Crystallisation processes play a crucial role in the pharmaceutical and agrochemical industries, where particle size and shape distributions (PSSD) significantly affect product quality and performance. Accurate process models are essential for design and optimisation, yet they are often hindered by overparameterisation stemming from mismatches between model complexity and data quality or availability. This overfitting results in models that fail to capture the underlying physics, leading to poor predictive capabilities, limited extrapolation beyond controlled laboratory conditions, and inefficient use of experimental resources. This work discusses a rigorous, valuable methodology and introduces a tool to address practical modelling challenges, guiding practitioners in academia and industry toward more predictive models. We propose a comprehensive framework for identifiability analysis to diagnose and enhance model feasibility, demonstrated through a case study of cooling crystallisation involving needle-like particles. A digital twin was developed to simulate the crystallisation process and replicate measurement devices, followed by an extensive identifiability analysis at both theoretical and practical levels, yielding actionable insights for real-world applications. Our results are threefold. First, we emphasise the constraints that identifiability imposes on modelling when parameters are not uniquely determinable (i.e., ill-posed). Second, we discuss the effects of device selection, data resolution, sampling frequency, and experimental design on model identifiability. Finally, we propose strategies for aligning model complexity with available data to prevent overparameterisation from theoretical formulation through to practical implementation.
Zaman et al. (Thu,) studied this question.