Reliability prediction of self-propelled mining machines

Customer industry: Mining
Problem: Given a load of time series data collected by onboard sensors of excavating equipment operated by the world largest silver mining company, provide a solution to decrease the number of unexpected machine breakdowns.
Solution: Long Short Term Memory (LSTM) deep recurrent neural network capable of reliably estimating time to failure, that is the number of consecutive eight-hour-long shifts before the breakdown.
Technology: Python (pandas, numpy), PyTorch, CUDA, Google Colab, Spark