Artificial intelligence and quality management – opportunities and challenges

Many experts and entrepreneurs have high hopes for artificial intelligence (AI) processes. The demands on production and service companies to improve the quality of their products by eliciting process knowledge will increase, according to these experts. AI with the Deep Learning or Machine Learning processes offers great advantages here. What impact does this have on quality management?

The decision-makers in companies have clear ideas about where AI can bring the greatest benefit in the future. Automation, predictions for optimising maintenance and interaction between humans and machines are the most promising segments, they say. These are the results of a survey of about 1,400 IT managers by the software company IFS. In the DACH region (Germany, Austria, Switzerland), quality management was at the top of the list of AI application areas with 22 per cent. Data analytics and business intelligence followed with 18 and 16 per cent respectively.

Guarantee a clean system transition

Why is AI so attractive in quality management in particular? AI can deliver results quickly for quality management, is one possible answer. However, the statements made by AI must be reliable and trustworthy, especially in quality management. Currently, companies usually use the knowledge and experience of employees when analysing process data for quality management. AI could provide other, additional insights here. When implementing AI, the results of the system in particular must be closely examined. In a transitional step, the results provided by the AI should be compared with the results of the previous system.

Core topic machine learning

AI tries to emulate how humans solve problems through different approaches. In this method, an algorithm is provided with training data in order to independently find solutions to unknown problems. Here, the algorithm does not rely on rules from the operator but uses abstract solutions that it has independently developed during its learning phase. Artificial neural networks are part of machine learning, they work in a comparable way to the human brain. The majority of exposed representatives of AI use the method of artificial neural networks. Well-known representatives of this are, for example, Microsoft's image recognition, which already worked more reliably than humans in 2015, or Google's Deep Mind algorithm, which defeated the then Go champion Lee Sedol a year later.

Field of application Statistical process control

AI methods can also make a significant contribution to making production more reliable in QM. AI-based software is able to generate new insights from already known information. These solutions are able to replace existing processes for processing information in QM with more effective and efficient procedures. Statistical process control (SPC), for example, can benefit from AI in QM. In this process, defined variables such as the number of components produced on a machine are documented over the duration of production. The data collected allows conclusions to be drawn about the quality of the manufacturing process. This makes it possible to implement an early warning system when conspicuous data or critical trends are observed.

Statistical expertise required

In order to enable an interpretation of the measurement data, their relevance for the production process must be evaluated. However, this requires information on how the desired distribution of the measured values should look. The user defines the applicable strategy for this, after which the algorithm starts hypothesis tests. However, this procedure requires that the operators have expertise in statistical correlations and the process to be monitored. By using AI, however, the procedure can be made easier. A purposefully trained AI can be very helpful here. The AI can largely replace the expert knowledge that was originally indispensable and significantly simplify the handling with regard to the regulation of processes. In this case, the monitoring of the process becomes more reliable and easier to use. In addition, monitoring can be faster, especially with more extensive process data sets.

Make decision-making transparent

Despite all the advantages of AI, there is one disadvantage whose importance should not be underestimated. With algorithms that work on the basis of artificial neural networks, the course of the process is not transparent. The system produces a result, but the way in which the AI has determined it is not apparent. The process of the AI's decision-making is hardly or not at all comprehensible. For this reason, these methods are also called black-box models. In contrast, in white-box models it is known how the decision-making process takes place. Either the software developers disclose the programming or the user defines the decision rules. AI software providers try to alleviate this transparency problem in different ways. In one approach, they run both systems, the new AI-based one and the old one, in parallel. If the results of the two processes are almost identical over a longer period of time, this builds trust among users. In addition, the AI software may also be able to calculate the quality of its proposed solutions. This can also be compared with the results of the current system.