Statistical Methods in Finance 2022

June 28 to July 2, 2022


Predictive Maintenance of ATMs

By Siuli Mukopadhyay
IIT, Bombay

Maintenance costs are a major part of the total operating costs of any manufacturing and production unit. With the aim to avoid unprecedented breakdowns, it is necessary to schedule regular maintenance by assigning technicians to check the equipment. This method, known as preventative maintenance is not cost-effective, since frequent maintenance can result in wasted labor and unnecessary travel costs, while too large a gap between visits can result in problems occurring without warning or the machine failing irrevocably. On the other hand, predictive maintenance (PdM) is a proactive maintenance strategy that tries to predict when a piece of equipment is likely to fail so that maintenance work can be scheduled and performed just before that happens. It monitors the performance and condition of equipment during its normal operation in order to reduce the likelihood of failures. It also reduces the additional costs of over-maintenance and improves asset reliability for customers. For the construction of a PdM model, we require data that records the internal state and condition of the equipment. This data is gathered through the use of various condition monitoring techniques and is accessible in the form of logs or system messages. For instance, in the case of bank ATMs (Automatic Teller Machines), all kinds of operations, as well as errors and warnings, are stored in its system logs. The valuable information retrieved from these messages is essential for detecting problems in the machine prior to their actual occurrence and thus proves to be relevant inputs for our PdM model. The goal of predictive maintenance is thus to predict whether the equipment is likely to fail, or not fail, in a given time window. We hereby choose ATMs as the example equipment in this project and we use real ATM event logs and maintenance records from an Indian bank as experimental data to evaluate our method on its feasibility and effectiveness.