Predictive Maintenance for critical Pump Systems & Plants in data sensitive environments

Rethink your Predictive Maintenance solution with Katulu FL Suite! Address new customer segments by offering AI without data transfer as part of your existing solutions.

Initial situation

Extend your cloud-based predictive maintenance solution

The cloud-based predictive maintenance solution of a pump manufacturer is to be expanded to include an additional offering aimed at operators of critical pump systems. Plant operators who run complex pump systems with a large number of different pumps do not want an isolated solution for each pump manufacturer; instead, these companies need a central system for all plants. In addition, such complex systems are often part of critical process steps, so that the use of a cloud-based solution for predictive maintenance must be viewed critically due to data transmission. Therefore, a specific offering based on the existing solution will be developed that takes into account the specific data availability, usage and security requirements of this customer group. In addition, more complex predictions of cavitation shall be enabled in order to completely avoid cavitation operations.

No data

The data for the perfect prediction is not in your own hands!

Accurate predictions of unpredictable events such as machine failures or malfunctions require very large amounts of data for each product type of a manufacturer. In addition to the continuous or periodic collection of machine and process data, the correlation to historical events such as failures, malfunctions or repairs is especially crucial. This makes it possible to predict the probability of certain events in order to make statements about the expected maintenance requirements. However, collecting this data directly in the field from customers often fails due to the fundamental problem AI application is facing in industry. Companies don't want to share their data. This is all the more true for industrial users who operate critical pump systems. Information related to critical infrastructure must be kept secret. Therefore, cloud-based predictive maintenance solutions are only adopted by certain customer groups. It is only possible to train an AI model that perfectly predicts unpredictable events based on data of particular customer groups.


Integration of AI predictive maintenance models into plant operators' existing systems.

The pump manufacturer's existing predictive maintenance solution will be extended with the Katulu FL Suite. Existing customers of the existing solution continue to transfer their data directly to the pump manufacturer's cloud-based solution, where the data is used tro train a central AI model. For more complex pump systems customer-specific AI models are calculated decentralized on the Edge, which benefit from the insights of the central model. In this hybrid approach of federated learning, all customers benefit from each other with the difference that the sensitive data of larger systems is not transferred to the cloud. The customized models enable real-time predictions with significantly higher accuracy to completely prevent machine failures. For prediction and early detection of cavitation, additional sensor technology captures audio data that is analyzed directly on the Edge. As a key digital added value, the customized models can be directly integrated into the customer's existing systems via interfaces, so that all information converges centrally in one system.

By using Katulu FL Suite Learning, the requirements of this customer segment can be excellently met, because the process data remains in the plant and the AI models for predictive maintenance can be integrated into the plant operators' existing systems.

Does this scenario sound familiar?

Let's talk about your use case. We look forward to the exchange.

First steps

The development of a decentralized model begins with data collection. Pump data sets are collected in the pilot plant or laboratory to train an initial model. It is essential to label the data sets according to failure modes. To do this, experiments are conducted with different machine parameters and cycle times, and the experimental results are evaluated to correlate identified characteristics of failure modes with the control data. Another very valuable data source are historical data that can be assigned to a specific fault case.

With this data, an initial AI model will be trained to identify key factors influencing operations, with the goal of enabling predictions of failures and malfunctions.

Katulu FL Suite screenshots

Minimum viable or viable product.

The first model is rolled out with the application to assist the user with predictive maintenance. Here, it is critical to capture the user's feedback on maintenance and failures in order to tag the data. To do this, Katulu mostly uses a mobile app for tablet or smartphone to keep the barrier to entry low. This initial version of the minimum viable product, including the required edge hardware, is provided to selected customers. Data collected with the application will be used along with captured machine data to train more robust models using federated learning.

Continuous Improvement

Continuous training and deployment cycles improve the model, and special model metrics (federated analytics) ensure that these improvements are also improvements to the model for the individual user. When this level of maturity is reached, it is the right time to train more advanced models that will help improve overall machine availability.

FL Suite Pipeline, Recurring Run and Logs screenshot

Are you ready for your journey into distributed machine learning?

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

Advantages of Katulu FL Suite

The challenges facing pump and compressor manufacturers are becoming increasingly complex, exemplified by the continuously increasing demands for overall plant effectiveness and the growing need for value-added digital services that adapt to customer requirements. The benefits of using Federated Learning for pump and compressor manufacturers are multi-faceted and an important step in meeting the new challenges.

Higher machine availability

Do you provide world-class pumps and compressors that rarely fail? Katulu helps you capture the rare failures of system-critical components to avoid machine breakdowns at an early stage in terms of predictive maintenance and sustainably extend the product life cycle.

Insights for product development

Katulu continuously provides you with exactly the anonymized insights from the field you need to improve your products. Learn from your pumps and compressors in the field without learning sensitive information about your users.

Better predictions

Through ongoing machine learning, with each iteration, pooled insights continuously flow back into and improve each customer's model. This iterative process increases the accuracy and robustness of the models and also avoids bias effects such as bias or bias despite customer-specific optimization. Through this, Katulu Federated Learning creates more reliable predictions of rare events such as machine failures.

True Data Sovereignty

Katulu protects the data sovereignty of EVERYONE. Pump and compressor manufacturers, operators and users benefit from mutual anonymized insight without sharing sensitive data.

Are you ready to use AI the right way?

Book a free consultation with our experts to talk AI without the headaches.

AI-based Injection Molding

Learn how process cability in injection molding is optimized across company boundaries with federated learning by Katulu - without data transfer from production sites.

AI-based Tool Wear and Health Monitoring

Learn how tool wear monitoring and prognostics can be applied across company boundaries with federated learning by Katulu - without data transfer from production sites.

Custom Digital Twins for Thermoprocessing

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