Increasing uptime through predictive maintenance

wattx set out to reduce downtimes of machines manufacturing Fast Moving Consumer Goods (FMCGs) utilizing modern sensor solutions combined with AI-powered predictive maintenance.

The challenge

The project is aiming to reduce downtimes of machines manufacturing Fast Moving Consumer Goods (FMCGs) - such as shampoos, toilet paper, or soft drinks, to name a few - utilizing modern sensor solutions combined with AI-powered predictive maintenance. The project idea came into existence during a hackathon hosted by an industrial partner who brought to our attention the incredibly high amount of downtime typically found in the field of FMCGs.Our own research confirmed not only that machine downtime on average adds up to an astonishing 25-40% of the overall production time, but also determined that a large part of that could be prevented.

Reducing preventable downtime via predictive maintenance

Manufacturing machines typically run at a certain operating efficiency that is usually expressed in percentages. A machine running at 70% efficiency, for instance, can have 30% of downtime. During such downtime, no items are being produced.

A certain amount of downtime is in itself normal and to be expected as machines need to regularly undergo maintenance or calibration. However, there is also downtime caused by machine malfunctions, which can in turn decrease production quality, stop production completely, or even lead to machines being permanently damaged. Most of these malfunctions can be detected by the machine itself which then halts and notifies an operator; only after such an issue is resolved, production can resume as planned.

Those malfunctions that cannot be detected automatically, call for constant human supervision. Supervision is usually expensive, prone to error, introduces higher latency, and thus, can lead to a significant decrease in production quality.

Our approach

This is where our product comes in. We not only detect those errors missed by the machine, but we also make predictions on those about to happen, give clear recommendations for action, and in this way, reduce downtime.

First, the error reports that are generated by the control system of each machine are analyzed and paired with the observations and recordings of the respective machine operators. Subsequently, through a thorough analysis of the error data, error hot spots are identified and specific sensors (such as vibration sensors, etc.) are attached to those hot spots. The software then correlates the sensor data with the malfunctions of the machine to build a model of the machine’s behavior. The readily build model is used to detect malfunctions before they can have an impact on the production process. Finally, the software sends out a warning to the user before a decrease in the process quality can occur.

The outcome

In case you want to chat about this project, please reach out to

Our software model is based on deep learning algorithms that self-learn using sensor data and error reports. Consequently, the training of our models requires little manual time from machine operators or engineers. Additionally, by using deep learning we are able to detect short and long term patterns, usually unrecognizable by the human eye, and our models are robust to noise and can, therefore, handle data captured by low-cost sensors. These advantages allow us to develop a failure detection system that is highly cost-efficient and effective.

We are currently looking for pilot customers, so if the problem sounds familiar to you – we would be happy to talk!

Explore our offerings

Innovation workshops
Test and vet your idea and learn our know-how in user-centered techniques and the construction of modern business models.
Proof of concepts
Validate problems, ideas, and save both time and cost before diving deep into the project.
Company Building
The alliance between your industrial and domain expertise and our know-how in modern technologies.
Want to learn more?