Predictive maintenance is a maintenance policy where the goal is to detect potential future maintenance risks in a system so that the maintenance process can be optimised before system faults occur. This paper describes a deep learning model that does not require domain expertise. Deep learning approaches have several benefits over explicit statistical modelling: (1) they require far less domain-specific knowledge; (2) if the underlying data-generating mechanism of assets changes, a deep learning model would only need to be retrained to learn these new changes; (3) they can capture non-linear and complex multidimensional relationships; and (4) they may outperform rule-based or statistical methods. The paper describes how the model predicts maintenance-relevant events, along with the cost of the upcoming event and the time when it will happen. The paper describes the use of a long short-term memory architecture for our deep learning model. By doing so, the cost values represent a real, quantitative value of the potential maintenance costs in the future of an asset. Event, cost and time prediction are all achieved with high accuracy. This allows for the development of maintenance solutions without the initial high degree of domain process knowledge required. The artificial intelligence model can be used to raise an alarm when the cost values exceed some threshold, when the frequency of high-cost events increases significantly over the lifetime of an asset, or when the expected cost exceeds the cost of maintenance.