic anomaly tracing network model is updated, and the basic anomaly tracing network model including the updated weight information is output as a target anomaly tracing network model.

9. A production line intelligent monitoring system, characterized in that: The production line intelligent monitoring system includes a processor and a machine-readable storage medium, in which machine-executable instructions are stored. The machine-executable instructions are loaded and executed by the processor to implement the artificial intelligence-based intelligent monitoring method for a milk powder production line as described in any one of claims 1 to 8.

Images