Nowadays, when the term 'algorithm' is used outside of mathematics lessons, it usually refers to a closely guarded secret: the rule according to which content is displayed to users of a digital platform. Google, Facebook, YouTube, TikTok - they all use such an algorithm. There are also countless potential applications for algorithms in the industrial environment and many are already in use. The next logical step is algorithms that use algorithms.

What is an algorithm?

Algorithms are not necessarily source codes. Basically, they are simple step-by-step instructions for completing tasks. Typical examples are cooking recipes, building instructions for furniture or model making or even route and driving directions. Algorithms achieve their full benefit when they are described in machine-readable form. This is because they then not only become a technical requirement for an automation project, but can also be directly integrated and implemented in it.

A simple example: Within a production line, a specific task is no longer to be performed manually in future, but automatically. To do this, the task to be performed must be described without interpretation and without gaps so that it can be implemented with suitable hardware and software and ultimately executed smoothly by the automation solution. It makes sense to create this description using suitable tools so that it is (also) available as source code for the automation software. Ideally, such tools also offer options for virtual testing of the requirement and its theoretical implementation.

The use of algorithms in an industrial environment is therefore by no means uncharted territory - time for the next step.

Internet platforms as a role model

Algorithms act as step-by-step instructions for processing tasks.

© shutterstock.com / Ulvur

The major internet platforms have developed selection and suggestion algorithms in order to show their users content that will keep them on the platform. In an industrial environment, specialists should always be provided with helpful content when they have reached the end of their own knowledge. This supply is particularly critical in the maintenance of technical systems.

The aim of maintenance is firstly to preserve the functionality of a system and secondly to restore it in the event of loss. The former is usually achieved through inspection and maintenance, the latter through troubleshooting and repair. As the system often has to be shut down for inspection and maintenance, costs are incurred in proportion to the duration due to production downtime. If unplanned downtimes occur despite everything, every second that can be saved in troubleshooting and repair is worth a lot of money. The better maintenance specialists are supported in their tasks, the more efficiently and, above all, reliably they can work.

This requires a selection and suggestion algorithm modeled on the Internet platforms: In the maintenance case, such an algorithm collects all available data and, if necessary, requests further information from the specialist. To collect the data, the algorithm in turn uses algorithms that are automatically executed by the system or made available to the specialist as work instructions.

Using algorithms Algorithms

Based on the data collected, the algorithm suggests options for handling the maintenance event in question. These options for action are - who would have thought it - available as algorithms that are implemented by the system itself or by a specialist. The core task of the selection and suggestion algorithm is therefore to execute the right algorithm or have it executed at the right time.

This requires three things:

• the algorithms for data collection and event handling,
• metadata for these algorithms,
• an algorithm for continuously improving the metadata.

For the algorithms for data collection and event handling, it is important that they are methodically structured, complete and created as individual, incremental sections. A proven procedure is to first create an overview of all components of a technical system that may be subject to maintenance measures. Next, the maintenance events that occur here and the symptoms that can be used to identify them must be stored for each previously determined component of the overall system. Finally, incremental algorithms must be defined that serve one of two purposes: Either they generate data and information to infer the causal event from the symptom, or they serve to treat the event.

The metadata for these incremental algorithms serve as a working basis for the selection and suggestion algorithm. On the one hand, this involves specifying which information is generated by the algorithm. The probability that a certain event will occur can be determined on the basis of this information. The algorithm will therefore suggest algorithms that further narrow down the existing event by excluding many events or verifying a specific event. In addition, metadata on the duration of the execution of an algorithm should be stored. If one of five possible events can be identified with certainty within seven minutes, that is good. However, if four of the events can be excluded with certainty within one minute, that is better. Depending on the specific application, further metadata may be useful.

The author: Johannes H. Diedrich is Head of Industrial Projects at Synostik.

© Synostics

Initially, the metadata can be collected and documented as an expert estimate. The learning algorithm for continuously improving the metadata checks it in practice and adjusts it if necessary. For example, an algorithm whose duration was originally estimated to be too short could be suggested less frequently over time, as its actual duration is now known. In this way, the metadata gets better and better and the selection and suggestion algorithm can fulfill its task better and better.