An integrated change-point detection framework for wind turbine monitoring and fault diagnosis using SCADA data

The high cost of wind turbine maintenance underscores the need for robust and reliable fault detection and condition monitoring techniques. Traditional approaches typically rely on a single detection method and predefined models of normal operation, making them vulnerable to false alarms and reduced adaptability under varying operating conditions. This study introduces a novel integrated Change-Point Detection (CPD) framework that combines three complementary statistical methods to enhance the accuracy and timeliness of fault diagnosis using SCADA data. The framework integrates: (1) a Chow test-based method for detecting single structural breaks; (2) a CUSUM test-based method for identifying multiple or unknown change points; and (3) a stationarity-based CPD method employing the Augmented Dickey–Fuller (ADF) test to detect abrupt changes in data stationarity without relying on predefined behaviour models. A multi-level fusion strategy synthesizes the outputs of all three methods, classifying turbine health into Warning, Alarming, and Shutdown states based on detection consensus and timing. This hybrid approach enables robust, early, and interpretable fault detection while supporting proactive maintenance decision-making. Validation using SCADA data from a commercial wind turbine with two consecutive generator faults demonstrates the framework’s effectiveness in predicting failures at an early stage, thereby reducing downtime and maintenance costs. • A novel integrated change-point detection framework for wind turbine faults. • Combines Chow, CUSUM, and ADF tests for robust fault detection using SCADA data. • Multi-level fusion strategy classifies turbine health into three alarm states. • Detects generator faults early, reducing downtime and maintenance costs. • Scalable framework adaptable to other turbine subsystems and farm conditions.