A method and system for monitoring production conditions

Abstract

This invention provides a method and system for monitoring production conditions. The method includes: collecting production condition parameters using sensors and conducting a comprehensive benefit assessment of finished products; training classifiers using single-class classification and supervised machine learning methods respectively; monitoring whether production conditions meet standards; and making adjustments at minimal cost when production conditions fail to meet standards. This invention can automatically monitor whether production conditions meet standards, mitigating the drawbacks of manual monitoring, ensuring the safety and high efficiency of the production process, reducing required manpower and material resources, and improving economic efficiency. Furthermore, it can quickly implement automatic monitoring using a single-class classifier when abnormal data is insufficient, while increasing the weight of the supervised learning classifier's prediction results as abnormal data accumulates, thus improving prediction accuracy. This adapts to various production condition monitoring application scenarios, making it highly practical and widely applicable.

Description

Technical Field

[0001] This invention belongs to the field of industrial control and production condition monitoring, and specifically relates to a production condition monitoring method and system. Background Technology

[0002] Production conditions encompass the production environment and related material conditions, factors, and states. The production of many products, such as chips, chemical products, biological products, pharmaceuticals, and handicrafts, has very strict requirements for production conditions. Production conditions are crucial factors affecting product quality, output, safe production, economic benefits, and social benefits. In many cases, there is a significant delay between the occurrence of abnormalities in production conditions and their detection in the product or related products; if abnormalities are not detected in a timely manner, serious consequences may have already occurred by the time they are discovered. Therefore, in industrial and agricultural production, continuous monitoring of production conditions is often necessary to ensure the safe, orderly, and efficient operation of the production process. With the increasing automation of modern industrial and agricultural production, the automatic monitoring of production conditions has received increasing attention.

[0003] Existing production condition monitoring mostly requires manual monitoring and adjustment, which is prone to problems due to the negligence and lack of experience of the monitoring personnel. Moreover, it usually requires a lot of manpower and material resources, which cannot meet the needs of modern large-scale production and intelligent manufacturing.

[0004] In addition, energy conservation is also an important factor to consider in production condition monitoring. Production conditions are generally a combination of multiple factors, and energy conservation also needs to be considered from multiple perspectives. When monitoring production conditions, in addition to ensuring overall product efficiency, energy conservation must also be taken into account.

[0005] How to achieve automated and intelligent monitoring and control of production conditions to meet the requirements of comprehensive product benefits and energy conservation has become an urgent problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for monitoring production conditions in order to meet the need for automated and intelligent monitoring of production conditions.

[0007] The first aspect of this invention provides a method for monitoring production conditions, comprising: collecting and storing production condition parameters through sensors, and conducting a comprehensive benefit assessment of the products coming off the production line; determining whether the corresponding production condition parameters are normal based on whether the comprehensive benefit assessment of the products coming off the production line meets the standards, and marking them accordingly; the comprehensive benefit assessment includes an assessment of one or more factors among product quality, production efficiency, environmental pollution risk, and energy consumption.

[0008] Using normal production condition parameters, a single-class classification machine learning method is used to train the first classifier to predict whether the production condition parameters are normal.

[0009] Using normal and abnormal production condition parameters and their labels, a second classifier is trained using supervised machine learning methods to predict whether the production condition parameters are normal.

[0010] The first and second classifiers are used respectively to predict whether the production condition parameters are normal; the weights of the prediction results of the first and second classifiers are preset; the prediction results of the production condition parameters are obtained based on the prediction results and weights of the first and second classifiers; if the prediction result of the production condition parameters is abnormal, the normal production condition parameters with the lowest adjustment cost compared with the current production condition parameters are searched within the preset range of production condition parameter changes according to the preset production condition adjustment cost calculation method, and this is taken as the target production condition parameter to adjust the production conditions; otherwise, the current production conditions are not changed.

[0011] Based on the above, using the data accumulated during the production process, the first and second classifiers are further trained and updated, and the weights of the prediction results of the first and second classifiers are updated.

[0012] Based on the above, as the amount of abnormal production condition parameter data accumulated during the production process increases, the weight of the prediction result of the first classifier is reduced, and the weight of the prediction result of the second classifier is increased.

[0013] Based on the above, the initial weight value of the first classifier is 1, and the initial weight value of the second classifier is 0.

[0014] Based on the above, the pre-set production condition adjustment cost calculation method includes a method for calculating the resources, costs, or time required to adjust the production conditions from a state where the production condition parameters are at the initial parameter values ​​to a state where the production condition parameters are at the target parameter values, or a combination thereof.

[0015] Based on the above, the preset range of production condition parameters includes a predefined set of normal production condition parameters.

[0016] Based on the above, the preset range of production condition parameters includes the range in which the preset production condition parameters change up or down with the current production condition parameters as the center.

[0017] Based on the above, the adjustment of production conditions includes: adjusting the production conditions by changing or setting the set values ​​of the production line and peripheral equipment according to the target production condition parameters.

[0018] Based on the above, the production condition parameters include one or more of the following: temperature, humidity, sound, light, pressure, displacement, speed, acceleration, voltage, current, flow rate, particle size, and concentration.

[0019] A second aspect of the present invention also provides a production condition monitoring system, comprising a training subsystem and a monitoring subsystem; the training subsystem includes a data acquisition and evaluation module, a first training module, and a second training module, used to train a first classifier and a second classifier using production condition parameters and comprehensive product benefit evaluation data; the monitoring subsystem includes a data acquisition module, a first prediction module, a second prediction module, a discrimination module, and a control module, used to monitor and control production conditions using the first classifier and the second classifier; wherein,

[0020] The data acquisition and evaluation module is used to collect production condition parameters through sensors and evaluate whether the comprehensive benefit assessment of the products off the production line meets the standards; based on whether the comprehensive benefit assessment of the products off the production line meets the standards, it determines whether the corresponding production condition parameters are normal and marks them; and sends the production condition parameters and corresponding markings to the first training module and the second training module.

[0021] The first training module receives production condition parameters and labels sent by the acquisition and evaluation module, and trains the first classifier using the normal production condition parameters and a single-class classification machine learning method.

[0022] The second training module receives production condition parameters and labels sent by the acquisition and evaluation module, and uses normal and abnormal production condition parameters and their labels to train the second classifier using supervised machine learning methods.

[0023] The acquisition module is used to collect production condition parameters through sensors and send them to the control module, the first prediction module, and the second prediction module.

[0024] The first prediction module is used to use the first classifier to predict whether the received production condition parameters are normal, and send the prediction result to the discrimination module.

[0025] The second prediction module is used to use the second classifier to predict whether the received production condition parameters are normal, and then send the prediction results to the discrimination module.

[0026] The discrimination module is used to obtain the discrimination result of the production condition parameters based on the prediction results sent by the first prediction module and the second prediction module, and based on the preset weights of the first classifier and the second classifier, and send it to the control module.

[0027] The control module is used to adjust the production conditions by changing, setting, or maintaining the set values ​​of the production line and peripheral equipment based on the received production condition parameters and the judgment results of the production condition parameters.

[0028] The acquisition and evaluation module is communicatively connected to the first training module and the second training module, respectively. The acquisition module and the discrimination module are communicatively connected to the first prediction module and the second prediction module, respectively. The control module is communicatively connected to the acquisition module and the discrimination module, so as to realize the above-mentioned production condition monitoring method.

[0029] This invention utilizes sensors to collect production condition parameters and performs a comprehensive benefit assessment of the finished products. Through analysis of the production condition parameters and the comprehensive benefit assessment results, classifiers are trained using both single-class classification and supervised machine learning methods to monitor whether production conditions meet standards and to make adjustments with minimal cost when conditions fail to meet them. Compared to existing technologies, this invention has significant substantive features and remarkable progress, specifically:

[0030] 1. The technical solution of this invention can realize automated and intelligent monitoring and control of production conditions, automatically monitor whether the production conditions meet the standards, and make adjustments at the lowest cost when the production conditions do not meet the standards. This can reduce the drawbacks of manual monitoring, ensure the safety and high efficiency of the production process, reduce the required manpower and material resources, reduce resource and energy consumption and production costs, and improve economic benefits.

[0031] 2. The technical solution of this invention can combine single-class classification and supervised machine learning methods. In the early stage of production condition monitoring, when there is insufficient abnormal data, it mainly uses a single-class classifier to quickly carry out automatic monitoring. As abnormal data in the production process accumulates, the weight of the prediction results of the supervised learning classifier is increased to improve the prediction accuracy. Therefore, it can adapt to various production condition monitoring application scenarios, has strong practicality and wide application range. Attached Figure Description

[0032] Figure 1 This is a flowchart of the production condition monitoring method provided in Embodiment 1 of the present invention.

[0033] Figure 2 This is a schematic diagram of the production condition monitoring system provided in Embodiment 3 of the present invention. Detailed Implementation

[0034] Example 1

[0035] This embodiment provides a method for monitoring production conditions, such as... Figure 1 As shown, it includes: collecting and storing production condition parameters through sensors, and conducting a comprehensive benefit assessment of the products coming off the production line; judging whether the corresponding production condition parameters are normal based on whether the comprehensive benefit assessment of the products coming off the production line meets the standards, and marking them accordingly; the comprehensive benefit assessment includes an assessment of one or more factors among product quality, production efficiency, environmental pollution risk, and energy consumption.

[0036] Using normal production condition parameters, a single-class classification machine learning method is used to train the first classifier to predict whether the production condition parameters are normal.

[0037] Using normal and abnormal production condition parameters and their labels, a second classifier is trained using supervised machine learning methods to predict whether the production condition parameters are normal.

[0038] The first and second classifiers are used respectively to predict whether the production condition parameters are normal; the weights of the prediction results of the first and second classifiers are preset; the prediction results of the production condition parameters are obtained based on the prediction results and weights of the first and second classifiers; if the prediction result of the production condition parameters is abnormal, the normal production condition parameters with the lowest adjustment cost compared with the current production condition parameters are searched within the preset range of production condition parameter changes according to the preset production condition adjustment cost calculation method, and this is taken as the target production condition parameter to adjust the production conditions; otherwise, the current production conditions are not changed.

[0039] In the first classifier and the second classifier, the classifier includes classification methods, algorithms, models and their parameters, and the form of the classifier includes computer software, devices, equipment, storage media, etc.

[0040] In this embodiment, the production condition parameters include one or more of the following: temperature, humidity, sound, gas, light, pressure, acceleration, voltage, and current. For each production condition parameter, one or more corresponding sensors can be used for data acquisition.

[0041] Product quality assessment can be conducted manually, or using methods such as image recognition and classification, or by employing detection methods based on sound, vibration, or light. For some types of products, destructive or invasive sampling inspections may also be performed. Product quality assessment results include scores, quality grades, or simple pass / fail statements.

[0042] Environmental pollution risks include the environmental pollution risks that may be caused by the product itself, as well as the environmental pollution risks caused by wastewater, exhaust gas, and waste generated during the production process.

[0043] Comprehensive benefit assessment is a quantitative assessment. It can be used to assess each factor that makes up the comprehensive benefit separately and use the assessment of each factor as a vector, or it can be used to organically combine the assessment of each factor into a single indicator.

[0044] The correspondence between the comprehensive benefit assessment results of a product and its corresponding production condition parameters usually refers to the correspondence between the comprehensive benefit assessment results of a product and the production condition parameters when producing the product.

[0045] Single-class classification refers to classification learning where only one class of data is available or focuses solely on one class; it's also called single-class learning. Since most production condition parameters are collected under normal production conditions, abnormal data is often scarce or missing, especially in the early stages of monitoring when only normal data is available. Therefore, single-class classification is necessary. Single-class classification uses the only available target data to establish a corresponding data description boundary, providing a description of the training sample set. This transforms the classification of unknown samples into a test of whether the unknown sample conforms to the learned description of the target sample. If it does, the unknown sample is accepted as a target class sample; otherwise, it is rejected. Single-class classification mainly includes probability density estimation methods, reconstruction methods, and support boundary methods.

[0046] Once enough time has been spent collecting production condition parameters and a sufficient amount of abnormal data has been accumulated, a supervised learning method can be used to train a classifier.

[0047] In this embodiment, data accumulated during the production process is used to continue training and updating the first and second classifiers, and the weights of their prediction results are also updated. As the amount of abnormal production condition parameter data accumulated during the production process increases, the weight of the first classifier's prediction result is decreased, while the weight of the second classifier's prediction result is increased. However, in the initial monitoring phase, when the number of abnormal production condition parameters used to train the second classifier is insufficient, the initial weight value of the first classifier is 1, and the initial weight value of the second classifier is 0.

[0048] This approach allows for rapid automated monitoring when there is insufficient abnormal data in the early stages, primarily using a single-class learner. As abnormal data accumulates during the production process, the weight of the supervised learning classifier's predictions is increased to improve prediction accuracy.

[0049] In this embodiment, the pre-set production condition adjustment cost calculation method includes a method for calculating the resources, costs, or time required to adjust the production conditions from the state where the production condition parameters are at the initial parameter values ​​to the state where the production condition parameters are at the target parameter values, or a combination thereof.

[0050] In this embodiment, adjusting the production conditions includes: changing or setting the settings of peripheral equipment according to the target production condition parameters to adjust the production conditions.

[0051] The peripheral equipment includes one or more of the following: air conditioner, humidifier, dehumidifier, pressure controller, light source controller, power controller, and flow controller.

[0052] In this embodiment, the preset range of production condition parameter variations includes a predefined set of normal production condition parameters. If the predicted result of the production condition parameters is abnormal, the normal production condition parameter with the lowest adjustment cost compared to the current production condition parameter is searched according to the pre-set production condition adjustment cost calculation method within this parameter set, and this parameter is used as the target production condition parameter to adjust the production conditions.

[0053] Example 2

[0054] This embodiment provides another method for monitoring production conditions, which differs from Embodiment 1 in that: in this embodiment, the preset range of production condition parameter changes includes the range within which the preset production condition parameter fluctuates around the current production condition parameter. The production condition parameter includes one or more indicators. If the predicted result of the production condition parameter is abnormal, then according to a pre-set production condition adjustment cost calculation method, within this range, based on a preset search step size, the effective values ​​of each indicator within the preset range of production condition parameter changes are obtained. Multiple candidate production condition parameters within this range are determined, and it is judged whether the candidate production condition parameter is a normal production condition parameter. The normal production condition parameter with the lowest adjustment cost compared to the current production condition parameter is selected as the target production condition parameter, and the production conditions are adjusted accordingly.

[0055] Example 3

[0056] This embodiment provides a production condition monitoring system, such as Figure 2 As shown, it includes a training subsystem and a monitoring subsystem. The training subsystem includes a data acquisition and evaluation module, a first training module, and a second training module, used to train a first classifier and a second classifier using production condition parameters and comprehensive product benefit evaluation data. The monitoring subsystem includes a data acquisition module, a first prediction module, a second prediction module, a discrimination module, and a control module, used to monitor and control production conditions using the first and second classifiers.

[0057] The data acquisition and evaluation module is used to collect production condition parameters through sensors and evaluate whether the comprehensive benefit assessment of the products off the production line meets the standards; based on whether the comprehensive benefit assessment of the products off the production line meets the standards, it determines whether the corresponding production condition parameters are normal and marks them; and sends the production condition parameters and corresponding markings to the first training module and the second training module.

[0058] The first training module receives production condition parameters and labels sent by the acquisition and evaluation module, and trains the first classifier using the normal production condition parameters and a single-class classification machine learning method.

[0059] The second training module receives production condition parameters and labels sent by the acquisition and evaluation module, and uses normal and abnormal production condition parameters and their labels to train the second classifier using supervised machine learning methods.

[0060] The acquisition module is used to collect production condition parameters through sensors and send them to the control module, the first prediction module, and the second prediction module.

[0061] The first prediction module is used to use the first classifier to predict whether the received production condition parameters are normal, and send the prediction result to the discrimination module.

[0062] The second prediction module is used to use the second classifier to predict whether the received production condition parameters are normal, and then send the prediction results to the discrimination module.

[0063] The discrimination module is used to obtain the discrimination result of the production condition parameters based on the prediction results sent by the first prediction module and the second prediction module, and based on the preset weights of the first classifier and the second classifier, and send it to the control module.

[0064] The control module is used to adjust the production conditions by changing, setting, or maintaining the set values ​​of the production line and peripheral equipment based on the received production condition parameters and the judgment results of the production condition parameters.

[0065] The acquisition and evaluation module is communicatively connected to the first training module and the second training module, respectively. The acquisition module and the discrimination module are communicatively connected to the first prediction module and the second prediction module, respectively. The control module is communicatively connected to the acquisition module and the discrimination module, so as to realize the production condition monitoring method described in Example 1 or Example 2.

[0066] The training subsystem can send the first and second classifiers obtained through training to the monitoring subsystem via a communication connection, or distribute them to the monitoring subsystem via storage medium copying, data writing, or hardware devices or equipment.

[0067] Optionally, a training subsystem and a monitoring subsystem can be configured on the same production line. Through prior training and parameter configuration, a stable and reliable monitoring subsystem can be gradually obtained to continuously monitor the production conditions of the production line.

[0068] Optionally, data acquisition and evaluation modules can be configured on multiple identical or similar production lines to rapidly accumulate large amounts of data for training the subsystem.

[0069] Optionally, a training subsystem and a monitoring subsystem can be configured on a production line to gradually obtain a stable and reliable monitoring subsystem for monitoring production conditions on multiple production lines.

[0070] Alternatively, a first classifier and a second classifier can be obtained through a training subsystem, and then distributed to one or more monitoring subsystems. Different monitoring subsystems can use different production condition parameter prediction algorithms, set different weights for the first and second classifiers, or use different production condition adjustment methods.

[0071] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for monitoring production conditions, characterized in that, include: The system collects and stores production condition parameters through sensors, and conducts a comprehensive benefit assessment on the products that come off the production line. Based on whether the comprehensive benefit assessment of the products comes to standard, it determines whether the corresponding production condition parameters are normal and marks them accordingly. The comprehensive benefit assessment includes an evaluation of one or more factors, such as product quality, environmental pollution risk, and energy consumption. Using normal production condition parameters, a first classifier is trained using a single-class classification machine learning method to predict whether the production condition parameters are normal. Using normal and abnormal production condition parameters and their labels, a second classifier for predicting whether the production condition parameters are normal is trained using a supervised machine learning method. The first classifier and the second classifier are used respectively to predict whether the production condition parameters are normal; the weights of the prediction results of the first classifier and the second classifier are preset; the prediction results of the production condition parameters are obtained based on the prediction results and weights of the first classifier and the second classifier. If the predicted results of the production condition parameters are abnormal, the normal production condition parameters with the lowest adjustment cost compared with the current production condition parameters are searched within the preset range of production condition parameter changes according to the pre-set production condition adjustment cost calculation method, and this is used as the target production condition parameter to adjust the production conditions. Otherwise, the current production conditions will not be changed; It also includes: using data accumulated during the production process to continue training and updating the first and second classifiers, and updating the weights of the prediction results of the first and second classifiers; As the amount of abnormal production condition parameter data accumulated during the production process increases, the weight of the prediction result of the first classifier is reduced, and the weight of the prediction result of the second classifier is increased.

2. The production condition monitoring method according to claim 1, characterized in that: The initial weight value of the first classifier is 1, and the initial weight value of the second classifier is 0.

3. The production condition monitoring method according to claim 1, characterized in that: The aforementioned pre-set production condition adjustment cost calculation method includes a method for calculating the resources, costs, or time required to adjust the production conditions from a state where the production condition parameters are at the initial parameter values ​​to a state where the production condition parameters are at the target parameter values, or a combination thereof.

4. The production condition monitoring method according to claim 1, characterized in that: The preset range of production condition parameters includes a predefined set of normal production condition parameters.

5. The production condition monitoring method according to claim 1, characterized in that: The preset range of production condition parameters includes the range of variation of the preset production condition parameters above and below the current production condition parameters.

6. The production condition monitoring method according to claim 1, characterized in that, The aforementioned adjustment of production conditions includes: adjusting the production conditions by changing or setting the set values ​​of the production line and peripheral equipment according to the target production condition parameters.

7. The production condition monitoring method according to claim 1, characterized in that: The production condition parameters include one or more of the following: temperature, humidity, sound, light, pressure, displacement, speed, acceleration, voltage, current, flow rate, particle size, and concentration.

8. A production condition monitoring system, characterized in that, It includes a training subsystem and a monitoring subsystem. The training subsystem includes a data acquisition and evaluation module, a first training module, and a second training module, used to train a first classifier and a second classifier using production condition parameters and comprehensive product benefit evaluation data. The monitoring subsystem includes a data acquisition module, a first prediction module, a second prediction module, a discrimination module, and a control module, used to monitor and control production conditions using the first and second classifiers. The data acquisition and evaluation module is used to collect production condition parameters through sensors and evaluate whether the comprehensive benefit assessment of the products off the production line meets the standards; based on whether the comprehensive benefit assessment of the products off the production line meets the standards, it determines whether the corresponding production condition parameters are normal and marks them; and sends the production condition parameters and corresponding markings to the first training module and the second training module. The first training module receives production condition parameters and labels sent by the acquisition and evaluation module, and trains the first classifier using the normal production condition parameters and a single-class classification machine learning method. The second training module receives production condition parameters and labels sent by the acquisition and evaluation module, and uses normal and abnormal production condition parameters and their labels to train the second classifier using supervised machine learning methods. The acquisition module is used to collect production condition parameters through sensors and send them to the control module, the first prediction module, and the second prediction module. The first prediction module is used to use the first classifier to predict whether the received production condition parameters are normal, and send the prediction result to the discrimination module. The second prediction module is used to use the second classifier to predict whether the received production condition parameters are normal, and then send the prediction results to the discrimination module. The discrimination module is used to obtain the discrimination result of the production condition parameters based on the prediction results sent by the first prediction module and the second prediction module, and based on the preset weights of the first classifier and the second classifier, and send it to the control module. The control module is used to adjust the production conditions by changing, setting, or maintaining the set values ​​of the production line and peripheral equipment based on the received production condition parameters and the judgment results of the production condition parameters. The acquisition and evaluation module is communicatively connected to the first training module and the second training module, respectively. The acquisition module and the discrimination module are communicatively connected to the first prediction module and the second prediction module, respectively. The control module is communicatively connected to the acquisition module and the discrimination module, so as to realize the production condition monitoring method according to any one of claims 1-5 and 7.

Images