AI in industry: Turbocharging productivity and quality

Artificial intelligence (AI) is no longer future vision, but a reality that offers great potential for the manufacturing industry, increasing productivity and quality while reducing costs. In this context, AI supported process digitalisation and automation play just as important a role as the use of artificial intelligence in production and quality assurance. In this Blog, Matthias Puhr, Data Scientist at DCCS, offers insights into the specific applications of AI for industrial companies and highlights factors important to making sure AI projects deliver the desired business value. 

AI: A catalyst to solve complex problems

In abstract terms, AI can be called a collection of mature technologies that identify and analyse complex interrelations in data. In industry, AI can serve to digitalise processes and automate procedures. AI methods can be used to solve various multi-faceted issues that have previously brought other approaches to their knees.

Artificial intelligence can be used wherever ample (high-quality) data is available. In industrial environments, production processes are particularly suited to the use of AI. In modern production processes, large amounts of data are collected and stored that can later be used as a basis for AI applications.

AI boosting quality

There is great potential in quality assurance, where AI can recognise complex correlations from a wide variety of data streams and where many tasks are currently still performed manually by employees.  Often times, humans are incapable of dealing with the complexity and amount of available data or derive relationships or operating instructions from it. This is where AI can support reasonable, context related pre-filtering of data or even completely automate processes.

Image recognition & anomaly detection

Visual quality assurance is a prime example. The large advances made in the field of machine vision and image recognition and analysis allow for the fully automated visual inspection of components in many applications. The use of pre-trained or open source AI models combined with advanced anomaly detection makes it possible to develop AI algorithms that learn what a component needs to look like based on training images. In the production process, camera images of components can thus be analysed and defective products can be identified reliably. The large benefit of the anomaly detection approach is that deviations from a desired target component state are detected at an early stage, eliminating the need to specify all possible defects from the start. AI can significantly contribute to boosting the efficiency of quality assurance processes and markedly lower costs.

AI: precise, reliable, fast

The big advantages of artificial intelligence are constant reliable algorithm results, the detection of patterns and relationships in large amounts of data and the capacity to make decisions quickly and in an automated way.

In the quality assurance realm, results consistency is particularly important since identical input data always needs to lead to the same decisions. Contrary to human workers, AI tools also never become tired and can complete monotonous tasks much faster than a human ever could. Thus, intelligent systems can be used to boost productivity and achieve consistently high quality.

The use of AI not only proves useful in production, but also in many other areas. AI supported digitalisation and automation of processes can also deliver value. Document classification is a common use case. Among others, this includes automated request assessment or direct forwarding of e-mails to the responsible department. Provided ample training data is available, AI can deliver success rates far in excess of 90 percent.

AI projects delivering business value  

Although the market offers many standard AI solutions for particular use cases, these are usually inflexible and can only be used for very specific applications. Complex use cases or customer-specific tasks commonly require individual tailoring and the respective expertise of AI experts capable of developing a customised, value-adding solution.

AI projects start with a comprehensive analysis of the available data and the respective processes. Together with stakeholders and domain experts, the AI experts then develop a thorough understanding of the use case and tasks in question. This is followed by data preparation and (where necessary) cleansing, transformation and homogenisation. The form of AI training data must match the used algorithms. What algorithms match the solution depends on many factors. The available infrastructure and the requirements towards accuracy and precision play an important role in this context. In practice, models that deliver slightly worse results in shorter amounts of time while using less resources, are often preferred over more accurate models that are more expensive to run.

The AI model’s deployment is another important point. One must assure that the AI is available to all potential users and, ideally, seamlessly integrated into the exisiting system landscape. In industrial environments, the integration of AI into the production process or machine controls can also be a vital success factor.
Deployment is often overlooked which leads to many AI projects never leaving the proof of concept stage.

In practice, it often makes sense to have several iterations of the above mentioned project phases since new data sources are discovered and new information needs to be incorporated as the project advances. In the course of this iterative process, the AI model is constantly improved with minimum time invested to achieve an optimum result.

Practical tips for success 

Building on several years of experience in the implementation of AI projects together with industrial customers, we were able to derive a few success factors:


Ask the right questions

Wrong questions, assignments or prerequisites lead to solutions that don't carry much value. Beginning projects with the premise of “what problems can AI solve” or “do something with my data”, is counter-productive. It has been proven useful to ask the following questions before starting the project:

  • ▪    What challenges and issues are we dealing with in processes or production steps and how can AI support us?
  • ▪    What data/information can be used to derive rules?
  • ▪    Are there any elaborate, repetitive tasks and can they be automated?
  • ▪    Where can and should one use digitalisation?


Don’t look for a do-it-all solution

Reaching a do-it-all solution is often difficult or entirely uneconomical. Most often, it doesn't make sense to strive for a do-it-all solution or try to completely replace employees.
Often, partial solutions or the automation of partial processes delivers great value, quality improvements or cost savings. You should see AI as an assistance function that supports humans and can boost productivity.

Don't underestimate the overall project

AI and the associated programme code usually only make up a small part of the overall solution. Successful integration into the customer’s system landscape and infrastructure is vital for success.

In the overall project, one needs specific technical / domain expertise for the respective project, but also expertise in the fields of software development, AI and infrastructure.

As a long-standing digitalisation partner to industrial customers, DCCS will gladly accompany you through your AI project with its comprehensive expertise and make it a success together with you.