I used to work for a contract manufacturer building extremely sensitive sensors for a customer. These sensors had a finite life, and they were replaceable in the field. The customer planned to keep spares on hand, so we received a large order for spares.
When we were about 10 percent into delivering the spares order, we started to see field returns coming back. The reported defect was “erratic behavior shortly after installation” and “operating temperature out of spec.” Field returns were covered under warranty unless we could assign the failure to mishandling or some other customer-related fault.
The design of these sensors was such that they would operate correctly only if they were within a fraction of a degree Fahrenheit of the ideal temperature. The sensor was very well insulated, and the operating temperature was controlled in two zones by electrical resistance heaters. The temperature of each zone was sensed by a Wheatstone bridge containing two glass-bead thermistors. Each bridge output was connected to a PID (proportional integral derivative) controller that was in the customer's system, not in the sensor. There was also a monitoring thermistor used as independent temperature readout, for a total of five identical thermistors per sensor.
A glass-bead thermistor is a pretty simple device. It has platinum wires that are stretched close together on a frame, and beads of ceramic slurry are applied between them. The frames are then fired, and the “string” is clipped to form individual thermistor beads with two leads coming out one end. A glass tube is then placed over each bead and melted around it to form a hermetic seal, with the leads sticking out. The high reliability (hi-rel) ones are put through a thermal-shock test, moving them back and forth repeatedly between 400°F oil and liquid nitrogen. Only the strong survive.
The four control thermistors were secured into blind holes in the metal body of the sensor with epoxy, and the monitoring thermistor was surface-mounted to the outside of the sensor body (inside the insulation) with the same epoxy. The epoxy was selected to cure at the same temperature as the operating temperature of the sensor.