AI predictive analysis gets the green light

This article originally appeared in the July'23 magazine issue of Electronic Specifier Design – see ES's Magazine Archives for more featured publications.

Caroline Hayes finds out how and why the semiconductor company is branching out into software as a service (SaaS).

In August 2022, Infineon Technologies acquired Berlin startup Industrial Analytics and at PCIM 2023, the company was able to elaborate on how this addition to the Industrial Power Control business unit would be integrated into its software and services business and contribute to power efficiency in IoT design.

AI evaluates performance

Established in 2017, Industrial Analytics develops artificial intelligence solutions which for example monitor plants for early detection of critical developments, based on analysis and evaluation of vibrations. The AI solutions from Industrial Analytics not only evaluate data for predictive maintenance but also provide recommendations for action (prescriptive maintenance).

The fledging company specialises in vibration analysis and data analytics at the edge and in the Cloud with algorithms that provide insight into machinery and operations that can increase energy efficiency and extend the useful operating life of equipment through prescriptive maintenance.

An early project was in collaboration with the German Aerospace Institute and the University of Stuttgart to create an infrastructure for autonomous driving and maintenance of trains. It also worked with clients in the food and beverage industry and a utility company to optimise gas storage facilities.

Digital twins are not new but the conventional model has limitations, explains Daniel Scharfen, vice president of Green Power Analytics at Infineon’s Green Industrial Power division. “The reality is that statistical machine learning requires a lot of historical data to identify anomalies,” he says. “The constant data flow of these systems needs a new approach to identify patterns.”

Mechanical and electrical expertise

Infineon contributes the electrical expertise and Industrial Analytics contributes the electromechanical skillset in the digital twin design, continues Scharfen, for “explainable AI” without having to have technicians in the field.

Physics-based hybrid models deliver a physical implementation with AI to create a range of possible results, says Scharfen. The model can be viewed as a system, where parts interact and can influence other parts with a dynamic range of operations throughout the system.

They can also comprehend the underlying root causes of malfunctions and prescribe effective maintenance measures, says Infineon.

Data from both machinery and the operator on day-to-day activity is gathered through the Edge Analytics service. This combines data pre-processing at the equipment level with advanced analysis of relevant sensor data which is then transmitted to the customer for interpretation.

The digital twins are built in collaboration with customers, taking the physical correlations of machine components into account. The hybrid machine learning models are creating using statistical data from the machines, calculated data and direct operator feedback obtained by operators maintaining and operating the machine day in day out.

The AI algorithm learns from the operator feedback and will notify the operator in the future when the failure occurs again. Pre-trained models are used for different equipment and the scalable solution is based on an open IT architecture.

This model has less than 30% false positives and does not require training each time new data is acquired, unlike a statistical model. Instead, the physics model can transfer data from one system to another. For example, a production line with five machines does not need five test benches as a single model can be tested and be valid for all machines.

AI for power efficiency

For Infineon’s power modules, the SaaS can be used to estimate product lifetimes, thermal performance, assess if a module meets a specification, or model what differences result from the use of magnetics, transformers or capacitors and temperature changes. All of these factors can impact the longevity of a product.

Announcing the acquisition, in which Infineon owns 100% of Industrial Analytics’ shares, Peter Wawer, President of Infineon's Industrial Power Control division commented: "We intend to jointly expand Industrial Analytics’ business and to offer Infineon's industrial customers new AI solutions that complement our semiconductor portfolio."

For industrial automation, the AI hybrid models can improve runtime and system availability and in residential IoT systems it can be used for HVAC systems to understand a customer’s energy management but also to improve IoT equipment operation and energy efficiency.