In recent years, due to frequent updates of new products such as mobile phones, computers, IoT, and in-vehicle systems, the PCB industry has become increasingly diverse with a large number of small-scale products and production processes. Moreover, the parameters of production recipes change constantly as customers and products change, exacerbating the situation. This often requires senior employees to rely on their experience to adjust production parameters as best they can in a short period of time. Additionally, the complex process of PCB manufacturing makes it difficult to pinpoint the root cause of quality problems when they occur, as problems can often be attributed to multiple processes.
In light of this, many PCB companies have invested in systems such as MES, machine connectivity, AIoT, and BI on manufacturing sites to achieve transparency in factory data, real-time information, product traceability, and situational awareness. The goal is to improve quality and reduce issues such as defects through such systems, resulting in reduced overall costs, reduced work waste, and more effective material preparation.
In the early days of the PCB industry, whenever any problems arose, data would be collected and relevant quality engineering methods such as the Taguchi method and 6 Sigma would be used to identify the cause and make improvements. However, it has been observed that the quality of manufacturing personnel on the production floor is not consistent and the use of professional statistical software is not easily spread. Most importantly, due to the aging of senior employees and the difficulty of recruiting talents in recent years, the PCB industry is facing severe challenges in preserving, replicating, and spreading its own professional knowledge. However, in recent years, the potential for these problems to gradually improve has arisen. Due to the smart manufacturing trend mentioned earlier, many companies have introduced various systems and already have a large amount of production data. Given this particular situation, many companies are paying more attention to the development of artificial intelligence (AI), hoping that these production data can be extracted and intelligently applied using this technology, enabling companies to improve and profit. However, on the first mile to AI, the first challenge is the difficulty of recruiting AI professionals to the market. This has led to the problem of a shortage of AI talent in the PCB industry, and many PCB businesses have been restricted as a result.
Recognizing the pain points in the industry, Profet AI has created a 'virtual AI data scientist' for businesses that operates 24/7 without interruption.
In response to the demand for AI applications and talent in the manufacturing industry, Profet AI is dedicated to developing an AI platform for that industry. Its Automated Machine Learning (AutoML) platform aims to solve the problems of AI application and implementation for enterprises. Profet AI's developed platform is like a virtual data scientist for enterprises, allowing production experts to quickly conduct AI analysis with past production history data without having to write code or study complex algorithms. During the production process, Profet AI's tools assist R&D departments in optimizing product development, quality experts in identifying causes of abnormalities through tools, and process experts in quickly finding the optimal parameters to improve the production process. Through innovative AI machine learning techniques, the platform simplifies complicated algorithms into user-friendly data analysis tools and quickly establishes a standardized smart decision system, helping factories further expand production scales and improve production and product development efficiency. Meanwhile, Profet AI's platform is not only applied in the quality prediction of PCB production, but also helps various industries through its products, allowing 80% of key personnel in various fields to independently use AI, fostering a culture of using data to assist decision-making in the enterprise.
Some PCB manufacturers with ambitions to enhance product value-added or to expand the factory scale have started to introduce the innovative Profet AI "AutoML platform." It acts like a "virtual data scientist" and is an AI data analysis tool that makes operations simpler and more intuitive through visuals and graphics. It quickly empowers business experts with the ability to use AI by simply understanding their data. From uploading to completing analysis, only a few steps are required, allowing traditional manufacturing to quickly adopt AI and easily apply it in applications such as real-time quality prediction and production parameter optimization for PCBs, for example, as shown in Figure 1.
(Figure 1) Profet AI's "AI Automated Machine Learning (AutoML) Platform" acts like a virtual data scientist for enterprises.
Taking the PCB industry's Electroless Nickel Immersion Gold (ENIG) process as an example, it is a relatively stable production process, but there are still two issues that are expected to be solved through AI: Firstly, the process uses sampling inspection, and the laboratory's gold thickness measurement equipment is relatively expensive and the inspection time is longer, so it is not possible to carry out full-line inspections online. Therefore, during mass production, there is no guarantee that problematic batches will be detected; Secondly, gold is a precious metal and relatively expensive, and enterprises hope to reduce the use of gold and still meet customers' specifications.
Engineers generally adjust the relevant times, temperatures, and concentrations through experience-based rules, but too many factors such as product types, production lines, board sizes, and plating areas will lead on-site personnel to prioritize higher pass rates and tend to set up processes that are compliant but have higher gold thicknesses, which will result in significantly increased costs in the long run. After confirming the customer's clear requirements, Profet AI begins the relevant steps to assist the customer in solving problems and applying AI:
Step 1: Consolidating Production Data
The customer begins to organize production data from the past year in ENIG, including product batch numbers with inspection data (Y), product specification information, production settings and real-time parameters (nickel tank/gold tank), gold thickness values (target Y), etc. According to the logic of Y=F(X), they are structured and arranged into an Excel or CSV table format, and the structural data format is shown in Figure 2
(Figure 2) Structural data format according to the logic of Y=F(X)
Step 2: Building the AI Predictive Model
Upload the data table to the Prophet AI platform. The system will first perform data pre-processing, allowing the user to confirm the data quality again and understand the relevance of each feature value to the result Y. Then the data set will be automatically modeled after the personnel simply confirm the modeling content and plan settings and start the modeling, activating fully automatic AI machine learning modeling operations.
Step 3: Cause Analysis, Simulation Prediction, Virtual Inspection, and Parameter Optimization Recommendation (as shown in Figure 3)
l Cause Analysis:
After the model is built on the Profet AI platform, the best AI prediction model can be identified through model scores. From the prediction model, users can immediately analyze the relevant factors affecting the gold thickness and quickly provide feedback based on the ranking for quality improvement decision-making reference. In actual applications, process engineers can easily use the platform and quickly find causes even without a deep statistical background.
II Simulation & Prediction:
Based on the prediction model established from historical data, the engineers in the field can perform real-time parameter adjustment simulations on the platform. When the gold thickness as measured is too high or too low, platform simulation can be used by inputting values such as speed and temperature, and the system will automatically predict the gold thickness value after adjusting the parameters. This way, engineers can perform pre-adjustment simulations without having to go through physical experiments as in the past, reducing the time and costs associated with manual adjustment and experimentation.
III Virtual Inspection:
After building the predictive model through the Profet AI platform, it can be deployed offline on the production floor for virtual inspection. The scenario changes from a situation where only sampling inspections can be carried out on the manufacturing site to one where the AI system virtually monitors the quality of each batch of thickness, and when the predicted thickness is higher or lower than the specification value, quality engineers can inspect the product immediately, turning quality control from passive to active and reducing the proportion of actual customer complaints by reducing defective products.
IV Parameter Optimization Recommendation:
When the predictive model is accurate, the Profet AI platform can allow users to set the expected gold thickness value and have the system recommend the best optimization parameters for related processes. The scenario is in electroless nickel immersion gold (ENIG), where users/engineers expect the average gold thickness to be within a range that falls below the specification but still within the specification when shipping to customers, so that the overall cost of precious metals in the process can continue to decrease. Therefore, in the system, manufacturing engineers can set the expected gold thickness target first, and then, based on different production lines, product sizes, gold plating area, and so forth, the platform can recommend the corresponding nickel and gold slot setting values. This application plan systematizes the senior engineer's experience and integrates it into the ENIG expert system. Subsequently, even if new employees have a quick start problem or there is a manpower problem affecting overseas expansion, they can all apply and improve. Many companies have already put this scheme into practice.
(Figure 3) Platform Recommendation of Optimal Process Setting Parameters
Profet AI has changed the traditional operating mode of the PCB industry in terms of process and quality control, effectively transforming the experience of senior engineers into expert systems, making it more convenient and more efficient for users to utilize and identify the root causes of problems for continuous improvement. However, the most important thing is that many user companies have found that when a large amount of production data from different processes can generate more value through the platform, a data-driven and continuous improvement culture has gradually been established within the company and a habit of collecting data assets has been cultivated. When this culture is deeply rooted, the company will be able to improve both tangible and intangible benefits and increase profits.
More than 100 companies, PCB manufacturers and beyond, have used the Profet AI platform and the practical methodologies of one-day on-boarding and one-week landing, including a majority of leading companies in fields such as IC packaging, optoelectronics, electronics