Under practical conditions, the power quality disturbances have a complex nature, and are often corrupted with external noise. This calls for robust detection and classification using appropriate signal processing and classification techniques. In this study, a time-frequency-scale transform is presented as a detection tool with high-noise immunity. It is a variant of chirplet transform adapted for power quality studies, which incorporates a Hann window and is capable of shifting and scaling operations. A number of simulated and real power quality disturbances are detected and classified under various noise levels for performance assessment. Performance of the time-frequency-scale transform is dependent on window length, and hence three different classifiers are employed to study the effect of window length variation on classification accuracy. Detection and feature extraction in time-frequency-scale transform is somewhat similar to Stockwell transform; therefore, a suitable comparison is shown wherever required. Results of the proposed methodology are found to be appreciable. Moreover, observations from this work can serve as a foundation for formulation of optimised classification problems involving manoeuvrable window signal processing techniques.