Why machine learning
Application in intelligent manufacturing
When it comes to intelligent manufacturing, we have to talk about "machine replacement". If we use robots, automatic control equipment or assembly line automation to replace the traditional production line operators and material personnel, we can achieve the purpose of "reducing staff, increasing efficiency, improving quality and ensuring safety". The application of machine learning in the manufacturing process is to further enable the intelligent manufacturing system, Realize the ability to replace or assist managers and professionals to make decisions on uncertain business.
Dikw model brings data, information, knowledge and wisdom into a pyramid like hierarchy, showing how data is transformed into information, knowledge and wisdom step by step. After the system collects the original data, it gets the logical information through processing. Then it gets the rules and knowledge, forms the ability of action and completes the task by refining the relationship between the information. Finally, it uses the induction and synthesis of all kinds of knowledge to form the prediction ability of focusing on the future uncertain business. Only in this way can the system truly realize the perception, analysis, reasoning and decision-making Decision making and control function.
Take a simplified example to understand the above process. The system collects real-time temperature through sensors, and then associates the data with other data (such as batch, bar code, machine, raw material, product quality grade, etc.), and calculates various statistical values of temperature points in the production process. These information can be used for process control according to known knowledge (process requirements), The model can also be concluded by correlation analysis. When new suppliers' raw materials or production lines appear, the process requirements can be optimized to achieve the best production capacity and quality.
02
What are the applications of machine learning in Intelligent Manufacturing
Machine learning is a special research on how the computer simulates or realizes human learning behavior, in order to obtain new knowledge or skills, reorganize the existing knowledge structure and make it continuously improve and optimize. It is a method to improve the ability of information to knowledge extraction and knowledge induction.
According to the definition of industrial big data set in the white paper on industrial big data issued by the Ministry of industry and information technology, industrial data includes enterprise informatization data, data collected by the Internet of things and external related cross-border data, and machine learning has become one of the main methods of industrial big data analysis and mining.
Expert system and pattern recognition technology in modern manufacturing process have been widely used in visual recognition, natural language understanding, robotics and other disciplines in manufacturing system. The original expert system defines the experience and experimental data of business professionals in the form of rules, and then integrates the algorithm of mathematical programming to find the optimal solution of the problem according to the given conditions, such as dealing with multi-objective dynamic programming in scheduling; Pattern recognition is based on the characteristics that have been set, through the method of parameter setting to give the recognition model, so as to achieve the purpose of discrimination, focusing on solving the perception problems of small data changes and single business objectives, such as production signal processing, image recognition and SPC control. Machine learning can use standard algorithms, learning history samples to select and extract features to build and continuously optimize the model, so that the original system in the enterprise increases the ability of autonomous learning, solves the uncertain business in the production process, and improves the intelligent level of the system.
For example, during the implementation of the production scheduling system, the implementation consultant will confirm the rules with experienced dispatchers. For example, due to process constraints, products must be ranked in line a rather than line B. due to less switching time, products a should be ranked first and then products B should be ranked. The maximum number of production batches is 100, and the minimum number is 40, In the system, the scheduling results are obtained by mathematical programming; Machine learning first establishes the scheduling task model and measurement indicators, then extracts the characteristics that affect the measurement indicators through the main cause analysis of the final execution results of a large number of production plans, and then adjusts and optimizes the rule parameters such as the interval of production batch size with the model, and even induces new rules to set the interval of production batch size, And the learning process is continuous, which can be adjusted according to the latest characteristics, avoiding the traditional way that experts modify the rule parameters regularly.
03
How to apply machine learning in intelligent manufacturing
One way to apply machine learning to intelligent manufacturing system is to build a single system with the function of machine learning. The other way is to build an enterprise level machine learning platform to provide machine learning ability and services for other systems in the enterprise. The latter machine learning platform system architecture can be divided into data acquisition layer, source data layer, data storage layer, data storage layer Data analysis layer and application layer.
The data acquisition layer is mainly used to collect the original data needed by machine learning and provide the learning data set for the platform. According to rami model, data acquisition layer mainly comes from external system, enterprise system, factory system, workshop system, control system, field equipment and intelligent products. The external system mainly collects the upstream and downstream supply chain data and external data related to the enterprise, such as competitive product analysis data, etc; Enterprise system mainly collects enterprise orders, customer information, inventory information, etc; The factory system mainly collects master plan, equipment account, etc; Workshop system collects work order information, quality information, operation log, monitoring video, etc; The control system provides data of production process, environment and energy consumption; Collect instrument readings, start stop signals and other data from field equipment; Intelligent products produced by intelligent manufacturing can collect data of product operation and maintenance.
Manufacturing big data process - 1 with trademark
The source data layer stores the data and information obtained from the data acquisition layer, and uses the relational database to store the organized information; Real time database stores compressed time series data; Use file system to store log and video files. In addition, if the machine learning platform is needed for real-time data flow calculation, the application layer system needs to be transformed, and the data is directly sent to the message queue of the data storage layer for processing. In this part, a new path can be added to the enterprise service bus to reduce the impact on the original system.
Machine learning platform can regularly extract the data from the source data layer to the value pair database or object database of the data storage layer, while the data in the real-time database can be sent to the message queue by active push, and the files in the file system can be saved to the distributed file system by file extraction.
Manufacturing big data process - 2 with trademarks
The data analysis layer extracts the sample features from the data storage layer, and generally adopts the batch data processing method of timing trigger. For example, after a work order is finished or when the shift is changed, it gets the samples needed for machine learning, divides the samples into two parts: training set and verification set, and uses clustering, regression, neural network and other algorithms to train the model, Then the model is evaluated and the model parameters are adjusted through the validation set.
Manufacturing big data process - 4 brands included
The trained and verified models can be published. For the prediction models (recommendation model, classification, neural network) obtained by machine learning, the prediction results are mainly fed back according to the input in the scenes with high real-time requirements. Stream data processing is used to monitor message queue or file increment to get test set, and then model calculation is used to get test results, which are fed back to data application layer for use. For example, according to the real-time instrument data to recommend the best set of equipment process parameters for production, or predict the quality of abnormal sent to the control system for alarm shutdown. The application of this kind of model can also use edge computing to release the prediction model to the industrial control system of the production site or embed it in the system.
Manufacturing big data process - 3 with trademarks
04
epilogue
Machine learning has a broad application prospect in the field of intelligent manufacturing, but in the application, business analysts and data analysts need to work closely together to clarify the analysis objectives and feasibility of machine learning from the business objectives and solving practical problems. This paper introduces a feasible application architecture for manufacturing enterprises, hoping to throw a brick to attract jade, To provide ideas for practitioners in the field of intelligent manufacturing, and form the best solution for enterprises.


Previous: Common parts of automation equipment
Next: No Data