Optimize tool utilization and improve product quality.

Determine the lifespan and wear condition of machine tools for any combination of machine, material and process.

Initial situation

Better quality products at lower cost

Operators have learned through experience when a tool is starting to wear down, but this method is not very accurate resulting in poor quality products, unused tool life, and unplanned downtime. A manufacturer would like to address this issue and offer a digital service forecasting the Remaining Useful Life (RUL) and asses the tool wear condition, helping its customers produce better quality products at a lower cost.

No data available

The needed data is captured at the machine, but unavailable for machine learning

Data availability is critical in any AI project, and often the deal breaker. Here's the problem: No single company can collect sufficient data for the various tool, machine and work material combinations. Moreover, collecting data directly from the users usually fails as they don't want to share their data for fear of losing their competitive edge with the data. . So while the necessary data is already captured at the machine, it's unavailable for machine learning – Katulu calls this the Distributed Data Dilemma.


Decentralized AI for tool wear prognostics

Katulu's unique decentralized approach to AI allows learning across different sites, machines, tools, and work materials, while all sensitive data remains securely with the users so that no one needs to fear losing their competitiveness. This solves the Distributed Data Dilemma and provides access to the relevant data for a state-of-the-art industrial tool wear prognostics solution, enabling online tool wear condition estimates, such as wear depth, RUL and conditions forecasts.

Does this scenario sound familiar?

Let us talk about your use case. We look forward to hearing from you.

How to start building an AI for tool wear monitoring and prognostics

The development of a decentralized (federated) model begins, like any other model, with initial experiment data acquisition. Collect data such as forces, torques and motor currents in a technical center or laboratory. It's important to only collect data also available in production environments so that learnings are applicable in production. A laboratory process measuring wear as well as mechanisms for surrogate labeling from data can provide the necessary tool wear labels. The next step is to extract the physical properties of the machine tool process using signal processing techniques such as Katulu's Signals Library. The extracted features are then used along with the data to train a machine learning model based on Katulu's temporal deep learning architecture for tool wear monitoring and prognostics.

You can use the Katulu Uniwear Dataset as well as the Uniwear Modelling Example as a starting point.

Katulu FL Suite screenshots

MVP or The first functional product

With the first model ready, the data flow between machine and edge devices running the tool wear model needs to be established based on the desired deployment scenario. One also needs to design a process for capturing user feedback either direct or indirect – Maybe a button to agree or disagree with the predicted tool condition within a tool management solution or a different data source like product quality assessment as a proxy for tool wear. Now, it's ready for prime time. Start with a small beta to resolve initial kinks and then roll out to more users providing access to the necessary data and allowing for more robust models.

Clustered Federated Learning (CFL) in tool wear

A federated learning solution for machine tools should be able to handle different cutting conditions, work materials and tool characteristics. Katulu's clustered federated learning algorithms provide a significant uplift in better detection of tool wear compared to non-clustered federated learning. Moreover, the dynamic nature of machine tools usage among different machines, and materials and irregular scheduling of cutting processes make dynamic clustered federated learning an excellent approach for distributed settings.

Continuous improvement

The local online learning setting for our tool wear MVP provides near real-time model building and utilization. This continuous learning captures changes in cutting conditions, workpiece and tool materials.

Learnings transferred from different participants, improve baseline AI modules over time. Along with the gradually introduction of new signal characteristics and architecture fine-tuning will result in a better and better solution over time.

FL Suite Pipeline, Recurring Run and Logs screenshot

Are you ready for your journey into distributed machine learning?

We look forward to supporting you with your ideas and projects.

Advantages of Katulu FL Suite

The challenges in machine tools are complex, such as multiple workpiece materials used in the production lines and irregular use of the same tools in diverse schedules. Moreover, machine tool processing may occur in different physical locations and entities with data privacy requirements.

Increasing tools utility and productivity

Achieving higher productivity via increased tool utility requires AI-assisted tool wear condition monitoring. Katulu's technologies create the necessary infrastructure and tools to accomplish this goal seamlessly.

Higher machine availability via remaining useful life

Do you have a world-class machine tooling or related processing that rarely breaks down? Katulu helps you prevent infrequent failures due to unknown remaining useful life (RUL) of machine tools with intelligent tool condition monitoring. The AI-assisted approach brings sustainability into the product lifecycle and stops unscheduled maintenance events.

Increase process stability

Create the conditions for increased process stability for all machines, tools and materials combinations in the field with Katulu and improve quality with increased efficiency. Katulu's machine tool solution prevents machine tool breakage that would degrade the strength of the overall manufacturing operation and early detecting high tool wear.

Harnessing knowledge

Whether it's employee turnover or demographic change, Katulu makes it easy for you and your users to capture and leverage empirical process knowledge across organizational boundaries. Because your industrial users learn from each other without learning about each other through the use of Katulu Federated Learning, your users are still protected.

Automatic clustering as an enabler of AI

Machines are used in different environments and processes and are designed accordingly. This makes it difficult to determine which set ups can be trained together when no information is shared.

Katulu Federated Learning identifies which machines and assemblies can be trained together according to their design and automatically groups them into clusters that can be trained together. Changes to the design of a machine are automatically detected and result in an adjustment to the grouping. This ensures that each machine - no matter how individual - learns from other comparable machines or assemblies.

True data sovereignty

Protect everyone's data sovereignty. Machine builders, tool manufacturers and industrial users benefit from gaining knowledge without sharing sensitive data.

Are you ready to use AI the right way?

Book a free consultation with our experts to talk about unlocking your AI potential.

Custom Digital Twins for Thermoprocessing

Learn how a manufacturer of thermoprocessing equipment enables the simulation of manufacturing processes for each customer on the screen.