In Big Data Insider, Dmitry Baykov, Team/Tech Data Science Lead at DataArt, explains how their team developed a predictive maintenance platform using IoT and ML to prevent equipment failures and malfunctions, implementing an event-driven architecture, utilizing AWS cloud services, and achieving high detection accuracy with a custom neural network model, resulting in extended machine service life and reduced spare parts inventory for the customer.
"The customer's large sorting center with conveyor belts, sensors and robots risked a standstill due to possible component failure. This outage could cost as much as $1.5 million per hour."
"In just four months we released the first software release using these time and cost saving approaches: Use of the AWS cloud to host and automatically scale our system... Secure data using standard AWS encryption... Use of an event-driven architecture to allow real-time operation of the system... Manage large amounts of data by creating backup pipelines and later switching to read-only raw data storage to reduce costs... Achievement of high detection accuracy with a custom neural network model on Keras and TensorFlow after initial deployment of pre-built containers on SageMaker."
"Thanks to the implementation of the platform, the customer was able to increase the service life of the machines by up to 40 percent and reduce the spare parts inventory by up to 20 percent."
The original article in German can be found here.
