MAINTLET project to promote scientific discovery by decreasing the maintenance costs and downtime of scientific instruments

Maintenance of scientific instruments may seem like an unglamorous topic: when we think of telescopes and microscopes, we want to think of gazing at Saturn’s rings or watching amoebas crawl across slides—not service contracts and spare parts. But the costs involved in maintaining and repairing experimental equipment can be staggering. In some scientific environments, 15% to 60% of the total costs can originate in maintenance activities, and about 33% of the money spent on maintenance is wasted on unnecessary and preventable activities. Furthermore, when scientific instruments fail, research grinds to a halt. The situation is especially bad in universities, where instruments are kept in service for longer periods, and with smaller budgets, than in industry settings.

Enter MAINTLET, a new NSF-funded effort to minimize maintenance costs and ensure instrument availability. MAINTLET will develop an advanced sensory network cyber-infrastructure that uses modern AI-guided big data methods to help labs identify patterns that indicate when to purchase parts and services, and to minimize interruptions due to repairs and maintenance.

The project is being led by CSL Director Klara Nahrstedt, who is also a Grainger Distinguished Chair in Engineering and a professor in computer science. She explains that MAINTLET’s vision emerged during her prior work on laboratory equipment monitoring in cleanrooms. In particular, the earlier NSF-sponsored SENSELET project looked at monitoring of the microclimates in cleanrooms where instruments were operating, to help verify semiconductor fabrication experiments’ validity. “But then we started to discuss with the lab managers, faculty, and students in the Holonyak Micro and Nanotechnology Laboratory (HMNTL) and Materials Research Laboratory (MRL) that this kind of external sensory observation could also be very well used, not just for understanding the microclimate around those instruments, but... for starting to track how these instruments age.”

MAINTLET will provide lab staff with two types of information. First, to make preventive and predictive maintenance more effective, it will perform simulations to identify problems in their early stages, before malfunctions occur. Data collected by sensors near the instruments—including, for example, acoustic, water flow, and temperature sensors—will be used as input for the simulations. The idea is that such data will include evidence of emerging condition problems.

Second, if a developing problem nevertheless goes undetected until it’s too late to prevent a malfunction, MAINTLET will offer an additional line of defense: it will improve reactive maintenance with trained failure detectors that can detect failures in real time, so that repairs can happen at the earliest possible time.

The MAINTLET system will comprise sensors; edge devices, such as Raspberry Pi computers, that support reactive maintenance services; WiFi and Zigbee access points and networks that interconnect sensors and edge and cloud devices; and a private cloud that supports predictive and preventive maintenance services. The project will initially focus on aging pumps, as pumps are ubiquitous in labs and fail frequently.

Nahrstedt thinks MAINTLET will offer two high-level benefits. First, the money saved on maintenance will translate into “thousands of dollars that could be then invested, maybe, in replacing some of the older equipment.” Perhaps certain expensive pieces could be upgraded, say, every 15 years instead of every 20. Second, lab staff today must often expend effort learning about long-obsolete systems instead of cutting-edge ones.

“Our labs have still scientific instruments that are running Windows NT!” she said. With more efficient maintenance, staff can study new technologies instead.  

The two-year, $1 million NSF grant is entitled “MAINTLET: Advanced Sensory Network Cyber-Infrastructure for Smart Maintenance in Campus Scientific Laboratories.” The Co-PIs include John Dallesasse and Mark McCollum (HMNTL), Gianni Pezzarossi (Engineering IT), and Mauro Sardela (MRL). Findings will be available at the project’s website, https://t2c2.csl.illinois.edu/maintlet/; MAINTLET software will be distributed as open source via GitHub.