Thermo-mechanical performance evaluation of hybrid NiTi/CF-PEKK composite laminates using experiments and machine learning approaches

• Thermo-mechanical behaviour of hybrid NiTi/CF-PEKK composite laminates (HNCLs) was experimentally characterized. • Experimental results were used to train different machine learning (ML) models. • Austenitic–martensitic transitions in NiTi and the brittle-to-ductile shift in PEKK together governed the tensile response. • GA-optimized neural network (GA-BP-NN) provided the most accurate stress prediction, achieving MAE ≈12.6 MPa and RMSE ≈10 MPa. • HNCLs exhibited minimal degradation of performance up to 150°C, demonstrating their adequacy for aerospace applications under extreme conditions. This work presents the first systematic investigation of the temperature-dependent mechanical behavior in hybrid NiTi/CF-PolyEtherKetoneKetone (PEKK) composite laminates (HNCLs), which has remained unexplored to date. The HNCLs were tested under tensile conditions in the range of 50°C to 175°C at an interval of 25°C, which covers the transformation temperatures of NiTi used in the experiments as well as the glass transition temperature (T g ) of PEKK. To optimize and predict the performance of HNCLs, various machine learning (ML) algorithms were trained using experimental datasets. The results showed distinct behavior over a range of temperatures due to the interplay between the phases of NiTi and the behavior of PEKK corresponding to the testing temperature. The thermomechanical response of HNCLs was governed by the interaction between NiTi phase transformations and the temperature-induced transition from brittle to ductile in PEKK. The fracture morphology observed through SEM revealed different failure mechanisms at various temperatures, including fibre pull-out, fibre-matrix debonding, and viscoelastic deformation of PEKK above its T g . ML models showed that, comparatively, the neural network based on genetic algorithm (GA-BP-NN) outperformed support vector regression (SVR), Decision Tree, and BP-NN in predicting tensile response, exhibiting the lowest mean absolute error (MAE) of around 12–14 MPa (corresponding to 2–3% of maximum stress levels), root mean square error (RMSE) of around 10–20 MPa (corresponding to 2 – 4% of maximum stress level), and mean absolute percentage error (MAPE) as low as 5.6%. Moreover, GA-BP-NN outperformed baseline regression model by 12.1%.