Description
Our cutting-edge toolkit enables Predictive Maintenance (PdM) in dynamic and complex settings. It includes auto-tunable anomaly detection techniques for asset state monitoring, comprehensive PdM frameworks tailored for vehicle fleets, implementation of evaluation metrics and key performance indicators (KPIs), and an innovative contextualized anomaly detection framework designed to deliver robust and explainable PdM solutions. These tools are meticulously crafted to enhance reliability, efficiency, and decision-making in predictive maintenance operations.
Predictive Maintenance in dynamic environments poses significant challenges, including data and concept drifts, lack of labeled data, and the constraints of streaming environments. To address these, our approach leverages unsupervised anomaly detection techniques to monitor asset states, eliminating the dependency on labeled datasets. Auto-configurable solutions ensure robustness against data drifts and enhance usability for end-users. Additionally, context-aware methods are employed to provide explainability and improve decision-making processes. All solutions are designed for online deployment, enabling real-time analytics and swift responses.
Predictive Maintenance Toolset
- Contextualized Anomaly Detection
Github Repository
Leverages multimodal information to establish the context for anomaly detection methods used during monitoring, offering valuable explanations towards robust decision-making. - Automated Thresholding
Github Repository
Provides implementations of auto-tunable and dynamic thresholding techniques for anomaly detection. - Parameter-free Distance-Based Anomaly Detection
Github Repository
Implements distance-based anomaly detection for streaming data, requiring no parameter configuration to identify anomalous data. - Monitoring of Homogeneous Fleets
Github Repository
Facilitates health monitoring of homogeneous fleets through clustering techniques and wisdom-of-the-crowd approaches. - Monitoring of Heterogeneous Fleets
Github Repository
Monitoring the health of heterogeneous fleets through a Predictive Maintenance framework that integrates feature transformation and online semi-supervised learning.
People
- Apostolos Giannoulidis (PhD Candidate)
- Anastasios Gounaris (Professor)