The application of machine learning to high cost, low volume (HCLV) manufacture is challenging due to prohibitive costs and low data volumes. An example HCLV application is linear friction welding (LFW) of Blisks (Bladed Disks). LFW is a solid-state joining process, typically used in high integrity aerospace applications. The successful application of machine learning (ML) has the potential to predict quality metrics and enable timely interventions to machine maintenance for avoidance of machine damage or deterioration. This paper proposes a methodology that combines expert knowledge with machine learning to minimise the quantity of weld data required to generate a robust and accurate ML model. Expert knowledge incorporation requires methods of elicitation, capture, standardisation and quantification of information (it can be qualitative, experiential and subjective) and conversion to a quantitative, data driven and digital format for input into a ML algorithm. This paper will describe the methodology developed to enable a combined data science and engineering approach to address complex manufacturing problems. If successful, this methodology will be used as a standard framework for application to HCLV manufacture.