The widespread adoption of areas such as Machine Learning, the establishment of Industry 4.0, and the various techniques and information available to companies today foster the need to incorporate advanced control and monitoring tools, such as predictive failure systems, into asset management. While there are various documented cases of trained ML models yielding good results, there is still a lack of clarity on how to address all the stages that an analysis like this requires in a general manner, considering that it must be valid across different areas and different data characteristics. This article presents and describes a workflow that encompasses this methodological proposal for the development of failure forecasting systems, which was then applied to the case of a mining conveyor belt in Chile. The study and its application case result in a successful integration between data from a Distributed Control System (DCS), a Digital Twin, and an operational logbook, as well as precision and recall values exceeding 0.83 in the best cases of the various trained algorithms with data transformed into new variables and the application of principal component analysis (PCA). This is done both for failure prediction in general and for fault type-oriented forecasting Based on this, the paper presents a transferable methodological proposal that is adaptable to various data sources without relying on specific assets or physical process information. Its main strength lies in reducing dependence on maintenance data for anomaly detection. However, this approach lacks validation and raises clarity issues, diverging from the Functional and Informational Requirements outlined by other authors. Despite these challenges, the model shows acceptable results, and the potential to integrate operational data allows for further development. Future iterations may focus on improving calculation times and addressing the challenge of identifying the origins or causes of predicted events.