Anyone who works with aerospace EHA testing knows the feeling: the oscilloscope captures every signal, but deciding what failed still comes down to manual file-by-file review.
An electro-hydrostatic actuator (EHA) is a core actuation component for flight control surfaces; the ailerons, rudder, and elevator all depend on it. The test center has already set up a centralized control platform, and process management and data consolidation run end to end. Each unit under test runs 40 to 200 minutes per batch with one sample point per second, capturing a dozen or more channels of electrical, hydraulic, control, and status signals. The problem is that this time-series data is saved as local files with no anomaly detection and no health assessment, and reading trend charts relies entirely on experience.
Even more troublesome: when one EHA’s tracking error runs high, it could be unit-to-unit variation or a material defect in the same batch of seals. The traditional approach can only confirm a batch problem after a second and third unit also throw alarms. By the time it is confirmed, whatever was going to be affected has already been affected.
There is an even more hidden case: a growing tracking error and a rising motor current both appear in seal defects and in controller software defects. The final-inspection readings are identical, but the responses are completely different. A seal defect means going back to the supplier; a software defect means changing the code. Judge the wrong direction and both time and cost are wasted.
At their core, these three situations are the same problem: the data has arrived, but the defect propagation chain has not.
Three pain points in EHA testing
A single unit’s anomaly is visible, but a batch defect cannot be confirmed
On the performance test line, the two-chamber pressure difference of the EHA-P01 unit under test falls from 25 MPa to 16 MPa, tracking error jumps from 0.05 mm to 0.28 mm, and product health score drops to 58. All of this is plainly visible on the oscilloscope. But P01 uses the same batch of seals as P02 and P03. Do they have the same problem? Local file storage cannot quickly compare multiple units side by side. By the time P02’s pressure difference also drops to 19 MPa and its health score to 75, then P03’s pressure difference drops to 22 MPa and its health score to 85, all three units have already shown trouble before anyone reacts. By then the batch pass rate has already fallen from 98% to 72%.
Motor current rises, but wear and software defects look similar
On the endurance test line, the motor current of the EHA-D01 unit under test oscillates from 85 A to a 65-145 A band, and tracking error rises from 0.05 mm to 0.22 mm. These same symptoms appear in seal wear and in controller software defects. The causal chain of seal wear is: seal degradation → internal cylinder leakage → pressure difference falling → pump compensation raising pressure → motor overload. The causal chain of the software defect is: excessive PID gain → control output oscillation → motor current and speed fluctuating violently → pump pressure pulsation → oil temperature spiking. The two chains look almost identical in final inspection, but the root causes are completely different, and so are the responses.
When one unit fails, whether the same model is affected stays uncertain
EHA-D01’s health score drops to 52 because of the PID calibration error in controller software V2.3.1. But D02 and D03 use the same controller model. D02 shows growing tracking error lagging 3 batches behind, and D03 only shows signs 6 batches behind. In the traditional mode, confirmation has to wait until they also throw alarms. But by the time D02’s health score drops to 72 and D03’s to 83, testing of that model has already been fully suspended.
Turning signals into judgments
Object modeling: turning measurement points into entities with operational meaning
EHA test data is not a pile of scattered points. The EHA-P01 unit on the performance test line has motor speed, motor current, pump outlet pressure, cylinder two-chamber pressure difference, tracking error, and product health score, and these attributes are physically coupled. Every 1 MPa drop in the cylinder two-chamber pressure difference lowers cylinder output force by 2 kN. Every 10 A rise in motor current raises pump outlet pressure by about 3 MPa. Only by binding these attributes to the same unit-under-test object, instead of scattering them across a dozen folders, is cross-attribute root-cause tracing possible.
More importantly, twelve EHA units under test are distributed across three test lines, and the consolidated assessment server in the quality assessment center keeps computing the anomalous-unit count, the batch pass rate, and the average health score. When the anomalous-unit count jumps from 0 to 1, the signal is not just “one unit has a problem” but “a batch of units may have a problem,” and that signal is invisible when you look at any single unit’s files.
Real-time analysis with event linkage: putting alarms back in process context
A traditional threshold alarm only tells you “the cylinder two-chamber pressure difference is below 20 MPa”; it does not tell you how that relates to tracking error, oil temperature, or motor current. TDengine IDMP’s approach is to read alarms back in their process context. When the consolidated assessment server counts more than two anomalous units, it triggers a same-batch anomaly linkage event. This is not an isolated numeric alarm but an operational judgment that “a same-batch or same-model defect may exist.”
For the endurance test line, when the motor current takes on an oscillating pattern (for example jumping from 85 A into a 65-145 A band), the AI anomaly detection algorithm recognizes the waveform pattern directly, with no need for a predefined hard threshold like “oscillation amplitude above 30 A.” Current oscillation itself is a strong signal that the controller output is unstable, which appears several minutes earlier than waiting for tracking error to exceed 0.15 mm.
Process analysis with AI-assisted insight: from experience-driven to evidence-driven
A seal defect is characterized by gradual accumulation: the cylinder two-chamber pressure difference declines slowly from 25 MPa, and the tracking error, oil temperature, pump outlet pressure, and motor current behind it drift in sequence, forming a cascade propagation chain. The root-cause attribute (pressure difference) has the longest anomaly window, and attributes farther from the root cause have shorter windows and longer lags.
A controller software defect is characterized by single-point failure plus cascade: tracking error and control error rise first, the motor current oscillates next, then motor speed, pump outlet pressure, and cylinder output force fluctuate in turn, and oil temperature rises last. The two causal chains look similar in final inspection, but the propagation order and the waveform signatures are completely different. Gradual accumulation is a monotonic drift; a software defect is oscillatory divergence.
AI anomaly detection can identify unfamiliar failure modes from anomalous patterns in the time-series waveform. But once it is recognized, the alarm has to be placed back in the object context of the asset tree, comparing data across same-batch and same-model units, before you can tell whether this is an individual problem or a fleet-wide problem.
Closing the loop on a typical anomaly scenario
Scenario 1: aileron EHA batch seal material defect
The seal supplier’s vulcanization process for that batch of fluororubber material was abnormal (curing temperature 8 °C too low), which caused microcracks in the piston seal that grew progressively under the high-pressure, high-frequency test conditions.
Step 1: Detection. On the consolidated assessment server dashboard, the anomalous-unit count rises from 0 to 1, triggering an L1 alarm.
Step 2: Trace back. Drill down to the performance test line and open the EHA-P01 cylinder two-chamber pressure difference trend chart. Starting from batch T-045, the pressure difference keeps falling from 25 MPa: first to 23.5 MPa, then to 18 MPa, and finally to 16 MPa, a 36% drop. At the same time, tracking error rises from 0.05 mm to 0.28 mm, oil temperature from 55 °C to 78 °C, pump outlet pressure compensates upward to 26 MPa, motor current compensates upward to 125 A, and motor temperature rises to 92 °C. The full causal chain is: seal crack → internal cylinder leakage → pressure difference falling → position accuracy falling → pump compensation raising pressure → motor overload → oil heating → health score deteriorating to 58.
Step 3: Confirm. Compare against the same-period data of EHA-P02 and P03. From batch T-049, P02’s pressure difference falls to 19 MPa, tracking error rises to 0.15 mm, and health score drops to 75. From batch T-057, P03’s pressure difference falls to 22 MPa, tracking error rises to 0.08 mm, and health score drops to 85. All three units show the pressure difference declining in sequence, confirming a same-batch seal defect. P01 is judged nonconforming; P02 and P03 need extended test observation.
Causal chain: abnormal seal material vulcanization process → microcracks in the piston seal → cylinder two-chamber pressure difference falling → position accuracy falling → pump compensation raising pressure → motor overload → oil temperature rising → product health score steadily deteriorating
Scenario 2: elevator EHA same-model controller software defect
Control software version V2.3.1 had a PID gain calibration error (Kp = 1.8, normal value 1.2). When the LVDT position sensor zero drift accumulated past 0.05 mm, the software bug was triggered, and the controller’s overcorrection made the output oscillate.
Step 1: Detection. On the EHA-D01 dashboard, the motor current trend chart shows violent oscillation: jumping from 85 A into a 65-145 A band, an amplitude of ±40 A. The AI anomaly detection algorithm recognizes the abnormal current waveform and triggers an alarm.
Step 2: Trace back. Open the tracking error and control error trend charts. Tracking error keeps growing from batch D-020, rising from 0.05 mm to 0.22 mm; control error rises from 0.03 mm to 0.18 mm. Continue drilling down: motor speed fluctuates to 2200-3800 rpm, pump outlet pressure pulsates to 17-25 MPa, cylinder output force fluctuates to 35-65 kN, and oil temperature rises to 82 °C. The full causal chain is: excessive PID gain → zero drift accumulation triggers the bug → controller overcorrection → motor current and speed oscillate violently → pump pressure pulsation → cylinder displacement and output force oscillation → oil temperature spiking → health score plunging to 52.
Step 3: Confirm. Compare against the same-period data of EHA-D02 and D03. From batch D-025, D02’s tracking error rises to 0.12 mm, motor current oscillates at 75-95 A, and health score drops to 72. From batch D-031, D03’s tracking error rises to 0.08 mm, motor current fluctuates mildly at 80-90 A, and health score drops to 83. All three units use the same controller model, confirming a same-model software defect, and testing of that model is fully suspended. D01 is judged nonconforming; D02 and D03 need to upgrade the software to V2.3.2 and retest.
Causal chain: control software PID gain calibration error → LVDT zero drift accumulation triggers the software bug → controller overcorrection and output oscillation → motor current and speed oscillate violently → pump pressure pulsation → cylinder displacement and output force oscillation → oil temperature spiking → product health score plunging
From manual file-by-file review to tracing causes
The ceiling on the value of EHA test center digitization is not set by how many signal channels are captured or how many files are stored, but by whether the signals can be turned into an actionable judgment.
- Object modeling: bind the measurement points scattered across files to the unit-under-test object, so the physical coupling between attributes becomes visible
- Real-time analysis with event linkage: move from “a value exceeded its limit” to “a batch of units may have a problem”, and only with alarms read back in process context can you tell seal wear from a software bug
- Process analysis with AI-assisted insight: move from “root cause by experience” to “causal chain traced through data,” reading the pressure-difference window length for gradual accumulation and the current waveform pattern for oscillatory divergence
TDengine organizes test signals into objects with operational meaning, turns threshold alarms into explainable process judgments, and turns data capability into quality-control capability that supports anomaly detection, same-batch tracing, and same-model early warning.
TDengine comes with a high-performance, distributed time-series database, Industrial Ontology modeling, and an Industrial Agent Runtime, providing a full-stack solution for industrial data streams from collection and storage to real-time analytics, visualization, event management, and root-cause analysis. To learn more about TDengine, visit www.tdengine.com and try it for free.
Try it yourself
Install and deploy TDengine Visit the TDengine Download Center, select TDengine All-in-One, choose the deployment platform and architecture that matches your environment, and follow the guided steps to complete the installation.
Load the sample data
On first activation, choose Servo System Time-Series Data Monitoring and Analysis on the sample data loading screen and wait for it to finish loading.
If you have already activated the product, click your avatar in the top-right corner, select Management Console, choose Sample Data on the left, then select Servo System Time-Series Data Monitoring and Analysis. Wait a few minutes for the data to load.
Servo System Time-Series Data Monitoring and Analysis
