Overview
ISO 13379-2:2015 - "Condition monitoring and diagnostics of machines - Data interpretation and diagnostics techniques - Part 2: Data‑driven applications" provides procedures for implementing data‑driven monitoring and diagnostic methods. It is focused on methods and workflows used by specialist staff (typically in a monitoring centre) to detect anomalies and identify faults by comparing observed process data to models of normal behaviour. Data sources include DCS/data historians and specialised monitoring systems.
Key topics and technical requirements
- Scope and purpose
- Procedures to build, initialize, tune and apply data‑driven models for condition monitoring and diagnostics.
- Training approaches for monitoring (trained on normal behaviour) and diagnosis (trained on normal and fault conditions).
- Data preparation
- Data cleaning: identify and handle missing, noisy, out‑of‑range or stuck data; warning to exercise caution before deleting historical records.
- Interpolation issues: beware of historian compression/linear interpolation that can distort correlations.
- Resampling: resample data to appropriate rates (retain timestamps for transients; coarser sampling under steady state).
- Model development
- Select relevant features (raw or derived), grouped by function (mechanical, electrical, thermal).
- Train models using data covering all expected operating conditions; build separate models for distinct operational modes where required.
- Prepare and execute model tests; evaluate model performance and set alarms based on residuals (differences between observed and expected values).
- Diagnostic methods
- Pattern recognition and classification approaches for fault identification.
- Examples of data‑driven algorithms noted: AAKR, PLS, SVM, Mahalanobis‑Taguchi (MT).
- Practical recommendations
- Competence needed to build and maintain models (deep statistical knowledge not mandatory, but practical experience is).
- Consider additional sensors if existing data do not cover critical failure modes.
Applications
- Predictive maintenance and anomaly detection for rotating equipment (e.g., turbines, pumps), process plants, HVAC and electrical systems.
- Integration with plant historians, SCADA or specialised vibration/monitoring systems to detect abnormal behaviour and trigger alarms.
- Use cases include online monitoring centres, condition monitoring programs, reliability and criticality analyses, and root‑cause diagnostics.
Who should use this standard
- Condition monitoring engineers and technicians
- Reliability and maintenance managers
- Data scientists and analysts working in industrial asset management
- Monitoring centre operators and system integrators
Related standards
- ISO 13379-1 (General guidelines)
- ISO 13379-3 (Knowledge‑based applications)
- ISO 13372 (Vocabulary)
- ISO 17359 (Implementation procedures and audits)
Keywords: ISO 13379-2, condition monitoring, data-driven monitoring, machine diagnostics, predictive maintenance, data cleaning, model development, anomaly detection, pattern recognition.
Open preview (PDF)Condition monitoring and diagnostics of machines — Data interpretation and diagnostics techniques — Part 2: Data-driven applications