From cloud connection to predictive maintenance algorithms
The keynote speech “AI for machine builders. From Cloud Connection to Predictive Maintenance Algorithm” was given jointly by Philipp Mayer, Managing Director of codestryke, and Felix Kraft, CFO and co-founder of ai-omatic. Codestryke is a young company specialising in the Internet of Things (IoT) for mechanical engineering, ai-omatic harnesses data via artificial intelligence to develop productive predictive maintenance applications for customers.
Predictive maintenance is about neither servicing machines too early and unnecessarily replacing parts, nor waiting until damage occurs, which often results in longer downtimes and thus higher costs. Predictive maintenance aims to use available data to predict when parts need to be replaced to ensure that machines are operating optimally and as efficiently as possible. Predictive maintenance requires data that is collected via IoT networks.
Since 2017, codestryke’s focus in its more than 50 projects has been to develop IoT connections in a scalable way so that all machines can be connected. In Mayer’s experience, connecting individual machines as an example case is not useful in practice.
At the beginning of an IoT project, Codestryke takes stock. What data is already available? What other data needs to be retrieved and how often? And how can it be accessed using hardware or sensors? The proof of concept evaluates the concept fit; after feedback from the customer, it is further developed and rolled out to all machines. Old machines are retrofitted with IoT capabilities, new ones are integrated. Once the connection is complete, it is optimised.
Ensuring long-term success
In order to successfully implement an IoT project, Mayer says, it is important to work towards collecting the right data from the very beginning in order to contribute to the efficiency of the operation. The technical feasibility must be considered, because a company that is to be networked rarely consists only of new machines that can be easily integrated. Last but not least, the choice of the AI application is relevant and also determines the success of a project.
Mayer has experienced a number of challenges, specifically in the complexity of projects. The better complexity can be reduced, the sooner his company can ensure that data remains a source of information and does not become a problem. Implementing IoT is a huge change, he says, which requires appropriate support from the company—staff must be available, and responsibilities must be clear so that the project can move forward. Ideally, all machines that are to be connected should be managed centrally, as otherwise difficulties could arise in coordination. To keep the project successful in the long term, security and flexibility should be ensured by allowing over-the-air (OTA) updates from anywhere in the world.
As a solution to these challenges, codestryke has developed the software Vergelink. Vergelink compiles data in a third-party system, runs hardware-independently and can be used in many areas. The software automatically recognises machines in the system and data points are easily selectable. Mayer gave insight into an application example in which several hundred machines already in use at a customer’s were subsequently connected to the cloud in record time with the help of the software—without any downtime of the machines. Since the connection could be made remotely, travel costs in the six-figure range were avoided.
Data secured—and now?
Once the data flow is secured, companies like ai-omatic come into play. The software company processes the data using an algorithm, which according to Kraft is the heart of the company and independent of the hardware, as he emphasised.
The ai-omatic approach combines statistical evaluations with a neural network. A digital twin of the system to be monitored is created, from which the algorithm learns what the normal state looks like. New data is compared with the normal state. An alarm is triggered via the system if data deviates from the target values. The special feature of the ai-omatic solution is that no past data is necessary. The system does not learn from past mistakes but is directed towards the future.
Condition monitoring functions via a simple visualisation. It is clearly recognisable in which area of the sensor system the deviation occurs. The anomaly score, which is being developed for the digital twin, is able to detect deviations much earlier than conventional methods and visualises them for users. Using an example that juxtaposed both methods, Kraft showed that the AI solution could detect bearing damage two weeks in advance.
Ai-omatic offers its software under its own logo or as proprietary solutions. This is a big advantage for manufacturers, because they usually do not have the capacity to develop their own solutions—but if they can also offer predictive maintenance, this is often a competitive advantage.