In recent years, the influence and application of Industry 4.0 and big data on manufacturing have grown ever more widespread and far-reaching. For factory quality control, let me give a few application examples of the concrete items affected, drawn from my own experience.
1. Full AOI inspection at a certain station (usually IQC or OQC), combined with using SPC to determine how to handle inspection of the other processes across the entire plant. This type of application has several characteristics:

A. The data volume at this station is usually very large. For example, while other stations sample five units per batch, this station inspects every unit (25,000 per batch, with 10 batches produced per day, meaning 250,000 pcs/day); over a month or even a year, the data volume is enormous.
Because it is full inspection, SPC statistical theory does not really apply, yet from a plant-wide quality perspective the data still needs to be consolidated together, which is inefficient and prone to interference.

B. Besides wanting to see the raw data, the customer also wants to see trends such as the Max, Min, Avg, PPK, and Sigma of each batch. But in reality, what truly needs to be examined is not the raw data (as proven over these years with many customers, what truly needs analysis is the raw data of the batches where anomalies occurred).

C. Once a control item with an anomaly is caught, they want to know which process parameters went wrong, i.e., P to Q.
In my actual consulting career, there are already standard solutions and relevant tools that can be applied to solve these problems.

For example: for each fully inspected batch, if only inspection analysis is needed, use the preliminary raw-data handling within SPC to process each batch into PPK, MAX, MIN, AVG, and so on for filtering, then build a trend chart as described above, which can also be combined with the sampling data of other stations. If you need to see the raw-data distribution, you only need to separately pull the original data of a particular batch and then analyze it (usually the data distribution of batches with anomalies).

If you want to perform process-parameter analysis, you must first collect the process stations preceding that inspection station along with the parameters most correlated with it (these parameters can have their correlation detected with scientific engineering methods, and the most highly correlated items can be grouped through cluster analysis).
After clustering, you can keep collecting the relevant data and verify these correlated P and Q relationships, then use mathematical models to analyze and further predict whether the P-to-Q engineering model can determine the optimal combination of P process parameters to achieve the best balance of both quality and yield. This can already be done in real cases; anyone interested may leave their contact information and arrange a separate time to discuss.

 
2. You already have (multiple) process parameters and inspection results, and a general understanding of their correlation. You now have automated machines that can output multiple process parameters, as well as automated machines that can collect quality results. The customer demands continuous refinement but does not know how to do it best. For example, Tai-X-Electronics requires its supplier’s coating thickness specification to be between 100-110 um (a generally acceptable coating THICKness is 80-140, but TSXC offers double the usual margin). The company produces 10,000 units, of which only 1,500 fall within 100-110; Tai-X-Electronics wants 7,000, while other customers together want only 3,000. Should you produce or not? If you produce, there will be many unsellable units; if you do not, the profitable order disappears.

Solution: Use the AIQ system to establish the correlation and mathematical model between the P parameters and the Q results, then collect a period of real data to verify and adjust the mathematical model. This AI mathematical model can self-correct, and as the data accumulates, it can grow on its own using neural networks and artificial-intelligence models. Once a suitable mathematical model is built, the system can recommend which combinations of P to use in order to produce the maximum quantity of products within the 100-110 specification. In this way, you obtain the optimal solution, and such scientific data can be replicated and continuously reused.

 
Everyone can give some thought to which possible application needs we can refine in our use of factory big data. The two examples just given are merely a starting point to inspire further ideas. In 2023, there is a free online AIQ seminar every quarter; if you are interested, you are welcome to register and join, leaving your questions so we can discuss and improve together!