Definition
IEPE (Integrated Electronics Piezoelectric) accelerometers integrate the piezoelectric sensing element with a preamplifier and output a voltage signal through constant current supply (usually 4mA). They require an IEPE data acquisition device or a gateway with IEPE input. Compared with MEMS, IEPE has a wider frequency response and a larger dynamic range, making it suitable for precise vibration monitoring.
Working Principle
IEPE uses a two-wire system: a constant current source (2–20mA, commonly 4mA) supplies power through a coaxial cable, the charge generated by the accelerometer is converted to voltage by the built-in amplifier, and output superimposed on a DC bias. The acquisition end extracts the AC signal via AC coupling and samples it using an ADC. The supply voltage must meet the sensor's operating range (usually 18–30V DC). The coaxial cable can be several hundred meters long and has strong anti-interference capability. Sensitivity is usually expressed in mV/g and needs to match the range of the acquisition device.
Comparison with MEMS
MEMS accelerometers are small in size, low in cost, and low in power consumption, suitable for embedded and high-volume applications, with a frequency response typically 0–several kHz; IEPE frequency response can reach 0.5Hz–tens of kHz, with a large dynamic range, suitable for precise vibration, FFT spectral analysis, and bearing fault diagnosis. IEPE is widely used in industrial predictive maintenance and rotating machinery monitoring; MEMS is commonly used in consumer electronics and structural health monitoring.
Use Cases
Rotating machinery (motors, fans, pumps, compressors) vibration monitoring; bearing and gear fault diagnosis; FFT spectrum analysis, envelope analysis; structural health monitoring; modal analysis. Edge gateways and data collectors need to provide IEPE input interfaces (constant current power + AC coupling + ADC), support multi-channel synchronous acquisition, and report data via Modbus, OPC UA, MQTT to the predictive maintenance platform.
Constant-current excitation and bias voltage
A piezoelectric sensor with built-in electronics is excited by a constant current. The vibration signal is an alternating voltage riding on a DC bias. The collector has to provide a suitable excitation current, take the vibration out with AC coupling or by removing the DC in digital processing, and scale it to acceleration by the sensitivity. Bias voltage is a health indication. Near the supply voltage, it is usually an open circuit. Near zero, it is usually a short, or excitation that was never sent. A normal bias only says the circuit is connected. It does not say the sensitivity direction or the mounting direction is correct. Sensitivity is given as a coefficient such as millivolts per g. The coefficient and the collector range have to be configured as a pair. Changing the probe without changing the coefficient scales the whole spectrum.
The conditions include an excitation voltage high enough to cover the cable drop, cable capacitance that does not filter off the high frequencies of interest, and a rigid joint between the sensor base and the surface being measured. A magnetic mount suits a walk-around check. It does not suit long-term high-frequency work. Wax or hand pressure is only for a rough look. When several channels are compared in phase, they should be sampled at the same time. Balancing a machine from channels polled one after another turns the time difference into a phase difference. For a low-frequency measurement, confirm the sensor's own lower frequency limit. The time constant of the AC coupling cuts off still more of the low end. Outside the probe's temperature range the sensitivity changes, the reading stays continuous, and there is no communication fault to report.
Cable capacitance and range saturation
Connecting this kind of probe to an ordinary voltage input, with no constant current, produces an output near zero or a drift, and the operator often thinks the probe is damaged. On a loop with the wrong current, the bias can fall outside the acquisition window, the alternating part is clipped, the time waveform becomes flat-topped, and the spectrum grows a string of harmonics that look a lot like mechanical looseness. If the range is chosen for normal vibration and no margin is left for a shock, the startup transient saturates, and a diagnosis treats that saturated spectrum as looseness or misalignment.
A ground loop comes from different potentials at the two ends of the shield. Supply frequency and its harmonics sit on top of the vibration and do not follow speed. Single-ended grounding and an insulated mount can show whether that is the path. An ordinary connector in the middle of the cable, instead of a coaxial one, brings in triboelectric noise, and the waveform jumps when the cable is shaken. Coiling a long cable into a small loop does not shorten its electrical length. The capacitance is still there. Several probes that share one excitation source, without the constant-current and isolation requirements met, cause crosstalk between channels. A report to the platform should include the bias, whether the channel saturated, and the sensitivity coefficient. One acceleration RMS value cannot say whether the number can be trusted.
Mounting stiffness and sampling channels together
The probe base and the surface being measured have to meet rigidly. A magnetic mount can be used for a walk-around. Long-term monitoring and high-frequency analysis should use a stud or a dedicated adhesive. Hand pressure and a thick paint layer filter off the high frequencies. The spectrum amplitude falls and the bias stays normal. When two channels are compared in phase for balancing, they must be sampled together. The time difference from polling is taken as a mechanical phase. The sensitivity coefficient travels with the probe. Changing the probe without changing the coefficient scales every peak together, and the trend looks like the operating condition changed.
Check excitation current and coupling against the collector port. An ordinary voltage port produces an output near zero. The range has to cover the startup shock. Harmonics after clipping look like looseness. Grounding the shield at one end can show whether supply frequency comes from a ground loop. Supply frequency does not follow speed. A non-coaxial joint in the middle of the cable makes noise when it is shaken, and that is easy to call an intermittent mechanical fault. Keep the bias, the saturation flag, and the coefficient next to every value on the platform. A single RMS value cannot be reviewed.
Judge bias voltage against that probe's typical range. Do not share one very wide threshold across every channel. If the threshold is too wide, weak excitation has already clipped the signal and the bias still looks normal. Cable capacitance limits the high frequencies that can be measured. After a longer cable is fitted, bearing high frequencies that used to be visible disappear. The measurement condition changed. The fault did not go away. Water in a connector, or shield strands touching the center conductor, looks the same as a damaged probe. Replace the cable before replacing the probe. When several channels share a mounting surface, one loose stud makes the phase of the neighboring channels abnormal together. Leakage in the collector's coupling capacitor drifts the low frequencies and lifts the spectrum near zero frequency. Write the bias, the cable length, and the mounting method into the point file, so a change of person or shift can still explain why the amplitude differs from last time.