Process Mining (PM) has shown great potential in business domains; however, its application in heavy industries remains limited. The main reasons are the dominance of raw sensor data, which requires preprocessing and abstraction, and the variability of industrial process execution. Based on mining cases studied in various research teams, this paper presents lessons that illustrate both the challenges and opportunities of applying PM in industrial contexts. We identified two key challenges: (1) constructing suitable event logs from heterogeneous sensor data, supported by domain knowledge, and (2) selecting modeling approaches that cope with process complexity and variability while serving the analytical objective. Our research focused on event log creation through case identification, event abstraction, and labeling techniques, including recent advances in utilizing Large Language Models (LLMs) for event abstraction. We also compared imperative, declarative, and hybrid modeling paradigms, highlighting their different capacities to represent variability of real-life processes. Based on mining experiences, the paper presents lessons transferable to other heavy industry sectors, demonstrating the potential of PM to analyze complex processes and support data-driven decision-making.