How Siemens lowered its PCB assembly defect rate through linked data analysis.

The introduction of causal AI has opened a new chapter in PCB production. As part of a proof-of-concept project, the Siemens Motion Control plant in Erlangen, Germany, successfully reduced the error rate in PCB production to below the demanding Six Sigma wall. This achievement demonstrates how holistic data analysis and AI-supported root cause analysis can optimize production standards (Figure 1). As a pioneer of digital transformation, Siemens leverages causal AI to revolutionize production processes. By seamlessly linking and analyzing all data across the entire value chain, they achieve a new level of transparency and efficiency. The result: optimal product quality, cost savings, accelerated processes and a decisive competitive advantage.

Figure 1. Electronics production at Siemens’ Erlangen plant.

What is Causal AI?

Causal AI offers an innovative approach to AI that focuses on discovering cause-and-effect relationships in complex data. Unlike predictive methods that only forecast what will happen, causal AI explains why certain events occur. This enables manufacturers to identify patterns in data and gain deeper insights into their processes. As a result, they can optimize production, reduce downtime, improve product quality and increase efficiency.

Causal AI is suited to production lines where full traceability of individual parts is available. The technology is rapidly gaining momentum for Industry 4.0, as shown in the latest Rockwell Smart Manufacturing Report. In a 2024 survey of 1,500 decision-makers from 17 major manufacturing countries, causal AI ranked first in planned investments over the next 12 months. Decision-makers view it as one of the best technology solutions to support the workforce and offers an exceptional return on investment (ROI). The Siemens plant in Erlangen, a digital flagship plant, specializes in producing high-precision industrial drives for automation and motion control. At the heart of these drives are specially designed circuit boards. Production is a complex process that is prone to errors and involves several steps (Figure 2):

Figure 2. Standard PCB assembly line, with AI tied to inspection.

1. Print the PCB with solder paste using a product-specific stencil.
2. Check the quantity and position of the solder paste for each component land.
3. Place electronic components on the board (50 - 1,500 components per board).
4. Solder the circuit board using the reflow process based on product-specific temperature profiles.
5. Identify faulty component placements, solder bridges and open solder joints in the final automatic optical inspection (AOI) (Figure 3).

Figure 3. Production steps with resulting data per fiducial (reference point).

Although the PCB production process remained highly stable, the results failed to meet the required quality targets: a defect rate of less than 3.4 per million possibilities and the required Six Sigma level. This necessitated a thorough overhaul of quality control and extensive data analysis.

The Data Puzzle

PCB production collects 500 million data points every day, generating a wealth of information that often gets viewed in isolation. Until now, data-driven problem analysis along a production line requires extensive manual data engineering to integrate all data sources and create a complete baseline. As a result, a comprehensive understanding of the entire production process to detect errors and optimize manufacturing has typically been missed.

Faced with a complex data landscape in PCB production, Siemens made a shift by implementing causal AI technology. The goal was to analyze data comprehensively across all production steps to significantly reduce the error rate and achieve Six Sigma quality levels. As noted, traditional predictive models predict only future events (what will happen?). Causal AI makes it possible to identify underlying causes of production problems (why is this happening?). In healthcare, analysts examine the patient, while in manufacturing, they focus on the produced part.

This causal information related to production issues forms the basis for targeted intervention and optimized PCB manufacturing. Once teams identify the causes of desired or undesired events (or effects, they can rectify errors and reinforce positive results.

The novel Xplain Data ObjectAnalytics technology provides a comprehensive 360-degree view of each PCB, which serves as the central object of analysis. The method uses an object-centric approach, combining all relevant data into a central object model with its main and sub-objects. In PCB manufacturing, the PCB is the main object, while each component on the PCB represents a sub-object (Figure 4). For each of the 500 components, readings are collected from every stage of the production process. By integrating all data into a central object model, the analysis reveals correlations and dependencies accurately. For example, technicians can analyze design parameters, solder paste print results and inspection together to identify problems that only occur when combining these factors. Using a mouse, a technician can create a graphical analysis of the root causes of all problems that occur.

Figure 4. The object tree for PCB production. The premise allows analysis of multiple individual processes to identify problems that only occur when combining these factors.

A Breakthrough in PCB Quality

Causal AI technology was introduced in three steps: object-centric consolidation of production data, identification of causal relationships and continuous optimization. This systematic approach drastically reduced the production error rate. A significant success in PCB production was breaking the Six Sigma barrier (fewer than 3.4 defects per million possibilities). Automated analysis now makes it possible to identify potential causes of defects at an early stage of production and take targeted action.

The AOI identified previously undetectable causes of design-related failures using the novel causal AI processes. As a result, Siemens significantly improved the quality of its PCB production and broke the Six Sigma wall. Causal AI is a game changer for us in electronics production. Siemens can now fully automate quality work, especially root cause analysis.

By implementing the holistic ObjectAnalytics data model, which is based on the entire object tree, Siemens is, for the first time, able to simultaneously analyze the effects of quality in process and quality in design in a holistic way. For Siemens, this represents a breakthrough in how to improve manufacturing quality.

Next Steps

The next step involves implementing this solution in daily operations to sustainably increase the contribution to value creation.

To maintain this achieved quality, Siemens plans to implement an autonomous tool to continuously monitor the production process (process control) and provide the quality management team with early warnings of potential problems and causes that could endanger the production line. This will lay the foundation for AI-supported collaborative problem-solving in quality and process engineering. Siemens also plans to extend the integration of causal methods to assembly of the entire device.

Conducting a cost-benefit analysis was another important aspect of the project. Within a year, the investment in causal AI technology paid for itself. In addition to significantly reducing the error rate and improving production quality, it opens completely new possibilities for supporting and easing the workload of Siemens’ quality and process experts.

Erik Schwulera is lead IoT@Manufacturing and Six Sigma director, Siemens Digital Industries (siemens.com).

• Prev
• Next