Reliability, maintenance and OEE

A robot cell earns its keep only while it runs. A cell that meets its cycle time on the acceptance test but stops for twenty minutes every few hours for jams, faults and adjustments may produce far less than planned, and the payback calculation of Lesson 11 collapses. This lesson covers how manufacturers measure the effectiveness of automated equipment with overall equipment effectiveness (OEE), how reliability and availability are modeled, what maintenance strategies keep robots running, and how condition monitoring and data analysis catch problems before they stop production. Industrial robots themselves are very reliable, with manufacturers commonly quoting mean times between failures of tens of thousands of hours or more for the arm and controller; in practice, most cell downtime comes from the tooling, sensors, part presentation, upstream and downstream equipment, and programs that do not recover gracefully from abnormal situations.

Overall equipment effectiveness

OEE, popularized by Seiichi Nakajima as part of total productive maintenance (TPM) in Japan in the 1970s and 1980s, measures how much of the planned production time is truly productive. It is the product of three factors. Availability is the share of planned production time during which the equipment actually runs. Performance is the share of the ideal output achieved while running, capturing slow cycles and short stops. Quality is the share of output that is good first time.

Multiplying out the terms shows that OEE also equals the ideal cycle time multiplied by the good count, divided by the planned production time: the share of planned time that would have been needed to make the good parts at the ideal rate. The six big losses of TPM map onto the three factors: breakdowns and set-up or adjustment losses reduce availability; idling and minor stops, and reduced speed, reduce performance; start-up rejects and production rejects reduce quality.

An OEE of 85 percent is often quoted as world class for discrete manufacturing, a benchmark that dates back to Nakajima's work, though real figures vary widely by industry and many plants run between 40 and 70 percent. OEE is most useful for tracking one cell or line over time and for directing improvement at the largest losses, not for comparing different kinds of equipment. Because the three factors multiply, a modest weakness in each produces a surprisingly low total: 90 percent in each gives only 73 percent overall.

# OEE from a shift log, and a Pareto of downtime causes
log = {"planned_min": 450, "ideal_cycle_s": 52, "total": 405, "good": 391,
       "stops_min": {"gripper part missing": 14, "weld wire jam": 9, "fixture sensor": 6,
                     "upstream starved": 5, "robot collision alarm": 4, "changeover": 22}}
run = log["planned_min"] - sum(log["stops_min"].values())
A = run / log["planned_min"]
P = log["total"] * log["ideal_cycle_s"] / 60 / run
Q = log["good"] / log["total"]
print(f"A={A:.1%}  P={P:.1%}  Q={Q:.1%}  OEE={A*P*Q:.1%}")
total = sum(log["stops_min"].values()); cum = 0
for cause, m in sorted(log["stops_min"].items(), key=lambda kv: -kv[1]):
    cum += m
    print(f"{cause:22s} {m:3d} min  cumulative {cum/total:5.1%}")

A=86.7%  P=90.0%  Q=96.5%  OEE=75.3%
changeover              22 min  cumulative 36.7%
gripper part missing    14 min  cumulative 60.0%
weld wire jam            9 min  cumulative 75.0%
fixture sensor           6 min  cumulative 85.0%
upstream starved         5 min  cumulative 93.3%
robot collision alarm    4 min  cumulative 100.0%

A Pareto chart ranks causes by their contribution, reflecting the common observation that a few causes account for most of the losses. In the example, changeovers and missing parts in the gripper together account for more than half the downtime, which points to two improvement projects: reducing changeover time, perhaps by the single-minute exchange of die (SMED) techniques developed by Shigeo Shingo, and fixing the part presentation that leads to missed grips.

Reliability and availability

Reliability is the probability that a component or system performs its function without failure for a given time. During the useful-life period, when failures occur randomly at a roughly constant rate \( \lambda \), reliability follows the exponential distribution, and the mean time between failures (MTBF) is \( 1/\lambda \). Availability combines how often things fail with how long they take to fix, measured by the mean time to repair (MTTR).

Components in series, where any failure stops the cell, multiply their availabilities, so a cell's availability is lower than that of any of its parts. Redundancy, in which either of two parallel components can do the job, raises availability: the combination fails only when both are down at once.

This kind of calculation includes only repair time. In practice, a technician must notice the fault, walk to the cell, diagnose it, find a spare part and restart. Mean time to repair as experienced by production is often several times the hands-on repair time, which is why clear fault messages, spare parts stored near the line and operator training to handle simple recoveries have such large effects on availability.

The bathtub curve and failure modes

The failure rate of many products over their lives follows a bathtub curve: a falling rate of early failures caused by manufacturing defects and installation errors, a long period of low and roughly constant random failures, and a rising rate of wear-out failures at the end of life. For robots, early failures appear during commissioning and the first months, which is one reason for run-in periods before acceptance; wear-out affects gearboxes, bearings, cables, seals and batteries. The Weibull distribution, with a shape parameter below 1 for early failures, near 1 for random failures and above 1 for wear-out, is widely used to model these phases.

Maintenance strategies

Reactive maintenance, also called run-to-failure, repairs equipment after it breaks. It is sensible only for inexpensive, non-critical items with quick replacement. Preventive maintenance replaces or services components at fixed intervals of time or usage, following the manufacturer's schedule; robot manuals specify inspection, lubrication, battery replacement and cable checks at intervals such as every few thousand hours. Condition-based maintenance acts when measurements show that a component is deteriorating. Predictive maintenance goes further and forecasts when failure will occur, using trends and models, so that work can be scheduled at a convenient time before it happens. Reliability-centered maintenance (RCM), originally developed in the aviation industry, is a method for choosing the right strategy for each failure mode based on its consequences.

Robot manufacturers offer remote monitoring services that collect data such as motor torques, temperatures, gearbox condition indicators and fault histories, analyze it centrally, and alert users to likely failures. Users can also monitor their own data through OPC UA and MQTT, as Lesson 6 described. Useful indicators include the torque each axis needs for a repeated motion, which rises as gearboxes wear or lubrication fails; the time a vacuum gripper takes to reach its level, which rises as cups wear; the number of retries per hour on a vision pick; and cycle-time drift, which can reveal slow pneumatic actuators or degraded network communication.

Spare parts and maintainability

Spare parts strategy balances the cost of holding inventory against the cost of downtime. Critical, long-lead items such as a spare robot motor, a gearbox for the most heavily loaded axis or a spare gripper may be stocked even though they are expensive, because waiting days for delivery could cost far more in lost production. When failures occur randomly, the number of failures in a period follows a Poisson distribution, which allows planners to choose a stock level that covers demand with a chosen probability.

# How many spare grippers to stock? Poisson demand during the resupply lead time
from math import exp, factorial
n_grippers, mtbf_h, hours_per_year = 12, 6000, 6000
lead_time_weeks = 6
failures_per_year = n_grippers * hours_per_year / mtbf_h
mu = failures_per_year * lead_time_weeks / 52           # expected failures during resupply
def p_le(k, mu):
    return sum(exp(-mu) * mu**i / factorial(i) for i in range(k + 1))
print(f"expected failures per year: {failures_per_year:.1f}; during lead time: {mu:.2f}")
for stock in range(0, 5):
    print(f"stock {stock}: probability demand covered = {p_le(stock, mu):.3f}")

expected failures per year: 12.0; during lead time: 1.38
stock 0: probability demand covered = 0.250
stock 1: probability demand covered = 0.597
stock 2: probability demand covered = 0.837
stock 3: probability demand covered = 0.948
stock 4: probability demand covered = 0.986

Maintainability is designed in. Cells with accessible components, quick-change tooling, clear labeling, standardized parts across cells, good documentation and diagnostic messages that point to the failed component are faster to repair. A computerized maintenance management system (CMMS) schedules preventive tasks, records work orders and failure causes, and tracks spare parts, providing the data needed for MTBF and MTTR analysis. Program and configuration backups belong in the maintenance plan too: a controller failure or an accidental change should be recoverable in minutes from a known good version.

Continuous improvement

Reliability engineering and OEE tracking feed a continuous improvement cycle, often expressed as plan-do-check-act. When a fault recurs, structured root cause analysis, using tools such as the five whys or fishbone diagrams, looks beyond the immediate failure to the underlying cause. A gripper that drops parts might be traced to worn fingers, then to finger material unsuited to an oily part, then to a part change that no one told the integrator about. Fixing the root cause, here by improving engineering change communication, prevents a whole class of future failures. Operators who run the cell every day are often the best source of improvement ideas, which is a central principle of TPM.