The design method of maximum fire test and its app

2022-07-31
  • Detail

Experimental design method and its application in China with the deepening of reform and opening up, the western advanced civilization represented by market economy and its methodology are more and more accepted by domestic enterprises. In the fields of quality management, product (medicine, chemical products, food, high-tech products, national defense, etc.) R & D, process improvement and other fields, statistical methods have increasingly become the standard configuration for enterprise operation

as a relatively complex and advanced statistical method in the field of quality management, experimental design has also begun to be gradually accepted and popularized in China. In fact, the experimental design is no stranger to our academic circles. For example, uniform design was initiated by Professor Fang Kaitai (left in the figure below) and Wang Yuan, an academician of the Chinese Academy of Sciences. It is an effective test technology for dealing with multi-factor and multi-level test design. It can complete the development and research of complex scientific research topics with less test times

in recent years, mathematics departments and statistics departments of some domestic universities have gradually started to offer special experiment design courses, such as Tsinghua University, University of Electronic Science and technology, Fudan University and other universities. Some domestic industry-leading enterprises, such as Sinopec, Huawei Technology, PetroChina, Baosteel and other enterprises, have also begun to adopt DOE methods in quality management, product research and development, process improvement and other fields

although DOE is more and more accepted by domestic production, learning and research institutions that do not consider the temperature requirements of the mechanics room, we still see that the domestic research and promotion of DOE is still relatively shallow. Taking the above-mentioned enterprises as an example, Sinopec's subordinate Petrochemical Research Institute and Shanghai Petrochemical Research Institute should be the ace units in China's petrochemical research field. However, whether it is Beijing's Academy of Sciences or Shanghai Petrochemical Research Institute, the use of DOE in oil R & D, process improvement, quality management and other fields only stays at the level of partial factors and orthogonal design

at present, the high-end experimental design methods commonly used in the industry, such as customized design, screening design, space filling design and other advanced experimental design methods (Advanced DOE), are still relatively rare in both domestic statistical teaching, scientific research and industrial applications, but have gradually expanded

the application of DOE by western enterprises has already started on a large scale. For example, Georgia aerospace design center, the top unit of American aerospace and aviation design, used customer design without exception when developing top secret weapon systems of the United States, such as missiles and fighter planes. In the civil field, such as Intel, HP, apple and other companies, advanced experimental design methods are used in the product R & D and quality improvement stages

according to the development history and application sequence of DOE method, we briefly introduce what is advanced test design method and its corresponding traditional test design method. As shown in the figure below, traditional DOE includes partial factor design, full factor design, response surface design, extended design, mixture design and Taguchi design. Correspondingly, advanced DOE mainly includes space filling, nonlinear and customized design

at present, there are few professional software tools that can realize DOE (experimental design), among which the most authoritative is the desktop statistical analysis software JMP from SAS group, the world's largest statistical software supplier. First of all, the DOE content of JMP is the most complete. In addition to traditional DOE such as partial factor, complete factor, response surface design, extended design, mixture design and Taguchi design, it also includes advanced DOE such as space filling, nonlinear and customized design. Secondly, the DOE function of JMP is the most powerful. In addition to integrating the traditional statistical modeling, graphic display and other analysis methods, it also integrates unique methods such as simulation, i-best and D-Best comparison, and simple data mining to strengthen the analysis effect. Thirdly, the DOE implementation of JMP is the most convenient. The number of factors, levels, test times, etc. can be customized. Users can build tests according to the requirements of actual problems without any modification. Furthermore, the test design of JMP also integrates the simulator function. Users can directly simulate the new scheme obtained from the test to minimize the risk of failure

there are many materials about traditional doe, so I won't go into details here. The author still focuses on the more efficient advanced doe

first, let's introduce the space filling design, which is suitable for the test occasions where there is no random error but the control system deviation is emphasized. As we all know, the three concepts of randomization, blocking and replication are the basic principles that we repeatedly emphasize when doing routine trial design. However, when we pay much less attention to random errors than to the model itself, emphasizing the above three principles will not make the best use of available resources. At this time, we should pay attention to the systematic deviation, that is, the difference between the approximate model and the real mathematical function. The goal of space filling design is to limit the system deviation. The size of the system deviation is closely related to the representativeness of the test points. Through the methods of ball filling design, uniform design and minimum potential, the test scheme of space filling design can obtain the best coverage, thus laying a foundation for obtaining the most informative test result data to provide decision support

secondly, let's talk about nonlinear design, which is applicable to models that require high-precision construction of parameters that are nonlinear. In some engineering technology and social science experiment design fields, we often encounter the research problem of nonlinear model. Due to the complexity and particularity of non-linear analysis, many people will use polynomial model to approximate description and simplify the problem. However, when we have high requirements for the application of the model, the above-mentioned processing methods are inadequate. In fact, the theory of nonlinear design and modeling has gradually matured. Newton iterative method and other technologies allow users to generate nonlinear optimal design and optimal extended data, so as to fit the model with nonlinear parameters. Compared with the standard polynomial model, when the LFT model is used to describe the corresponding process, it can produce more accurate process behavior prediction, that is to say, the model is better consistent with the actual problems

what is worth mentioning is the custom design. Its flexible and convenient design style and common and consistent analysis mode have brightened the eyes of many people who have been defeated repeatedly in the traditional DOE field and greatly increased their confidence. For ordinary non statistical professionals, just hearing a lot of professional terms such as response surface, mixing materials, Latin hypercube and so on is like falling clouds, and the following analysis reports are even more like a heavenly book. This situation makes people shy away from the experimental design because the proportion of plastic packaging in the total output value of the packaging industry in the market has exceeded 30%, and the experimental design restores the essence of the experimental design in the most accessible language, As long as you define the input factors and output responses of the product or process you are studying, the current test budget, and the focus and purpose of the analysis, the custom design generator will design the test plan that best meets your requirements. In addition, the actual test data and specific analysis results, such as model formula and optimal level combination, are vividly displayed in front of you. 1. First of all, daily maintenance breaks through the limitation of traditional DOE that is "rigid and professional", and is affectionately called "DOE that can be tailored" by many engineers in European and American enterprises

let's give a simple example to experience the unique charm of customized design. For example, in the experimental design of a market research, you want to understand the psychological preferences of the target customers. The functional factor level includes the worst (1), medium (2) and best (3), while the price factor level includes high (1), medium (2) and low (3). According to the permutation and combination method, there may be 9 horizontal combinations, but in fact, in this example, the company may not be willing to sell products with the best performance at the lowest price in the market, so you need to exclude the combination of the best (3) in the function and the low (3) in the price when making the test plan. At this time, the traditional DOE (whether full factor design or partial factor design) can not realize the customized factor level constraint. The resulting test plan lacks practical significance, and the resulting analysis results lack credibility. However, the customized design can provide the flexibility to reject the combination of specific factor levels, and cleverly solve such problems that have plagued DOE users for a long time

the above methods can be implemented by professional software JMP, so as to further improve the work efficiency of using doe. Interested readers may give it a try. (end)

Copyright © 2011 JIN SHI