How Spryfox Pioneered a Smarter Future for Heavy Machinery
thyssenkrupp Mining Technologies GmbH* is a global leader in the mining sector, offering cutting-edge systems and services for the extraction, processing, and handling of raw materials. Known for its engineering excellence and innovation, the company operates highly complex machinery, such as stone grinders, that play a vital role in industrial material processing.
The Challenge: Downtime and Wear-Related Costs in Stone Grinding Machinery
In high-capacity industrial environments, unplanned downtime can cause significant financial and operational disruption. thyssenkrupp’s stone grinders, which perform the comminution of ores and other minerals, are subject to continuous wear and eventual failure, particularly at the grinding rolls’ surfaces. Replacing these components on a fixed schedule often leads to either premature servicing or costly failures.
The company needed a more intelligent and dynamic maintenance strategy, one that could anticipate failures before they occurred and adjust based on the machine’s real-world usage.
However, several hurdles stood in the way:
- inconsistent data availability
- harsh operational environments that compromised sensor performance
- a lack of failure-specific data, and
- the challenge of scaling predictive models across different machines and sites.
Spryfox's Solution: A Pragmatic, Principle-Based AI Approach
Spryfox partnered with thyssenkrupp to design and implement a predictive maintenance solution powered by explainable AI. Their approach was guided by six core principles:
- Laying a Data Foundation
Spryfox began by establishing robust data collection mechanisms, aggregating up to two years of operational data from stone grinders. They accounted for sensor gaps and hardware failures by using data augmentation and designing models that could perform under imperfect data conditions.
- Designing the Right Target Variables
Instead of jumping directly into modelling, Spryfox fused laser wear measurements into reliable, meaningful target variables that captured both long-term wear and short-term degradation. These variables were validated using expert knowledge and cross-correlated with sensor data such as pressure, temperature, and input volume.
- Building Resilient, Adaptive Models
Recognizing that real-world operations are messy, Spryfox developed models that could handle missing data and sensor failures. They implemented fallback models and surrogate sensor logic, ensuring the system delivered insights even under degraded input conditions. Continuous monitoring and automatic re-training mechanisms were built in to adapt to machine aging and evolving operating environments.
- Ensuring Explainability
To drive user trust and adoption, Spryfox prioritized model transparency. They employed decision-tree-based methods that clearly showed which operating conditions led to higher wear. These explainable models enabled subject matter experts at thyssenkrupp to verify predictions and understand the root causes of alerts.
- Scalability Through Transfer Learning
Rather than training siloed models for each machine, Spryfox used transfer learning and few-shot learning techniques. This allowed the team to deploy base models quickly across similar machines and refine them over time, dramatically shortening the time to operational value.
- Operational Integration
Spryfox engineered an end-to-end deployment architecture combining edge and cloud. Industry-grade PCs were connected directly to machines for real-time inference and API access. High volume process data was transferred into a central system which handled data storage, model retraining, and distribution back to the edge devices.
Key Results
The collaboration yielded a robust predictive maintenance framework that is already producing measurable benefits for thyssenkrupp:
- Reduced unplanned downtime through early failure warnings
- Improved scheduling of roll replacements, minimizing unnecessary maintenance
- Increased trust and adoption among field engineers due to model explainability
- A scalable system architecture supporting multiple machines and sites
Looking Ahead
With predictive maintenance now embedded into core operations, thyssenkrupp and Spryfox are exploring further AI-enabled innovations across mining technologies. The partnership illustrates how data-driven strategies, when executed thoughtfully, can deliver real-world transformation.
If your business is looking to reduce operational risk and unlock the value of machine data, contact us today for an obligation-free consultation.
*thyssenkrupp has since sold the Mining Technologies business unit to the Danish company FLSmidth.