Predictive Quality

Predictive Quality: Data-based quality testing with our autfactory software

100% final inspection? No need for that !

autfactory evaluates components during production—and shows early on whether a part will be OK or not OK before it reaches the final test bench. With real-time data and machine learning, you can detect deviations immediately and check only where it is really critical.

For predictive quality to function reliably on the shop floor, one thing is essential: data that AI models can actually work with. We ensure clean production data (collection, context, preparation). For the AI models, we collaborate with IconPro, specialists in AI-supported evaluations. The result: a robust predictive quality solution that works under real production conditions.

Ideal for manufacturing companies with the following characteristics:

• Variant manufacturing & cost-intensive production (many work steps, high added value per part)
• Complex manufacturing processes with many influencing factors (e.g., material properties, heat treatment, coating) and many process parameters
• High testing costs (time-consuming tests, expensive testing equipment)
• Critical quality characteristics with high risk in case of deviations
• Limited testing capacities (e.g., measuring room, EOL test bench)

AI on the shop floor: Reduce inspection effort and scrap

Artificial intelligence and deep learning on the store floor are no longer dreams of the future. We are also working intensively with these technologies and have gained valuable insights for quality assurance in practice as part of a research project. In collaboration with our AI partner, we were able to demonstrate the following after data preparation, analysis, and training of the artificial intelligence:

We know which components are NOK before they are tested on the final test bench!

Reduction of testing times through intelligent algorithms

Our systems analyze production and testing data in real time to speed up testing processes. Through the targeted use of statistical methods and machine learning, we reduce unnecessary inspections and focus on critical areas. This reduces the testing effort and increases the throughput speed.

• Faster time to market: Reduced testing times accelerate your production cycles and bring products to market faster.
• Cost reduction: Less testing means reduced machine running times and energy consumption.
• Increased production capacity: More machine time for production thanks to optimized testing processes.
• Better use of resources: Staff deployment and test benches are planned more efficiently, thereby avoiding bottlenecks.

Quality predictions for error-free production

With predictive quality, potential quality problems can be identified and avoided during the production process. Our AI models analyze historical and current production data to identify trends at an early stage and eliminate sources of error. This minimizes waste, avoids costly reworking and increases the overall quality of your production.

• Early error detection: Problems are identified and prevented in advance.
• Less waste: Production errors can be eliminated at an early stage, saving material costs.
• Stable processes: AI-based predictions help minimize fluctuations in the production process.
• Less rework: By avoiding errors, the effort required for corrective measures is eliminated.

autfactory is your AI enabler: data in the quality that AI needs

AI models are only as good as the data they learn from. autfactory creates precisely this foundation.

We bundle machine and process data, production data, quality data, planning data, and operator inputs to generate a usable, uniform database.

We deliver data in a quality and structure that AI can actually work with—as a foundation for predictive quality, anomaly detection, but also for transparency (dashboards/KPIs) and OEE optimization.

autfactory Use Case: AI-supported early detection of rejects

With autfactory, you can identify rejects where they occur—not just at the end of the line. AI-supported early detection makes deviations visible at an early stage, allowing affected parts to be sent for reworking before they tie up time, capacity, and material in further process steps. This significantly reduces inspection time and equipment – and optimizes your existing infrastructure without additional capital investment. Because at the end of the day, data quality = result quality – only with consistent, reliable shop floor data can predictive quality be truly accurate.

Use case 1 – Automotive / Tier 1: Increased output with existing EOL test equipment (without a second test bench)

Initial situation: In many production lines, end-of-line testing (EOL) is a clear bottleneck. Although upstream processes still have reserves, throughput is limited by EOL. An additional test bench would be a solution, but it is usually expensive, requires space, and involves validation costs.

Predictive quality approach: Process data from assembly, screwdriving stations, leak testing, and measuring stations are linked to existing EOL data. On this basis, a predictive quality model can be trained that evaluates the probability of a “pass/fail” even before the EOL test.

Benefits:

• Increase in output without additional EOL test bench
• Reduced waiting times and blockages at the EOL
• Faster root cause analysis because error clusters and drivers become visible
• Better use of existing testing equipment instead of additional investments
  

Use case 2 – Metal processing: Identifying complex process relationships (heat treatment, many parameters)

Initial situation: In complex manufacturing processes, quality problems often arise not from a single parameter, but from interactions. Especially in heat treatment, batch changes, material variations, or process drift, scrap and rework can fluctuate—even though all individual values are within specification.

Predictive quality approach: A comprehensive component history (traceability) is established and process and quality data are consolidated across all stations. The predictive quality model not only evaluates limit violations, but also identifies critical combinations of parameters (e.g., furnace profile + batch + quenching conditions).

Benefits:

• More stable quality despite high process complexity
• Early warnings for process drift before waste occurs
• Transparent cause-and-effect relationships instead of trial and error
• Faster root cause analysis and more targeted process optimization
  

Use case 3 – Pharmaceuticals / Filling & Packaging: Avoid the cost of 10: Detect errors earlier, before they become expensive

Initial situation: In highly automated filling and packaging lines, small deviations (e.g., sealing, pressure/temperature drift, micro-leaks, labeling errors) can cause high follow-up costs. It becomes particularly expensive when errors are detected late—e.g., after final inspection or even after batch completion.

Predictive quality approach: Inline data from sensors, camera systems, and process parameters is linked to final inspections and quality data. A predictive quality model can continuously assess the risk per batch or process phase and identify deviations at an early stage—before they affect large quantities.

Benefits:

• Earlier detection of quality problems before batch completion
• Reduction of costly rework, additional inspections, and quarantine cases
• More stable production through drift monitoring
• Faster decisions on when targeted intervention is required
  

SoliDAIR: Predictive Quality – scientifically proven, industrially tested

With SoliDAIR, we are part of an EU-funded industrial project driving the introduction of AI, data, and robotics in manufacturing—with a clear focus on reliable, scalable quality control.

In our use case for automotive transmission assembly, the classic end-of-line inspection process for 50% of products is supplemented or replaced by AI-supported quality control. The goal: to detect errors preventively, better understand their causes, and derive proactive corrective measures directly in the process—through the interaction of humans and AI.

You benefit from predictive quality that not only “builds models” but has also been further developed under real-world conditions.