Humanoid Robot Joint System
Near-field Acoustic Holography Inspection Solution

A Specialized Acoustic Inspection Solution Based on the NAH1024 Ultra-Large Planar Array

I. Industry Background and Market Opportunity: The “Quietness Bottleneck” of Mass Production and a Blue Ocean in Inspection

Humanoid robots are accelerating from the laboratory toward industrialization and real-world deployment. As the core component of motion execution, the joint module directly determines motion smoothness, operational stability, and the human–robot interaction experience. According to industry research, more than 60% of humanoid robot operational failures are related to joint module vibration and noise. Vibration and noise have become a key pain point constraining their scaled mass production, and in particular they limit deployment in scenarios with high quietness requirements, such as healthcare, home service, and precision manufacturing.

The industry has entered an early commercialization stage. According to forecasts by industry research institutions, China’s humanoid robot shipments will reach approximately 50,000 units in 2026 and are expected to reach 446,000 units by 2030, with a compound annual growth rate (CAGR) of approximately 106% from 2025 to 2030, corresponding to a market size growing from approximately USD 2 billion in 2026 to approximately USD 15 billion in 2030. Core components such as harmonic reducers are ramping up in step, with leading manufacturers’ shipments expected to exceed 1.5 million units in 2027. Demand for acoustic quality inspection will surge in tandem with the scaled volume growth of joint modules.

It is worth noting that mainstream solutions on the market today focus mostly on general-purpose acoustic measurement or on the performance, durability, and precision testing of joint modules; specialized acoustic NVH inspection for humanoid robot joints remains a market blank. This blank is precisely the value entry point of this solution.

II. Solution Positioning: Specialized Acoustic NVH Inspection for Humanoid Robot Joints

Centered on a single NAH1024 ultra-large planar array, this solution focuses on precise acoustic localization at the joint and component level, concentrating resources on the core technology of near-field acoustic holography to deliver deeper, more specialized inspection capability for low-frequency source tracing. Its differentiation is built along three dimensions:

• Joint specialization—acoustic inspection aimed at components specific to humanoid robot joints—harmonic reducers, frameless torque motors, planetary gearboxes, roller screws, and others—rather than general-purpose acoustic measurement;

• A dedicated mid- and low-frequency array—a 1024-channel high-density array dedicated to solving the difficult problem of low-frequency source tracing in joints, breaking through the localization bottleneck of the “meter-scale spot” produced by conventional beamforming;

• Cost-effective domestic substitution—independently controllable holographic sound field technology providing domestic inspection capability with a better cost-performance ratio than imported solutions.

User profiles:

III. Technical Rationale: Why NAH Is Suited to Humanoid Robot Joint Inspection

The physical structure of the humanoid robot joint module determines the unique difficulties of its acoustic inspection—and these difficulties happen to fall squarely within the capability range of near-field acoustic holography technology. The following demonstrates the logic of this technical fit from the perspective of physical mechanism.

3.1 Densely Packed Sources: The “Microscopic Blind Zone” of Conventional Acoustics

The joint module highly integrates a frameless torque motor, a harmonic reducer, an encoder, and support bearings within a housing only tens to just over a hundred millimeters in diameter. The potential sources—the flexspline/circular spline meshing point, the flexible bearing, the motor stator and rotor, and the output bearing—are separated from one another by only a few millimeters to a few centimeters. Multiple sources overlap heavily in space to form a densely packed source cluster. Conventional acoustic imaging cameras and beamforming techniques can hardly distinguish individual sources within such a compact space, giving rise to a clear “microscopic blind zone.”

3.2 Low-Frequency Source Tracing: A Physical Necessity for Breaking the Diffraction Limit

Humanoid robot joint noise is dominated by low-frequency structural excitation: the flexspline deformation fundamental frequency of the harmonic reducer and its mesh stiffness fluctuation (2–4× the input rotation frequency), motor cogging torque fluctuation, and joint resonance frequencies are mostly concentrated in the low-frequency band of tens to several hundred hertz, while meshing frequency sidebands are distributed in the mid-frequency band. Low-frequency acoustic wavelengths can reach the meter scale, whereas the spatial resolution of conventional beamforming is constrained by the Rayleigh diffraction limit (on the order of half a wavelength). Localization spots for low-frequency sources therefore appear at the “meter scale”—large and blurry—and cannot lock onto the excitation point.

Near-field Acoustic Holography (NAH) fundamentally breaks through this limit: in the near field of the source, a high-density array simultaneously measures propagating waves and evanescent waves, and the sound field is reconstructed through an inverse propagation mathematical transform, achieving centimeter-level source localization accuracy (sub-wavelength resolution is attainable under ideal near-field and high-signal-to-noise-ratio conditions). This carries low-frequency source tracing across the gap from “meter-level blur” to a “fine origin.”

3.3 Evanescent Waves and High-Density Sampling: The Physical Core of NAH

The prerequisite for capturing evanescent waves is satisfying the spatial Nyquist sampling requirement—the spacing between array microphones must be small enough to recover the high spatial frequency components of the source. The NAH1024 ultra-large planar array uses 1024 microphones at approximately 30 mm spacing to form a 960 mm × 960 mm high-density sampling plane, which exactly satisfies the sampling density required for reconstructing mid- and low-frequency evanescent waves, providing the physical basis for breaking through the diffraction limit.

3.4 Transient Meshing Impacts: Millisecond-Level Synchronous Snapshots

Meshing impacts caused by tooth profile errors and tooth backlash are transient, broadband excitations that appear instantaneously only at specific phases and under specific operating conditions. If point-by-point scanning measurement is used, serial acquisition will lose transient features and produce spatial localization deviation due to time-varying operating conditions. The millisecond-level synchronous “one-touch snapshot” of the NAH1024’s 1024 channels completes full-field data acquisition with a single trigger, fully capturing transient meshing impacts and eliminating the localization error introduced by scanning movement.

3.5 Statement of Technical Boundaries: Rigorous Definition Is the Prerequisite for Delivery

As a solution founded on R&D rigor, we define the capability boundaries of the NAH1024 forthrightly:

IV. Core Inspection Targets and Typical Scenarios

The above targets focus on the joint module and its core components, covering core joints such as the shoulder, elbow, wrist, hip, knee, and ankle as well as the dexterous hand, and spanning the two mainstream configurations of rotary and linear actuators. For closed joint assemblies, the “surface hotspot + structural inference” approach is adopted; for dismantled components, direct near-field measurement is performed.

V. Application Value: Building a Closed Acoustic R&D Loop for Humanoid Robots

5.1 R&D: From “Black-Box Trial and Error” to “Precision Targeting”

Conventional noise reduction relies on empirical iteration: modifying tooth profiles, substituting materials, and adjusting assembly all require repeated trial-and-error validation. The three-dimensional sound pressure / particle velocity / sound intensity contour maps output by NAH can point directly to acoustic radiation hotspots on the structural surface; combined with structural analysis to infer the excitation source, noise reduction shifts from “trial and error” to “targeting,” significantly shortening R&D cycles and reducing development costs.

5.2 Supply Chain: Quantitative Batch Consistency Evaluation of Reducers

Harmonic reducers are a key component in domestic substitution, and differences in batch consistency translate directly into whole-machine NVH dispersion. The quantitative acoustic field fingerprints provided by NAH enable batch-by-batch acoustic consistency evaluation of incoming reducers, rapidly identifying batch-level defects such as tooth profile deviation and assembly tightness and gatekeeping the quality of core components.

5.3 Process: Acoustic Reverse Verification of Assembly Quality

Assembly deviations such as shaft train parallelism, bearing clearance, and preload significantly alter sound field characteristics. NAH inspection can serve as a means of reverse verification of assembly quality, inferring the type and location of assembly deviation from sound field characteristics and guiding the standardization of assembly processes.

5.4 Operation and Maintenance: Predictive Diagnosis of Early Wear and Lubrication Failure

Early bearing wear and lubrication failure manifest weak acoustic signatures in the sound field before indicators such as vibration and temperature rise do. The high sensitivity of NAH can capture these weak acoustic precursors during the incipient stage of a fault, supporting predictive maintenance of humanoid robot joints, extending the service life of core components, and reducing O&M costs.

5.5 Standards: Supporting the Development of Acoustic Quality Standards for Humanoid Robots

No unified acoustic quality standard yet exists for humanoid robot joint modules. The quantitative, reproducible three-dimensional sound field data provided by NAH can serve as the technical foundation for standard development, advancing the industry from “empirical judgment” toward “data-driven standardization.”

VI. Technical Parameters

6.1 NAH1024 Ultra-Large Planar Array

Notes: (1) The sensitivity figure is quoted as −24 dB; its reference basis (e.g., re 1 V/Pa or dBFS) is to be specified. (2) The 65 dB SNR figure is applicable to the microphone array; whether it is a per-channel, array-level, or system-level value is to be specified. (3) The “50 Hz – 5 kHz” effective reconstruction band is bounded by the array’s spatial sampling limit (approximately 5.7 kHz at 30 mm spacing), as described in Section 3.5.

VII. Summary

The acoustic quality of humanoid robots has become a key bottleneck on the path to mass production and deployment in quietness-sensitive scenarios, and what the industry currently lacks is precisely the inspection capability of “precise localization and source tracing.” By virtue of three physical advantages—“capturing evanescent waves, breaking through the diffraction limit, and high-density synchronous sampling”—near-field acoustic holography technology is naturally suited to the acoustic characteristics of humanoid robot joints—“densely packed, low-frequency dominant, transient excitation”—and can upgrade noise reduction from “black-box trial and error” to “mechanism-level precise localization.”

This solution is a single-point, in-depth solution centered on the NAH1024 ultra-large planar array. It focuses on the market blank of “specialized acoustic NVH inspection for humanoid robot joints,” covers five value dimensions—R&D source tracing, supply chain evaluation, assembly verification, O&M diagnosis, and standards development—and ensures deliverability through rigorous definition of technical boundaries, helping joint modules achieve “quiet, stable motion” and accelerating high-quality mass production of the industry.