The application of Machine Learning has the potential to address a common problem in MRO inventory and maintenance management. Machine Learning is a branch of Artificial Intelligence (AI) with algorithms that can learn from and make predictions on data. This capability can be deployed to infer missing pieces of critical information to improve inventory and maintenance management.
Missing and/or inaccurate data is a pervasive problem in MRO inventory and maintenance management. For example, the criticality of inventory items is an important piece of information in determining optimal stocking levels but it is frequently missing and/or not well defined in many organizations’ inventory management systems. Similarly, maintenance work orders need to contain a detailed breakdown of labor hours for analysis, process improvement and management but these are frequently missing as the data are typically manually recorded and oftentimes technicians do not consider it a priority to do so.
One way to overcome this problem of missing data is to infer or predict the missing values from related pieces of data that are accurately recorded in the database. In the case of missing item criticality, one may seek to infer its value from related information such as the criticality of the equipment to which the item has been issued, the priority of the work orders associated with the item, the item’s supplier, commodity code, Quality Assurance (QA) inspection code etc.
In the formal terminology of Machine Learning, the missing criticality which we are trying to infer is referred to as the output or target value while the related information used to make the inference, such as the equipment criticality, are referred to as predictors or features. When the number of predictors is small, it is possible infer the output by using a rules-based system with Boolean logic such as: the item’s criticality is A IF it has been issued to equipment X AND/OR it has work orders with priority Y AND/OR it has been purchased from supplier Z AND/OR … ELSE the criticality is B IF …. However, when the number of predictors becomes large, then the complexity of the resulting logic makes such a rules-based solution impractical. In such a case, a Machine Learning algorithm may be developed to perform the inference by learning from an appropriate training data set. If the output has discrete values such as item criticality, its prediction constitutes a classification problem.
Conversely, if the output is a real number such as labor cost, then its prediction becomes a regression problem. In either case, the algorithm uses a mathematical equation with multiple coefficients to infer the output from the predictors.
The coefficients are determined by training the algorithm with a training data set where the output is known for a given set of predictors. This training data set may be generated by having a sample of the data parsed by human subject matter experts (SMEs) to identify the appropriate item criticalities based on the given predictors. Once this is done, the algorithm learns from the training data by adjusting its internal coefficients such that its output matches those generated by the human SMEs as closely as possible without overfitting. Once the algorithm has been trained, it encapsulates the expertise of the SMEs and it can then be applied to the rest of the data to infer the desired output.
The Machine Learning approach has two advantages over the rules-based system. Firstly, it is able to handle a larger number of predictors without being overwhelmed by the resulting complexity of the problem. A larger number of predictors, if appropriately selected, have the potential to provide a more accurate inference of the desired output. Secondly, most real-world data contain noise which is difficult for rules-based systems handle as it requires multiple levels of exceptions which again rapidly becomes overwhelmed by complexity. Conversely, Machine Learning algorithms, by their design, are more robust to noisy data.
Learn how Intelligent MRO is the smart way to manage MRO materials and spares