When a power outage in 2021 shut down a fab for six hours, it affected 30,000 semiconductor wafers and cost the company about $33.85 million, or $5.64 million/hour. This unplanned downtime was not the result of equipment breakdown and the cost incurred was purely from the lost wafers in production. But it goes to explain why, even with varying costs of such disruptions across different types of manufacturing operations, organizations are keen to avoid unplanned downtime.
Is preventive maintenance enough?
Preventive maintenance to avoid breakdowns is akin to regularly taking a car to the service center. The service schedule is based either on a regular time interval, usually a year, or better still the number of miles the car has been driven. Traditional preventive maintenance certainly many advantages over reactive maintenance — repairing equipment when it breaks down or when smoke is perceived — saving both equipment and downtime costs. Regular maintenance means higher upfront costs and more time spent planning and budgeting, but the resulting predictability helps increase production efficiency, lowers losses due to unplanned outages during time-sensitive processes, increases equipment life, and enables businesses to manage costs and needed profitability.
At first glance preventive maintenance seems enough, yet it leaves much to be desired. Continuing with the car analogy, the manufacturer-set servicing requirements are at best averages of anticipated driving conditions. They do not account for how one drives a car, the number of hard stops and starts, or the load normally carried by the car. As averages go, they cover everything without meeting anything, any condition, specific to a car. That is why, manufacturing equipment, just like cars, still breaks down in the middle of nowhere in a manufacturing process. At best, preventive maintenance’s set schedule may lead to unnecessary maintenance of equipment and, therefore, predicted but unnecessary costs and downtime.
Earlier maintenance with condition monitoring
This reactive maintenance strategy utilizes microphones, heat sensors, and smoke detectors, to alert companies when failure is imminent and response urgent, however disruptive it may be to production processes and schedules. Maintenance at this stage often also demands higher-cost part replacement and repair.
Better understanding of equipment lifecycle has led to improvements on this concept and other parameters added for monitoring (Figure 1). Since industrial equipment prone to breakdowns typically involves moving parts — motors and pumps used in conveyors, robots, and fluid supply — the estimated equipment life stages to failure shown in Figure 1 relate closely to such equipment.
Organizations can peer earlier into fault conditions with the monitoring of additional parameters and undertake preventive or even predictive maintenance. Widely used sensors for such condition monitoring include the following.
IR sensors/cameras: IR sensors are at the heart of thermography equipment used to continuously or frequently check equipment temperature. This enables detection of “hot spots” that result from such faults as fuses near failure and sparking from incorrect or corroded electrical terminations. Lower-cost implementations may use RTD and thermocouples.
Particle counters: Oils are used not only in hydraulic systems but as lubricants in gear boxes, transmissions, and motor bearings across nearly all industrial processes. Particle counters help determine the contamination that occurs as oils break down or pick up detritus along their journey smoothening motion. Particle count data thus informs about equipment condition, such as abrasive bearing wear.
Automatic optical particle counters are among the most widely used methods and utilize white light or lasers to detect shadows or light scattering for counting. The equipment is often certified to ISO 11500 and the codification of the particle count data is determined by ISO 4406:99 so that consistent assessments are achieved.
Energy monitoring: Variation in motor current, speed, and power result from variable-speed drives used in industry to improve electrical efficiency in HVAC systems, robotics, conveyors, and other motion-dominated applications.
Energy monitoring and trends analysis at various points in a process or at particular equipment can reveal important information about equipment health. Anomalous energy consumption patterns often indicate either unsafe operating conditions or an ongoing fault condition, say due to worn-out bearings or electrical wiring issues. Even a cheaper, direct measurement motor current can inform about eccentric rotors, rotor bar issues, winding issues, and bearing problems.
Vibration sensors: Vibration and attending data analysis offers up critical information on bearing condition, gear meshing, pump cavitation, rotor misalignment, and load condition. Piezoelectric or microelectromechanical system (MEMS) accelerometers and sensor modules, like the three-axis ADcmXL3021 MEMS module, are easily added to equipment. The information they extract relates so closely to underlying problems and so early in fault development (Figure 1) that equipment life can be extended without significant overhead.
Vibration sensing and assessment is governed by several standards, including ISO 5348 for mechanical mounting of accelerometers, ISO 10816 for evaluating machine vibration by measurements on non-rotating parts, and ISO 7919 for assessing machine vibration by taking measurements on rotating shafts.
Ultrasonic sensors/microphones: Acoustic sensors for the ultrasonic range can detect barely noticeable pressure leaks, bearing problems, gear meshing, and pump cavitation. They are considered the first line of defense because they give a very early warning of a potential problem, such as by detecting slight increases in friction in rotating equipment.
Both electret and MEMS microphones are available for use in harsh and relatively benign industrial conditions, respectively.
Analyzing sensor data for predictions
While the measurements provided by sensors are useful for diagnosing problems, setting up automated alert thresholds is challenging. For instance, any variance in the nature of the process or production recipe and even changing environmental conditions due to changing weather conditions with seasons of the year can cause false alarms. Industry therefore uses multiple sensors to not only track equipment through its entire life but to ensure there are no informational gaps in condition monitoring.
Sensors and threshold alerts focus on machine status at the time of measurement. Predictive maintenance on the other hand focuses on fault detection before it becomes obvious and, with extensive analysis, even before the inception of the fault so that it can be avoided altogether.
For instance, platforms like the iCOMOX leverage an array of technologies to help build out a predictive maintenance strategy. The platform utilizes the low-power ADuCM4050 ARM Cortex M4 MCU, low-noise ADXL356 three-axis accelerometer as a vibration sensor, ADXL362 accelerometer as a low-g wake-up trigger, ADT7410 temperature sensor with ±0.5°C accuracy, a Bosch BMM150 three-axis magnetic field sensor, and an IM69D130 digital dual-backplate MEMS microphone.
Embedded artificial intelligence (AI) in iCOMOX prioritizes and rationalizes sensor data so that lower data flow and lower power consumption are achieved from the edge to the cloud where the analytical heavy lifting is usually done.
When multivariate analysis is brought to bear on combined condition and process historical data, fault prediction and predictive maintenance precision is increased. Deployment of deep learning algorithms can help create and bring together process models with equipment health/life models to account for changing machine, process and environmental conditions. This allows manufacturers to predict and plan maintenance when it least impacts productivity.
Onwards to self-maintenance?
Even as predictive maintenance is going mainstream, leading equipment companies are already eyeing self-monitoring of machines as well as moderate levels of self-maintenance. For instance, in the semiconductor etching equipment industry, tools can detect the presence of the electrostatic chuck on which wafers are placed, check its alignment and calibration, and ascertain its condition. Replacing the chuck with a new one from in-built storage is a matter of relatively simple automation utilized with consideration of ongoing processes and upcoming opportunities for a planned downtime.
Ushering in this future are specialists at Arrow, advising engineers about sensors, compute platforms and cloud services as well as leveraging a vast ecosystem of partners who support optimized operations by enabling predictive maintenance.
Contact Arrow to upgrade your manufacturing operations.