Predictive maintenance (PdM) using Machine learning (ML) is a top-rated business case with respect to the availability of data and potential business value for future sustainability and competitiveness in the manufacturing industry. However, applying ML within actual industrial practice of PdM is a complex and challenging task due to high dimensionality and lack of labeled data. To cope with this challenge, this paper presents a systematic framework based on an unsupervised ML approach by aiming to construct health indicators, which has a crucial impact on making the data meaningful and usable for monitoring machine performance (health) in PdM applications. The results are presented by using real-world industrial data coming from a manufacturing company. In conclusion, the designed health indicators can be used to monitor machine performance over time and further be used in a supervised setting for the purpose of prognostic like remaining useful life estimation in implementing PdM in the industry.