Before an industrial plant selects Machine Learning for Predictive Maintenance, it will likely develop criteria for evaluating potential vendor solutions. Because we have been through the vetting process multiple times, we wanted to share with you the most common requirements considered.
For reference, we are sharing the answers we typically provide.
|Hardware||Requirements to purchase additional equipment to support a solution.||To the extent that new hardware must be installed in plants to detect signals to be analyzed by a Machine Learning algorithm.|
|Time to Deploy||Length of time from signing a contract to receiving meaningful/actionable output from a solution.||An iterative process requiring the local O&M staff to support the vendors’ Big Data experts can result in extensive ramp-up periods.|
|Plant Resource Commitments(O&M for deployment)||Requirement for internal plant resources to be assigned to deploying the solution.||Some solutions depend on plant technicians to train the system on (1) the behavior of the asset and (2) critical production processes.|
|Plant Resource Commitments (Big Data for solution maintenance)||Requirement for internal plant resources to be assigned to analyzing data, performing Machine Learning activities.||Certain Machine Learning solutions depend on Big Data expertise at a plant level.|
|Machine Learning Methodology||Strength and adaptability of learning algorithms.||A Machine Learning algorithm is not a static equation that is built once and then reapplied. Artificial Intelligence-based Industrial Analytics applies algorithms that learn from and adapt to new data.|
|Reinforced Learning||Integration of asset behavioral data into Machine Learning algorithm.||The optimal Industrial Analytics solution incorporates plant-level data failure incident reporting. In this way, the prediction rates can be further improved.|
The above checklist is a starting point; each vendor has its own strengths and weaknesses. However, as a starting point, a vendor must provide a production plant with a comprehensive list of requirements from existing O&M resources and an explanation of how the Machine Learning algorithm is applied to operational data.