Over the years I have been asked about automation of production line quality control inspection – from incoming parts to outgoing finished goods. While there are more than a dozen audio analyzer vendors that have QC test suites, most of these have scripts that at a touch of a button (or footswitch) will trigger a Go/No-Go test. But what is missing is getting the parts, such as a woofer, off the production line and into one of Geoff Hill’s Tetrahedral Test Chambers - which enable stable and accurate measurements of loudspeakers without using an anechoic chamber - or placing a headphone onto an artificial ear, or a mic onto a reciprocal test jig. In all those situations, an operator is required to repeat the same gestures countless times... for days, weeks, months, and years.

In the last couple of years, factories have learned that the Achille’s heel in a smooth production flow is the incoming supply chain and stable employee staffing. While getting incoming parts suppliers under control can be out of reach of most purchasing departments, streamlined production lines require workers who don’t get COVID, will work 24-hour shifts 7 days a week, and don’t ask for raises.

What if instead of end-of-line people moving product to the test jigs, sorting and binning, or packing, we could use a robotic arm using its vision camera not just for guiding the arm to pick up and place the component, but also inspect the paint finish and then, after the analyzer does its audio pass/fail, get the device under test (DUT) back onto the Good or Not Good bin or even pack it up. While there are reasons why this is rare for audio contract manufacturers, Sanezoo has been perfecting universal visual inspection and flexible bin-picking solutions that are viable for the situations described. In fact, Sanezoo’s artificial intelligence vision systems integrated into robot arms are already in use by VW, DuPont, Bosch, and others.

At CES each year, Menlo Scientific has a technology showcase of solutions for the audio industry at the Venetian. Since by the new year most of us felt the dark clouds had passed, this year Sanezoo joined us in our CES suite and the ALTI convention floor hospitality suite to explore expanding its focus from the automotive industry to audio products, such as loudspeakers. Sanezoo’s platform enables manufacturing quality control and collision-free robot guidance by obtaining and generating data for continuous improvement of AI and analytical algorithms, thereby enabling better decisions, performance, quality, and safety.

Sanezoo’s focused use cases include 3D bin picking and precise placing to solve the unpredictability of loose parts and complex geometries of products on the conveyor belt, complemented with AI-Powered 3D surface inspection solutions on challenging surfaces. These applications are the result of passive 3D vision with intelligence powered by smart algorithms tuned for GPUs efficient at manipulating image processing due to their parallel structure and deep learning methods. The two image sensors of the 3D camera resemble two human eyes, while the intelligence resembles part of the human brain capacity.

A single-shot stereo camera enables a quick scan (<1ms), so that the scene can be moving in respect to the camera, and it is naturally fast. In comparison with other methods, it does not use a series of patterns shot at the scene. The vision system knows how the parts are oriented and picks and places them accurately as required by the next process steps. It does not have to use any special light, operating with lighting conditions from dark robotic cells to direct sunlight. It can also handle a variety of material, from shiny reflective metal to dark light absorbing rubber.

Part of the reason why robot arm vision picking systems have not yet been widely adopted by the consumer audio industry is related to the limitations in accuracy or speed of existing systems. With smarter vision and faster image capture, the optimum picking point can be identified with a collision-free path - with other parts, the container, the robot, and the environment - since the path is computed for each move. Typically, the 3D camera is placed on the robotic flange and can take a snapshot of the contents of the bin with a single shot. Dense high-resolution point clouds (3D representation of the scene) allow for precise 3D matching of the parts. The precision and speed are enabled by using a single passive 3D stereo camera, empowered with Sanezoo's smart algorithms and AI.

Of course, cost is another major consideration but is a fraction of what this level of sophistication was just a few years ago, and now without the flaws for a viable stable installation. An AI vision conveyor/bin-picking system that guides the robot arm is between $25k and $30k. This cost covers the eyes and some of brain of the system. This will mount to a 6 degrees of freedom (6DofF) robot arm.

The whole feeding robot cell for manipulating parts including the 6DoF robot, safety, PLC, gripping, automation, and mechanical parts as well as integration and programming will be around $100k. Then there is the cost of the audio analyzer along with another subsystem for cosmetic QC for surface visual quality control. Not cheap, but a fast worker that can go 24/7, will last for years, and won’t call in sick or quit has its appeal and ROI.

3D laser or structured light vision systems normally involve a number of challenges, from creating point clouds with inaccurate points and many outliers (invalid points); can be too big and heavy for fixing on the robotic arm, while the process scene cannot be moving.

Many 3D defects cannot be seen clearly or detected reliably by 2D cameras and are invisible to 3D scanners. In addition, some surfaces are shiny and are almost impossible to be scanned or photographed without disturbing light reflections. Light reflections are usually handled by special lighting techniques, such as dome light. These approaches, however, sweep away the fine 3D structure of materials. This is undesirable for the detection of small surface defects.

Many surfaces are far from uniform. For example, metal milling produces a variety of trochoidal marks, which are not defects. Such difficulties are not possible to address without the construction of special sectional lighting and advanced image analysis techniques.

We mentioned milling marks on the surfaces are well distinguished, together with their 3D structure. This enables the system to be tuned for classifying the marks as OK or defective, which depends on the tool wear progress, as well as how other surface artifacts can be handled.

The individual use of artificial intelligence methods, in combination with standard approaches used in machine learning, does not produce as good results as when combined with the sectional light of the 3D fine surface inspection system. One of the proprietary developments is a suitable design of sectional light with enough sections in combination with the composition and analysis for effective defect evaluation.

The sectional light must be designed in terms of geometry and hardware design (suitable LEDs, microcontroller control, and timing) so that the light design and its selected sections correspond to the chosen analytical software methods. This leads to the design of a special industrial sectional light. Also, due to the need for precise arrangement, the light implementation cannot simply be assembled from elements easily available on the market.

The Sanezoo UNITY combination of light and camera captures and evaluates surfaces of any kind, regardless of reflection or ambient light. And it is safe, as it has no lasers or infrared while covering a wide scale of resolutions.

It can find small defects and discontinuities like scratches, cracks, pores, or other dents as well as various deformations, burrs, deposits, or inclusions. It classifies defects by type and severity with pseudo-defects and false alarms eliminated thanks to fine surface structure reconstruction where defects are emphasized and clearly distinguishable.

Another use-case where Sanezoo has a viable solution is GRASP, a universal part picking and feeding arm. This system can pick randomly scattered parts from a box or from a conveyor belt - firmly, and accurately. It knows how each part is grasped, by which picking point, and from which direction. Firm and precise grasping is important for speedy robotic manipulation and short cycle times. The fast cycle time - from the start of scanning to the path calculation - is less than 2 seconds. The need for programing and setting up the robot, 3D camera, and vision applications are all greatly minimized in Sanezoo solutions.

In the production of audio parts and products, these systems already have clear potential in multiple applications. Both inspection and perhaps assembly of speaker backplates into magnetic structures come to mind, with picking combined with fine surface inspection. The typical cold forged T-yoke pole pieces can be dinged during tumbling and the area that will form the magnetic gap is critical.

Sanezoo is currently looking to partner with audio testing equipment manufacturers and their system integrators to create the next-generation Acoustic Testing Robotic Cell (ATRC) for the robotic picking and handling of audio products, namely speakers and speaker parts, headphones, earbuds, and headsets.

For example, for headphone manufacturers the robotic arm could pick a product from a belt, put it on an artificial ear, and remove it when the test is done. For a cost-effective approach with higher throughput, the robot arm could place the DUTs into an acoustic chamber for testing in batches (two to four headsets at a time) with one acoustic tester switching from one headset at a time. Visual inspection is an optional module.

Sanezoo has an application-specific testing program for samples that can be sent to start the process. For more information, contact Lubos Brzobohaty.

Production line audio testing poses many challenges such as noisy environments, harsh operating conditions, high throughput, relative limits, and more. In this article, Steve Temme, the founder and President of Listen, writes about the essentials of test and measurement applied to manufacturing environments and quality control. Production tests always benefit from proven practices and experience and that's what is shared in this article, outlining the main considerations to ensure a successful operation. This article was originally published in audioXpress, March 2023. Read the Full Article Now Available Here

This article examines an important patent filed in 2014 by F. Bruce Thigpen (Tallahassee, FL) and granted in June 2016. Appearing both as the inventor and assignee, Thigpen describes a rotary sound transducer having an improved output at higher frequencies. Thigpen has been working on this type of device for more than a decade and has advanced what was already a very difficult engineering achievement, to a more sophisticated structure that apparently raises the general performance and high frequency limits beyond what was thought possible with this architecture. This article was originally published in Voice Coil, July 2017.  Read the Full Article Now Available Here