Why Listen to Your Products

Every mechanical assembly produces sound. A properly built motor hums at predictable frequencies. A worn bearing generates broadband noise with characteristic spectral peaks. A misaligned gear set introduces tonal components that shift with speed. These acoustic signatures carry information that other test methods simply cannot capture — and that is what makes acoustic testing one of the most powerful tools available for end-of-line quality verification.

Traditional quality methods like dimensional inspection, torque checks, and visual examination catch many defects. But they often miss problems that only manifest during operation: a bearing with a microscopic surface defect, an eccentric rotor, a loose lamination stack, or a gear tooth with insufficient surface finish. Acoustic testing catches these issues because it evaluates the product doing what it was designed to do — running.

How Acoustic Testing Works

At its core, acoustic testing compares the sound signature of a product under test against a known-good reference. The process follows a straightforward sequence:

Signal acquisition. One or more microphones or accelerometers capture the sound or vibration produced by the product during a controlled run cycle. Microphones measure airborne sound pressure, while accelerometers measure structure-borne vibration directly on the housing or mounting surface. The choice depends on the application — accelerometers provide better signal-to-noise ratios in loud factory environments, while microphones offer non-contact measurement.

Signal conditioning and digitization. The raw analog signal passes through anti-aliasing filters and an analog-to-digital converter. Sample rates typically range from 20 kHz to 100 kHz depending on the frequency range of interest. For bearing defect detection, higher sample rates are often necessary to capture the high-frequency energy associated with surface imperfections.

Frequency analysis. The digitized signal is transformed from the time domain to the frequency domain using Fast Fourier Transform (FFT) algorithms. This produces a spectrum showing amplitude versus frequency, which reveals tonal components, harmonics, and broadband noise floors. Advanced systems also compute order analysis, envelope spectra, and cepstral analysis for specific defect signatures.

Classification. The measured spectrum is compared against acceptance criteria. These criteria might include overall sound pressure level limits, individual frequency band limits, order-related amplitude thresholds, or pattern-matching algorithms trained on known-good and known-defective samples. Machine learning classifiers have become increasingly common, using training datasets of several hundred to several thousand samples to build robust pass/fail models.

Common Applications

Acoustic testing is particularly effective for products with rotating or reciprocating components:

  • Electric motors — detecting rotor eccentricity, stator winding faults, bearing defects, and lamination looseness across the full speed range
  • Pumps and compressors — identifying cavitation, impeller imbalance, valve leakage, and excessive clearances
  • Gear assemblies — catching tooth profile errors, surface finish problems, misalignment, and backlash issues
  • Bearings — finding raceway defects, cage damage, contamination, and lubrication problems before they cause field failures
  • HVAC blower assemblies — verifying airflow noise levels meet specifications for residential and automotive comfort standards
  • Power tools — confirming that gear whine, motor noise, and overall sound quality meet consumer expectations

In the automotive supply chain, NVH (Noise, Vibration, and Harshness) testing has become a standard end-of-line requirement. Tier 1 suppliers shipping seat motors, window regulators, HVAC actuators, and electric power steering units routinely perform 100% acoustic inspection to meet OEM specifications.

Test Station Design

A well-designed acoustic test station balances measurement quality with production throughput. Key design considerations include:

Acoustic isolation. Factory environments generate significant background noise from conveyors, pneumatics, and adjacent equipment. Test enclosures with sound-absorbing lining (typically 20-30 dB insertion loss) isolate the product from ambient noise. Semi-anechoic chambers are used when free-field measurement conditions are required by the specification.

Fixturing and drive systems. The product must be mounted and driven in a way that represents its installed condition without introducing extraneous noise. Vibration-isolated fixtures, low-noise drive motors, and compliant couplings prevent fixture-borne noise from contaminating the measurement. For motor testing, dynamometers or controlled loads simulate real operating conditions.

Run cycle control. Most acoustic tests require the product to operate through a defined speed profile — a ramp-up, steady-state hold, and ramp-down sequence. The controller must synchronize data acquisition with the speed profile so that order analysis can separate speed-dependent tonal components from broadband noise.

Cycle time management. Production acoustic tests typically run between 3 and 15 seconds depending on the product and the defect types being targeted. Faster cycle times are possible when defects produce strong signatures, but some defects only appear at specific speeds or load conditions and require longer run profiles.

Integration With Automated Lines

Acoustic test stations rarely operate in isolation. They integrate into larger automated assembly and test systems where products flow from assembly operations through multiple test stations before reaching packaging. Integrating acoustic testing into these lines requires coordination with the material handling system — conveyors, pick-and-place units, or robotic transfers that move products into and out of the test fixture.

Data integration is equally important. Each test result must be linked to the product's serial number or lot identifier and stored in a central database. This traceability allows engineers to correlate acoustic test failures with upstream process parameters — for example, connecting a spike in bearing noise rejections to a change in grease supplier or press-fit force on a specific assembly station.

Modern acoustic test systems also support real-time statistical process control. Rather than simply pass/fail sorting, they track trends in acoustic parameters over time. A gradual increase in a specific frequency component might indicate tooling wear in an upstream machining operation, enabling corrective action before the defect rate increases.

Machine Learning in Acoustic Testing

Classical acoustic testing relies on expert-defined frequency band limits and threshold values. Setting these limits requires deep knowledge of the product's acoustic behavior and the spectral signatures associated with each defect type. This approach works well for products with clearly defined failure modes but becomes difficult when defect signatures overlap or when products have high unit-to-unit variation.

Machine learning classifiers address this challenge by learning the boundary between acceptable and defective products directly from labeled training data. Support vector machines, neural networks, and ensemble methods can all be effective. The key requirement is a training dataset that adequately represents the full range of normal variation as well as the defect types that must be detected.

In practice, building a robust training dataset is the most time-consuming part of deploying a machine learning acoustic test. Normal production samples are readily available, but defective samples for each failure mode must be collected or intentionally manufactured. A common approach is to start with classical threshold-based testing, collect a library of defective samples over several months of production, and then train a machine learning model that improves detection rates and reduces false rejects.

Getting Started With Acoustic Testing

For manufacturers considering acoustic testing, the process typically starts with a feasibility study. This involves collecting acoustic data from a set of known-good and known-defective products, analyzing the spectral differences, and determining whether reliable classification is achievable. Not every defect produces a detectable acoustic signature — and not every product benefits from acoustic testing. The feasibility study answers these questions before committing to a full production system.

Once feasibility is confirmed, the next steps involve specifying the test station hardware, developing the analysis algorithms or training the classifier, and integrating the station into the production line. The entire process from feasibility through production-ready deployment typically spans several phases of development, validation, and refinement.

Partner With AMD Machines

AMD Machines designs and builds automated test systems that include acoustic, functional, and dimensional inspection — integrated into turnkey production lines. Our engineering team has experience across motor, pump, gear, and bearing applications where acoustic testing delivers measurable quality improvements. Contact us to discuss your testing requirements.