Photovoltaic enterprise quality improvement tool: interactive visual data analysis - Power Circuit - Circuit Diagram

Along with the outbreak and escalation of the Japanese nuclear accident, the decline of the nuclear power industry, which originally occupied the largest share of new energy, is the trend of the times. Is another important new energy industry, the photovoltaic industry, back to have better development space?

China's photovoltaic market with opportunities and crisis

In this regard, industry experts have a more in-depth view: Although the nuclear power industry is not as good as before, but the photovoltaic industry is also undergoing a severe test. Since China became the world's largest producer of photovoltaic products in 2007, all aspects are gradually improving. However, it is worth noting that it is necessary to convert the government-subsidized drivers that are currently prevalent in the market into market-driven. Because relying entirely on government subsidies as a driving force for industry development has great risks. Once the policy changes or terminates, the industrial impact is huge. Only by establishing a real PV market system as soon as possible, so that the photovoltaic industry can operate according to the market rules, can we fully leverage the government subsidies and reduce the damage that policy changes may cause to the industry.

All practitioners in the photovoltaic industry (especially the leadership) must have such a clear understanding: regardless of the industry boom, the price of products in all aspects of the industry must continue to decline until the cost of photovoltaic power generation is actually lower than that of thermal power generation. Only the cost reduction is the real driving force for the development of the photovoltaic industry.

So how can we reduce costs? Many people often think directly about reducing the cost of raw materials such as polysilicon. In fact, after 2009, polysilicon costs have consumed a large part of the downside. New and real competition has begun to develop at the level of the company's process and quality management. Similar to the traditional semiconductor electronics industry, the one-off yield of photovoltaic products determines the strength of the production process, but also Determine the size of production costs.

Drawing on the experience of foreign famous photovoltaic companies such as First Solar, Ege Solar, and JA Solar, it is not difficult to find that the yield control of PV products cannot be separated from the refined analysis and decision-making based on data. This is a technical activity, and it is precisely the weakness of domestic PV companies. Of course, the quantitative analysis of advanced manufacturing processes is not blindly pursuing the complexity of data warehousing, the complexity of data modeling, but more and more emphasis on practicality. In recent years, a new way of data analysis - interactive visual data analysis - has been used by more and more leading companies to identify inferior costs and analyze their causes, develop improved solutions to optimize production processes, and ultimately improve one-off Yield and overall quality while reducing costs. The so-called interactive visual data analysis is to promote the use of more graphical tools to achieve two-way interaction between analysts and data through interactive graphics, so as to get rid of the dependence on traditional statistical analysis tools (or a graphical The simple way to use the advanced statistical analysis methods behind it) reduces the threshold for its use and provides better insight into the important information behind the data. Below we use a professional six sigma and quality management statistical analysis software JMP, which has been widely used in the photovoltaic industry, to improve the one-time yield improvement of solar cells, and briefly explain how to improve productivity and reduce productivity with interactive visual analysis. Inferior cost.
The process of simple solar cell production is basically divided into eight main steps, which can be roughly represented by the following flow chart.

The overall overall process of a simple solar cell

First, after collecting historical data, we hope to quickly find the best breakthrough for improvement and see which processes have the greatest impact on improving the overall straight-through rate. From JMP's predictive descriptors, we can see some clues: on the linear relationship between the actual yield of the eight steps and the overall actual pass-through rate, we clearly find the actual diffusion yield and the actual etching yield and the actual straight-through rate. The slope is significantly larger than the actual yield of other steps, and further changes the actual diffusion yield and the actual etching yield with the mouse, resulting in the largest change in the actual through rate, which can intuitively determine the diffusion and etching process. The most influential, they are the key processes in the production process.

Predictive morph analyzer analysis

Then, we naturally want to understand who is responsible for the proliferation of low etch yields. It is easy to find out by JMP's Pareto diagram that the main defects in the diffusion process are diffusion rework and diffusion of the internal debris. The main defects in the etching process are etching the fragments during the loading and etching the fragments in the machine. Therefore, we need to focus on and control the causes of these four defects.

Pareto graph analysis

Then, we began to use some of the more advanced analysis tools such as regression modeling, decision trees to find the cause. Of course, when it comes to advanced analysis tools, many people may be discouraged and feel too abstract and unintelligible. In fact, this kind of worry is really worrying now. In JMP, all analysis tools, whether simple or complex, can be viewed with engineers through various statistical graphs. For example, in this example, when we use the decision tree function of JMP software to perform factor analysis, we only need to click the split button on the analysis interface with the mouse to minimize the difference within the group step by step, and maximize the difference between groups. Principles group data to find a lot of valuable information:

1. Because in the process of grouping, the decision tree selects four variables of silicon wafer manufacturer, date, wafer lot and shift from many candidate variables, so it can be judged that they are the key factors affecting the defect rate.

2. If you pay attention to the factor of silicon wafer manufacturers, it is not difficult to find that the quality problems of the two factories are not big, but the quality problems of the other two factories are relatively serious. This is because from the graph on the left side of the decision tree, the green area representing the high yield rate is large, and its corresponding classification level is the silicon wafer manufacturer (A, D); otherwise, from the graph on the right side of the decision tree, it represents good products. The red area with a low rate is very large, and its corresponding classification level is the silicon wafer manufacturer (B, C).

3. In view of the date factor, the situation on November 1 was very bad. The yields of all the four silicon wafer manufacturers were not up to standard; and the two days on November 2 and 7 also had serious problems. All production batches of Class B of Class B are not up to standard, which is worthy of our in-depth on-site investigation.

Decision tree analysis

In fact, in actual work, the technicians of solar cell companies can use more interactive and visual data analysis methods to dig deep technical reasons and optimize the improvement plan. Due to space limitations, no further in-depth introduction will be made. It is. However, what is certain is: how to quickly achieve the dual improvement of quality management and cost control, mastering interactive visual data analysis will be an important issue for domestic PV companies.