How to estimate system time-to-failure without sensor data?

I'm working in the prediction of time-to-failure of vehicles. The available data are the vehicle characteristics, such as make, type of vehicle (truck, car, etc.), year of manufacture, weight, region where is employed and lots more. Also the fuel consumption data is available, with the odometer updates and fuel load. But the richest source of data I have are the work orders, these are the maintenance task performed on every vehicle. Some of its important attributes are date, failure description and system that failed.

With this data I'm intending to fit a model for time-to-failure prediction. I've research the recent literature and every author have access to sensor data, which they use to perform feature engineering and fit their fancy models.

Do you think the work orders approach still valid for the times we are living? What would be a good benchmark?