At a Glance
Analyze both device component service calls and records and examine device utilization logs events to model optimal maintenance schedule thresholds relative to acceptable downtime, costs, and scheduling.
Developed machine learning models to optimize preventative & scheduled maintenance of on-site medical instruments and analytical devices, analyzing over 400,000 service events on thousands of component types in laboratory equipment
Expertise & Technology
SFL Scientific worked with a global diagnostics and analytical instruments corporation, operating in over 140 countries, to optimize preventative maintenance (PM) scheduling and reduce unplanned downtime of thousands of systems and asset classes.
To create machine learning models that monitor and predict downtime, understanding the dataset’s key variables that influence timing of calibrations, checks, PMs, and repairs events was needed and reverse engineer the breakdown reason or trigger event.
A major component is the organization and Exploratory Data Analysis (EDA) of the legacy data. Aggregating service logs with separate on-board sensor and device usage data requires approaches to clean, load, and analyze multiple diverse streams.
Depending on management priorities, organizations may choose to pare down systems and objectives to prove out value in a 6-8 week pilot. In this case, senior leadership understood the importance of making a full analysis and SFL Scientific had buy-in to perform a comprehensive data aggregation effort of all available information.
SFL Scientific Solution
To achieve our multiple objectives, we analyzed the datasets for (a) Repair Events, (b) PMs, (c) Routine Checks/Calibrations, and performed a full EDA feature and time-series analysis to build robust models.
We examined time to fault per component and correlations between groups of devices, age, models, manufacturers, customer-sites/locations, language, and categorical distributions such as time of year, etc. Outcomes included understanding on a per SKU or device basis the window of failure and time to fail after installation or a service event.
The full analysis enabled deeper insights into predicting optimal maintenance schedules, forecasting downtime, and optimizing tech labor. SFL created both prescriptive and predictive models for failures and duration and developed scheduling optimization for end-user visualization.
As field failure events were heavily dependent on age of device, historic service history, and usage, embedding IoT & device sensors for identifying additional events that cause downtime and other costs, e.g., device underperformance, calibration, vibration, power consumption, etc., can optimize real-time warnings and augment central operations.
Enabled organization to reduce current costs, improve customer satisfaction with increased up-time, and create service packages targeted at specific clients & devices, expanding repair-based revenues.