Posters
See the list of poster presenters at the Technology Workshop 2026 – and check out their work!
For more details on each poster, click on the poster titles to read the abstract.
PO33: Enhanced Anomaly Detection Methodology and Comprehensive Results for Wind Turbines Using Deep Learning and Statistics on CMS Vibration Data
Breno Carvalho, Performance Engineer, Casa dos Ventos
Abstract
A WIND TURBINE GENERATOR IS A COMPLEX ROTATING MACHINE COMPRISED OF MULTIPLE CRITICAL SUBSYSTEMS. SINCE WIND TURBINES WITHIN THE SAME FARM EXHIBIT SIMILAR BEHAVIORS, COMPARING SIGNAL TRENDS FROM ONE TURBINE TO OTHERS IN THE SAME FLEET IS OFTEN EFFECTIVE. HOWEVER, FOR VIBRATION ANALYSIS, COMPARING A TURBINE'S OWN HISTORICAL DATA IS CRUCIAL, AS VIBRATION PATTERNS MAY VARY BASED ON EACH TURBINE'S UNIQUE MECHANICAL CHARACTERISTICS. VIBRATION DATA ANALYSIS REMAINS ONE OF THE MOST EFFICIENT METHODS FOR EARLY FAILURE DETECTION, ENABLING PROACTIVE MAINTENANCE PLANNING AND MINIMIZING DOWNTIME. IN THIS WORK, A PROCESS WAS REFINED AND EXPANDED TO MONITOR ROTATING SUBSYSTEMS BY IDENTIFYING ANOMALOUS TRENDS BASED ON FREQUENCY DOMAIN ANALYSIS OF VIBRATION DATA FROM THE TURBINE CONDITION MONITORING SYSTEM (CMS). AUTOENCODER MODELS WERE TRAINED USING DIFFERENT RANGES OF FREQUENCY DOMAIN VARIABLES TO FLAG OUTLIERS, AND A STATISTICAL MOVING AVERAGE METHODOLOGY WAS APPLIED TO IDENTIFY DIVERGING TRENDS. THESE TRENDS WERE THEN PRIORITIZED THROUGH A LINEAR REGRESSION MODEL TO ENSURE ACTIONABLE INSIGHTS. WITH THIS UPDATED METHODOLOGY, RELEVANT ANOMALOUS VARIABLES ASSOCIATED WITH MAJOR COMPONENT ISSUES WERE FLAGGED ACROSS MULTIPLE TURBINES, RESULTING IN THE SUCCESSFUL PREDICTIVE DETECTION OF MECHANICAL FAULTS. CASE STUDIES DEMONSTRATED HOW THE TOOL ENABLED THE IDENTIFICATION AND RESOLUTION OF POTENTIAL FAILURES, LEADING TO CLEAR REDUCTIONS IN DOWNTIME AND THE OPTIMIZED ALIGNMENT OF INTERVENTIONS WITH LOW WIND RESOURCE PERIODS, AS FORECASTED BY POWER MODELS. THE METHODOLOGY DEVELOPED HERE ALSO SHOWED A SIGNIFICANT CORRELATION WITH ANALYSES INVOLVING TRADITIONAL MATHEMATICAL TECHNIQUES FOR VIBRATION ANALYSIS. ADDITIONALLY, INTEGRATION INTO A PROPRIETARY PYTHON-BASED SYSTEM ENHANCED THE MONITORING PROCESS, STREAMLINING MAINTENANCE PLANNING AND EXECUTION. BY INCORPORATING THIS TOOL, THE OPERATIONAL RELIABILITY OF WIND TURBINES WAS SIGNIFICANTLY IMPROVED. THE FINDINGS EMPHASIZE THE VALUE OF COMBINING DATA-DRIVEN ANOMALY DETECTION WITH STRATEGIC MAINTENANCE PLANNING TO MAXIMIZE ENERGY-BASED AVAILABILITY, MINIMIZE OPERATIONAL DISRUPTIONS, AND ENHANCE THE OVERALL EFFICIENCY OF WIND FARMS.
No recording available for this poster.
