As the new energy vehicle and semiconductor industries continue to grow, a key link is drawing increasing attention: parts quality inspection. With inspection standards becoming ever stricter, the traditional model relying on human experience and sampling inspection is gradually becoming a hidden bottleneck for product quality and capacity ramp-up, while AI-powered ultrasonic quality inspection is emerging as a new solution.
On October 12, at the 2026 China International Industry Fair (CIIF), Hikvision won the 'Industrial Automation Award' for its AI ultrasound inspection technology, confirming the effectiveness of this intelligent approach.
According to reports, Hikvision has innovatively applied AI modeling, quantitative analysis, and classification and recognition to raw ultrasonic signals, much like 'installing' an expert who can read waveforms directly into the device, enabling it to accurately identify defects in parts. This turns what used to be an experience-dependent 'technical craft' into a standardized, automated intelligent process, truly lowering the barrier to using the technology and raising its industrial value.
Clear to See, Able to Read
Large Ultrasound Model Precisely Parses Raw Signals
Just as reading human ultrasound requires a professional doctor, ultrasonic inspection of industrial parts likewise requires certified professionals. This high barrier has long confined ultrasound technology mainly to specialized industries such as aerospace, making it difficult to popularize across broader industrial scenarios.
Addressing this industry need, Hikvision, drawing on its deep accumulated expertise in imaging and signal processing, has fully self-developed a series of ultrasonic non-destructive testing products, solving not only the 'clear imaging' problem but also, for the first time in the industry, applying AI modeling, quantitative analysis, and classification and recognition to raw ultrasonic signals, enabling devices to 'read' the quality information behind the waveforms.
But for AI to understand product quality, three difficulties must first be solved: the many 'clutter' signals can easily mask weak defects, leading to high missed-detection rates; when the material changes, the original algorithm fails, showing poor generalization; and defects on complex workpieces are difficult to identify and quantitatively analyze.
To this end, Hikvision has used its self-developed knowledge-enhanced signal reconstruction network and Large Ultrasound Model to solve the core challenges of ultrasonic quality inspection from four dimensions: first, adaptive noise separation filters out environmental interference so that weak defect signals are clearly distinguishable; second, the waveform-reading experience of senior inspectors is embedded into the algorithm, giving the AI the judgment of a 'veteran engineer'; third, through self-supervised pre-training and few-shot transfer, the model gains the ability to 'draw inferences' and generalize across materials and workpieces; fourth, automated quantitative analysis precisely computes a defect's position, size, type, area, and weld depth, with phased-array recognition precision reaching 0.1 mm, and for micro components such as MLCCs it can identify 5-micron-level defects—dozens of times finer than a human hair.
'Now, the AI ultrasound product alone can independently complete full-process interpretation, delivering not only non-destructive testing but also high efficiency and high precision,' said the head of Hikvision's ultrasound products, noting that AI ultrasound inspection can already achieve 100% fully automated full-process interpretation in typical scenarios.
'Inspect While Scanning' on the Production Line
Defect Recognition Accuracy ≥99.5%
Thanks to online waveform reading, ultrasonic testing can now be moved directly from the laboratory to the production line. Take the deployment in new energy vehicle parts inspection as an example: AI deep learning algorithms achieve end-to-end automatic determination of defect type, position, and size, improving single-workpiece inspection efficiency by more than 60% and complex-workpiece inspection efficiency by more than 1x, upgrading ultrasonic non-destructive testing from sampling to meeting the takt time of batch full inspection, and significantly reducing the costs of manual image interpretation and repeated testing.
In addition, with the flexible deployment model of 'AI analysis board card + non-standard equipment integration,' companies' overall investment is reduced by more than 50% compared with directly purchasing complete machines; by reducing destructive metallographic sampling, sample and material loss drops by 80%, further cutting rework, scrap, and quality disposal costs, creating sustainable cost-reduction and efficiency gains. In one actual implementation project, the defect recognition accuracy for new energy vehicle parts reached ≥99.5%, with a missed-detection rate ≤0.5%.
To date, Hikvision ultrasound has built a serialized product lineup including ultrasonic microscopes, three-axis/four-axis/six-axis automated equipment, intelligent all-in-one ultrasonic machines, portable devices, and multiple types of probes, with deployments in typical scenarios such as friction stir welding in automobiles, MLCCs, and wind turbine composite blades.
'At a time when intelligent manufacturing is penetrating ever faster, quality inspection is no longer the last link on the production line, but a decisive battle for industrial competitiveness,' said the head of Hikvision's ultrasound products. 'We believe that this deep synergy of "AI + manufacturing" will inject a continuous stream of intelligent momentum into the high-quality development of Chinese industry.'