First, the monitoring method of process state stability
Judging whether a process is in a stable state can be judged by the control chart [1], which is why European and American enterprises generally say that they control a stable process. Control chart discrimination is based on the rule that small probability events are not easy to occur, and if small probability events are observed, the process is considered to be affected by special factors. However, when using the control chart, we must pay attention to its conditions of use. In fact, the use of control charts must meet the following three conditions:
(1) data are independent of each other;
(2) The data should generally obey the normal distribution;
For example, we often use P-chart, NP-chart, U-chart and C-chart in the counting control chart, in which P-chart and NP-chart are usually suitable for product inspection results (qualified and unqualified) and generally meet binomial distribution, while U-chart and C-chart are usually suitable for counting the number of defect points per unit area or volume, such as the number of crystal points on the film, the number of defects on the casting and the number of bubbles on the glass. , generally obey Poisson distribution, and these two types of distribution only.
(3) There is only random error in process variation.
If the data does not meet the above three conditions, the control chart can be used to monitor the process only after the data is processed. For example, if the data distribution is not normal, the abnormal distribution can be transformed into normal distribution through appropriate transformation. The more standardized and general method here is to use Box-Cox method [3] for conversion.
Second, how to effectively eliminate the process variation in the industrial field?
In the industrial field, the stability of production process has always been regarded as an important goal of process control. However, in practice, due to the influence of many factors, it is not easy to make the process stable and realize optimization on this basis. Especially in fine chemical industry, material industry, pharmaceutical intermediates and other industries, its production process is usually a complex and multi-step dynamic process, which generally has the characteristics of small production batch, many varieties switching and complex and changeable process parameters, so it poses great challenges to optimize production efficiency, product yield and quality stability.
In the actual process improvement, we usually adopt the following analysis ideas for process anomalies:
1) Review and analyze the rationality of the whole process;
2) Through team brainstorming, analyze the problem from human, machine, material, method, environment, measurement and other factors (5M 1E) (as shown in figure1);
3) According to the understanding of process mechanism and combined with experience, make diagnosis and investigation.
Especially in the analysis of 5M 1E, many factors will be involved, which is also the main source of many difficulties and pain points in chemical process control. For example, different people have different understanding of the process and work habits, and the operation process may be thousands of people. Even in some enterprises with frequent batch switching and various raw materials, sometimes there are things such as feeding errors and inaccurate feeding; Due to the failure of effective preventive maintenance of production equipment, equipment anomalies often occur, resulting in a large number of production waste and quality accidents; After the quality abnormality occurs, due to the lack of process data, it is impossible to trace the quality abnormality effectively, let alone find out the real reason and form effective preventive measures; There are many factors (input variables) such as process improvement, yield improvement and formula optimization. However, design optimization by experiment (DOE) often leads to the omission of important factors due to screening factors or a large number of experiments, which affects the promotion efficiency of the overall optimization work.
There are also many chemical enterprises aiming at these problems, through the lean Six Sigma project, expecting to achieve the purpose of eliminating waste, reducing costs and increasing efficiency, and continuous improvement. As we all know, lean six sigma method is a real problem-solving and continuous improvement method based on data and statistics. The traditional lean six sigma project process is to first find and determine the problem points through lean methods, and then set the improvement goals according to the DMAIC method of six sigma, analyze the reasons, find the optimization scheme and solidify the improvement measures. Theoretically, this process is very rigorous and pragmatic. However, in practice, the sustainability of each lean six-sigma project is often less than expected, and the project effect generally reaches its peak in the project acceptance stage, and then gradually falls back. The reason is that although every step of Six Sigma project is based on data and statistical methods, it is an offline method, and the project measures and effects cannot be solidified and tracked online. After the acceptance of the project, in the solidification stage, the attention of project stakeholders gradually decreased, and the effect naturally declined. In addition, there are also some Six Sigma projects, which cost a lot of time and energy, but ultimately failed. The reason is mainly caused by two problems:
1) manual collection of factory data has some shortcomings, such as untimely and inaccurate, which leads to little improvement work;
2) Even with rich and high-quality real-time data, the analysis method relies on traditional statistical analysis methods, and it is difficult to dig deep into high-dimensional massive data and identify the important laws contained therein.
In order to cope with these difficulties, more enterprises began to choose digital transformation and build smarter intelligent manufacturing plants. By transforming the equipment into the Internet of Things, they built a digital twin system at the bottom, combining production execution management system, laboratory information management system, research and development information management system and so on. , so as to make the execution of each process more standardized and realize the maximum traceability of the whole process. Further, with the help of powerful artificial intelligence algorithm, we continue to deeply analyze and excavate the variation existing in the process, find out the important reasons for the variation, and improve it, so that the overall operation of the factory can achieve the goal of intelligence.
References:
[1]. Statistical process control? Reference manual? AIAG
[2]. Simple Statistics, written by Dawn Griffith.
[3]. "Six management statistics guide", waiting for the horse.
On the intelligence of national engineering;
Guogong Intelligence is a state-owned joint-stock high-tech enterprise specializing in providing the overall solution of artificial intelligence decision control and landing service for the process manufacturing industry. It focuses on using artificial intelligence, big data and other technologies to solve the intelligent manufacturing needs of complex scenes of the process manufacturing industry under massive data, and provides customers with the overall solution of "IOT+AI+OR" intelligent manufacturing artificial intelligence. At present, the company has become a leader in the field of artificial intelligence decision control in process manufacturing.
As a professional intelligent manufacturing service provider in China, SINIC has independently developed data brain analysis platform (MAI), intelligent manufacturing management platform (MES), Internet of Things data acquisition platform (SCADA), laboratory management system (LIMS) and dual-system equipment management system (EMS) based on artificial intelligence, which have been successfully applied in the industry.
Guogong Intelligent has been deeply involved in fine chemicals, materials, medicine, food, feed, agriculture and animal husbandry for a long time, with customers all over the country. We have successfully provided intelligent manufacturing services to customers such as Xi Anruilian, Haida Group, China Resources Sanjiu Pharmaceutical, Kangyuan Pharmaceutical, Fengyuan Group, Liming Group, Jiumu Chemical, Lan Fan Medical, New Age Health Industry Group and enron nano Group.
Guogong Intelligent adheres to the enterprise development concept of "benefiting the country and the people, being good at work", serves traditional manufacturing enterprises with high-end IT technology, promotes the transformation and upgrading of national manufacturing industry, and empowers China Zhizao with craftsman spirit! Strive to become the leader of scientific and technological innovation and industrial revolution, and contribute to the rise of China's real economy and the realization of Made in China 2025!
Common content of data brain
MAI-CLI is a data analysis platform integrating data scheduling, data cleaning, data calculation, data mining and data visualization. The system carries out human-computer interaction in a simple and easy-to-use dragging operation mode, which shields the complexity of data analysis and prediction business and greatly lowers the technical threshold of data analysis.
The whole data analysis business is built in the form of calculation flow. The platform realizes the unified scheduling of scattered data and the integration of multi-source data such as laboratory equipment, industrial sensors and information system interfaces.
The platform provides hundreds of algorithm functional components, including a large number of algorithm components such as application statistical model, machine learning model and deep learning model, and supports lean six sigma online improvement tools. The platform has the function of timing analysis, which can monitor tens of thousands of quality monitoring points at the same time, and realize the functions of self-diagnosis and self-providing solutions.