Industrial X-ray inspection looks like a mature market. X-rays have been used for decades to find foreign objects in food, inspect solder joints on printed circuit board assemblies (PCBAs), and detect flaws in industrial parts. The X-ray sources, detectors, and mechanical systems are well established, and leading overseas suppliers have spent decades refining them.
Yet Wang Wei, president of Hikrayin, sees it differently. “Industrial X-rays, I think, are only just beginning and remain at an early stage,” Wang said. “Look backward and you see a crowded market; look ahead and you see open territory.”
At this year’s China International Industry Fair, Hikrayin introduced a high-precision 3D automated X-ray inspection (AXI) system for semiconductor packaging. The company says the equipment combines submicron measurement, 3D image reconstruction, and defect analysis with consistent, reliable performance, extending its reach into advanced packaging and laboratory failure analysis.
In just a few years, Hikrayin has moved from millimeter-scale security screening to micron-scale industrial inspection and now submicron measurement. Alongside improvements in precision, penetration, and defect detection, it is investing heavily in artificial intelligence.
Apple’s newly unveiled foldable iPhone Duo has brought renewed attention to phone hinges, which can contain hundreds of parts. A magnifying glass can reveal surface problems, but it cannot reliably show internal defects or foreign objects that affect manufacturing yields.
Hikrayin began combining X-rays and AI to inspect hinges automatically several years ago. Wang told 36Kr that, when the process was still new, manufacturers struggled with foreign objects, cracks, voids, and missing components escaping detection. Inspecting every unit with AI-assisted X-rays reduced the defect rate by an order of magnitude, he said.
In meat processing, where X-rays have primarily detected bone fragments and foreign objects, the company has added image algorithms and large AI models to measure fat content. AI is also making it possible to inspect assembled laptops, springs inside eyebrow pencils, and automotive components, applications where X-rays were previously uncommon on production lines.
Founded in 2020, Hikrayin entered industrial inspection in 2021. Its business now covers electronics manufacturing, food, and other industries. Backed by Hikvision’s R&D capabilities, it says it ranks among China’s top three suppliers in several equipment niches.
The business is part of Hikvision’s expansion beyond security equipment. Since 2016, the company has extended its sensing and AI technologies into semiconductors, automotive technology, and the computer, communications, and consumer electronics industries. It has branched into eight businesses, including Hikrobot, Hikmicro, HikAuto, and Hikrayin, seeking to reposition itself as an AIoT company that combines artificial intelligence with the internet of things.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: What is the main technical breakthrough in the new submicron, high-precision 3D AXI system for semiconductor packaging?
Wang Wei (WW): Its main achievement is submicron X-ray measurement with high reliability and consistency. Submicron means less than one micron, taking us into the nanometer scale. At the same time, it must perform 3D image reconstruction and defect analysis.
It differs substantially from the equipment we made before. Previously, X-rays were mainly used for inspection during manufacturing, with the primary goal of finding defects. Besides production inspection, this submicron product can serve as measurement equipment in research institutes and laboratories. Inspection and measurement are not quite the same. Production inspection mainly asks whether a defect exists. In a laboratory, you also need to know the length, width, and dimensions of a structure. That demands greater precision, reliability, and consistency.
36Kr: Which gaps does it fill in Hikrayin’s product range?
WW: There are two main areas: semiconductors, particularly advanced packaging, and laboratory failure analysis and measurement. Our products are extending from their previous focus on production inspection into higher-precision measurement. There are not many companies worldwide that can enter this field. You could probably count them on two hands. Very few domestic manufacturers can enter this field, either. The industry as a whole is still at a relatively early stage.
36Kr: How does the product compare with those of leading overseas suppliers?
WW: Broadly speaking, China has mostly followed overseas developments in high-end X-ray equipment. But with the arrival of high-precision equipment like this, performance has essentially reached the level of the world’s leading products. In some specific applications, we can lead. For complex defect detection, for example, we can deploy more powerful AI models. Where the signal-to-noise ratio is poor, image enhancement algorithms can reveal defects that were previously difficult to see.
36Kr: What has the technical progression been since Hikrayin entered industrial inspection in 2021?
WW: We started with security screening equipment, which basically operated at the millimeter scale. We then entered industrial inspection at the micron scale. Our submicron product takes us another step forward. It comes down to two things: physical sensing capabilities, with greater precision and penetration, and inspection capabilities, with higher detection rates and fewer false alarms.
The AI algorithms have evolved alongside that. From the time we started making inspection equipment, we have actually used very little traditional pattern recognition. Our first generation already used deep learning. In recent years, we have gradually moved from small models to large ones. The overall direction has not changed, but the pace is driven by users’ needs. We have not rigidly prescribed which industry must come first or second. It is more about bringing X-rays and AI together and applying them to customers’ actual problems. Where the value is greater and the problem more urgent, we may prioritize developing that application with the customer.
Technically, there is no end in sight to the pursuit of greater precision and penetration. The same goes for AI: from pattern recognition to small models and large models, and perhaps eventually self-improvement, making equipment easier to use with less human intervention. It is a little like a world record in sports. The record stands, but sooner or later someone pushes it a little further. We have to keep challenging those limits in equipment, too.
36Kr: What does AI actually change about X-ray inspection?
WW: Looking backward, X-rays are a very mature technology. There have not been particularly large breakthroughs in the underlying technology over the past decade, just incremental improvements. From that perspective, it looks like a traditional industrial equipment sector. Many companies have been in it for ten or 20 years, accumulating deep expertise in materials, mechanics, and engineering. But looking ahead, you find an industry that is still very immature. The key is the combination of AI and X-rays. Previously, X-rays could produce images of the inside of an object, but many of those images could not be interpreted automatically. That restricted the equipment to laboratories or sample inspections. Once AI can interpret those images automatically, there is an opportunity to move from inspecting samples to inspecting every unit.
That shift directly determines whether the technology can scale. Only a few industrial X-ray applications have developed into sizable industries: detecting foreign objects in food, finding voids in PCBA solder joints, and industrial flaw detection. Why did they scale? Because they achieved automated assessment relatively early. Without automated inspection, a factory might need only one or two machines in a laboratory for failure analysis and sample checks. Once every unit can be inspected, each production line may need a machine. The number of machines needed changes completely.
I have always thought of X-rays as a general-purpose capability. They penetrate and form images. In theory, many products need to be checked for internal defects. The absence of X-rays from many applications in the past did not mean there was no demand. More often, automated inspection capabilities and the return on investment had not reached the necessary threshold.
36Kr: Is that similar to industrial cameras ten years ago?
WW: Yes. I often look back at the development of industrial cameras from 2015 to 2020. Before 2015, it was difficult to imagine so many of them on production lines. But as costs fell and AI and automated inspection improved, industrial cameras became a ubiquitous sensing tool in manufacturing.
X-rays may be going through something similar. Industrial cameras mainly look at surfaces; X-rays look inside. Many products have internal defects, but equipment used to be too expensive, or automated inspection was not possible. Now costs are falling and capabilities are improving. Both bottlenecks are being addressed at the same time.
36Kr: Can you give a representative example?
WW: Foldable phone hinges are a very good example. A hinge may contain hundreds of components. How were they inspected before? Often, someone looked through a magnifying glass. But that only shows the surface, and when people spend a long time examining such complex parts, they inevitably miss defects.
We began automated X-ray inspection of hinges several years ago. At the time, hinges involved a new manufacturing process, and foreign objects, cracks, voids, and missing components were escaping detection. Inspecting every unit with X-rays and AI reduced the defect rate by an order of magnitude. Combining X-rays with AI made inspecting every unit, previously very difficult, part of the production process.
There is also the inspection of fully assembled laptops. That is more complex than a hinge, but fundamentally the same: are screws missing, are there extra screws, or are there foreign objects? Those are all things X-rays are well suited to checking.
36Kr: Does the same reasoning apply to food inspection?
WW: Yes. Chicken breasts and thigh meat, for example, must not contain bone fragments. After we entered that application several years ago, customer trials showed that we could detect bones as thin as about one millimeter. Customers were quite surprised because overseas companies had largely dominated that field.
Building on that, we used the characteristics of the X-ray energy spectrum together with AI to measure fat content. Different proportions of fat and lean meat in pork, for example, affect both taste and price. Previously, perhaps only a handful of companies worldwide could do this, and the equipment was very expensive. Now that we have developed this capability, there is an opportunity for it to become more routine.
I often say that when new and traditional technologies meet, they produce sparks. Our job is to catch those sparks.
36Kr: Everyone is talking about large AI models. What is their most important practical value in industrial X-rays?
WW: One very important aspect of industrial AI is practical effectiveness. There are many open-source models today, so at first glance AI seems very mature. But back on the factory floor, relatively few companies do it well. Why? Industrial applications often do not give you much time. A new product may go from the sample stage to mass production in just one or two months. Can you develop an entirely new inspection category within that period and meet the customer’s requirements? That is a very high bar. Given a year or two, many companies could probably do it. But an industrial customer will not wait a year.
36Kr: Is few-shot learning important here?
WW: Very important. Traditionally, developing a hinge inspection model might require thousands or even tens of thousands of images, followed by labeling and training. Using the foundation provided by Hikvision’s Guanlan AI models, we may achieve good results for an entirely new category with one or two hundred, or a few hundred, images. If we have already worked on that category, such as one hinge design before moving to another, about 10 images may be enough to transfer the model relatively quickly.
If you compare a model to a child, others may have to teach it from scratch. Our child has already reached elementary or even middle school, so teaching specialized knowledge is naturally faster.
36Kr: What practical value does that offer in high-mix, low-volume manufacturing?
WW: First, time. Going from 10,000 images to 200 could make data preparation alone dozens of times faster.
Second, accuracy and false alarms. Industrial inspection ultimately comes down to results. Many products now require three nines, or 99.9%, or four nines, and customers will keep pushing toward five nines.
Third, generalization. Previously, you trained one model for each product. With a sufficiently strong foundation, once I have worked on one category, I need progressively less data for similar categories. What users experience is equipment that becomes easier and easier to use.
36Kr: Where else is AI being used besides defect recognition?
WW: In many places already. Equipment used to have numerous image parameters, such as gamma and grayscale settings, that required a specialist to adjust. With AI, we can adjust them with one click. The machine automatically selects relatively suitable parameters based on the image, reducing the level of expertise needed to operate it.
The same applies to image enhancement. An image is fundamentally signal and noise. Sometimes a defect is hard to see because noise overwhelms the signal. AI can learn the characteristics of that noise and remove it so the actual signal emerges. These capabilities have been deployed for three or four years.
Looking ahead, I think self-improvement deserves attention. Can equipment become easier to use and more intelligent with less human intervention? That may be an area of further change over the next few years.
36Kr: Electronics, food, and automotive parts are very different. How do the Guanlan models adapt across industries?
WW: We have a simple saying: “If it is visible, it is inspectable.” As long as a defect has features visible in an X-ray image, AI has a chance of addressing it.
On the surface, an electronic component, a food item, and a car wheel look completely different. But to the model, they are all X-ray images. Different defect features may be hidden within otherwise normal images, and we use the underlying large model to identify them and iterate quickly. The industries may look different, but fundamentally it still comes back to “AI plus X-ray.”
36Kr: The industry uses AXI to mean automated X-ray inspection. Hikrayin now talks about AI X-ray inspection. What does that change signify?
WW: “Automated” is certainly not wrong, but we want to put greater emphasis on AI. Much traditional equipment still relies on pattern recognition and is suited only to relatively simple applications. Hikrayin incorporated deep learning from its very first machine. We started with small models and are gradually moving toward large ones.
The first change is a substantial improvement in inspection against complex backgrounds. The second, more important change is that AI is beginning to acquire some ability to understand. Equipment used to tell you only, “There is a defect here.” In the future, it will be able to tell you what kind of defect it is. Beyond that, it could analyze earlier manufacturing processes and determine which parameter may have gone wrong. Inspection would then create a feedback loop: finding defects downstream and using that information to optimize upstream processes.
The more deeply AI is integrated, the more valuable the equipment becomes.
36Kr: Will competition increasingly center on AI?
WW: Hardware and AI are both important. The physical capabilities of X-rays still need to advance, with greater precision and penetration. That will always matter. But if we look at opportunities at this stage, the integration of AI and X-rays is changing faster, and more applications remain undeveloped. The opportunity may therefore be somewhat greater.
I have always believed that the future belongs to intelligent equipment. It will not simply be traditional machinery. You might even think of it as an “equipment robot,” with a growing ability to make judgments, execute tasks, and adjust itself automatically. Our market performance over the past five or six years has increasingly supported that view.
36Kr: Many people still associate Hikvision primarily with security. What role does Hikrayin play among its eight businesses?
WW: Hikvision is already about much more than security. We started thinking about its future direction more than a decade ago. We gradually extended the sensing capabilities developed for security into smart homes, automotive electronics, and industrial production.
Sensing methods themselves are also expanding. Previously, we mainly used visible light. Now we have extended into infrared thermal imaging, millimeter waves, X-rays, and sound waves. Hikrayin is taking Hikvision into X-rays and detection beyond visible light. Put simply, it is about seeing through the surface to what is inside.
36Kr: Within Hikvision’s broader industrial business, is Hikrayin a smaller example of its deepening industrial involvement?
WW: You could see it that way. Hikvision now provides products and solutions for industrial quality inspection, equipment monitoring, production operations, and workplace safety.
In industrial inspection, for example, besides X-ray products, there are ultrasonic nondestructive testing systems, extended-depth-of-field 3D digital microscopes, mass spectrometers, and chromatography instruments. These products take different technical approaches but face very similar industry trends: all are being combined with AI to find high-value applications.
We are gradually building more comprehensive sensing capabilities for industrial inspection. For a chip, for example, ultrasound is better suited to detecting cracks along planar layers, while an extended-depth-of-field microscope can reveal microscopic surface defects. If the issue ultimately turns out to involve the material, chromatography and mass spectrometry may be needed for qualitative and quantitative analysis.
Every technology has strengths and weaknesses. Hikvision developed a multidimensional sensing approach back in its security business. The toolbox contains different sensing capabilities that we combine according to the user’s application.
36Kr: How does Hikvision’s own manufacturing system help Hikrayin’s products?
WW: It helps a great deal. Many of our products are first deployed at scale in Hikvision’s factories. Some applications might not occur to outsiders. For example, we use X-rays to check whether accessories are missing from camera packaging. Fundamentally, X-rays penetrate objects. However good a product looks on the drawing board, it has to be tested in the field. Only in a real manufacturing environment do you find out where it works well and where it does not. We are fortunate to be able to validate products thoroughly in Hikvision’s own manufacturing operations before delivering mature solutions to external customers.
36Kr: As a relatively young industrial inspection brand, did you initially encounter customers who did not understand the offering?
WW: Certainly. Interestingly, when we first actively promoted it, the results were not particularly good. Later, customers saw our capabilities at exhibitions or through other channels and approached us themselves. Once we addressed a real problem, those relationships became very durable. Why? Because we solved problems they had not been able to solve before. That is why our business has never been organized entirely by industry.
We do not start by determining whether to focus on this or that industry next year. Rather, it comes down to where combining AI and X-rays can create value. As those applications accumulate, the industry may gradually develop an understanding of Hikrayin’s capabilities. Ultimately, it is about making good products, solving problems, and creating value for customers.
36Kr: Are there particularly unusual industrial applications?
WW: Many. Some eyebrow pencils contain springs that can jam if installed incorrectly. Previously, it was difficult to see inside. We use X-rays to determine whether the spring is properly fitted. Automotive temperature sensors and motors are other examples where X-rays were not traditionally used for production inspection. There are more and more cases like these. Customers’ quality control requirements are rising while equipment costs are falling. As costs decline, industries that would never previously have considered X-rays can begin to make the economics work.
36Kr: What stage of development has industrial X-ray inspection reached?
WW: X-ray security screening has moved beyond its initial stage into the early-to-middle phase. Industrial X-rays, I think, are only just beginning and remain at an early stage.
Look backward and you see a crowded market; look ahead and you see open territory. Looking back, many X-ray companies have been operating for ten or 20 years, and traditional applications are mature. Looking ahead, AI is opening new applications while costs decline, so a great deal has yet to be developed.
36Kr: Will these new inspection capabilities eventually become standard manufacturing processes?
WW: I think so. Today, one customer may be the first to adopt an inspection technology and achieve much better quality than its peers. Once end customers see that, they make the same demands of other suppliers. Today’s highest requirement may well become tomorrow’s minimum standard. Many technologies initially look customized, but after three or five years they may become a small standardized product within an industry, or even a standard process. It is dynamic. Our technology is advancing, industry requirements are rising, and standards keep changing. That creates powerful momentum of its own.
36Kr: Where might Chinese companies find opportunities against established overseas X-ray suppliers?
WW: If we look only at the very best physical X-ray capabilities, we are still catching up overall. But at the equipment level, AI will certainly be an important source of competitiveness. One clear impression from my own experience is that many excellent overseas X-ray companies have not done much to integrate AI into specific applications.
One reason is capability. Hikvision operates at a scale of roughly RMB 100 billion (USD 14.9 billion), with many sensing businesses supporting the Guanlan models. Not every X-ray company can have that kind of foundation.
Another is the business model. Industrial AI is often fundamentally about customization and service. An application may ultimately sell only a few machines, or a dozen or so. We are willing to do that; many overseas companies may not be.
China has a complete manufacturing supply chain and a large number of real applications. Technology and the market are close together, so new capabilities can quickly find their way into individual industries.
36Kr: How are Hikrayin’s different industrial businesses progressing?
WW: Several areas are doing well. Electronics manufacturing, food, and industrial inspection all have a vast range of applications that X-rays have not yet covered. After several years of development, we can essentially rank among the top three domestically in several equipment segments. It is difficult to say which will offer the greatest future opportunity because, as things stand, every direction has considerable room to grow.
36Kr: Will overseas markets become a priority?
WW: Yes. China’s high-end equipment capabilities will keep advancing as its overall manufacturing system develops. Hikrayin is not progressing alone; it is supported by improvements across China’s X-ray supply chain and manufacturing capabilities. I believe Chinese equipment will increasingly reach global markets, just as footwear and apparel, white goods, and phones have done, and as new energy vehicles are doing now. Overseas markets are certainly important to us.
36Kr: Do you have a clear global goal?
WW: Internally, we have a fairly clear short-to-medium-term vision: to offer the world’s best AI X-ray inspection solution. But it is difficult to give “best” a highly quantitative definition. For me, it ultimately means that users find it works well for them. We believe that, with Hikvision’s platform, our own efforts, and the improving capabilities of Chinese manufacturing as a whole, we have an opportunity to do this.
36Kr: In ten or 20 years, what would you like people to think of first when they hear “X-ray inspection”?
WW: Hikrayin. More specifically, I hope they think that Hikrayin led this wave of intelligent X-ray inspection. We were founded in 2020 and have been at it for several years. Looking back, around 2020 was also a turning point in intelligent security screening. We were pushing the integration of X-rays and AI, while the broader trends of the time were also helping us succeed.
As people in Zhejiang say, ride the wave. At this moment, you have to position yourself at the crest. If it is the right wave, the future is yours. At least you have a chance.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Xi for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.69 = USD 1 based on estimates as of October 22, 2026, unless otherwise stated. USD conversions are approximate and, where appropriate, rounded for ease of reference. They may not fully match prevailing exchange rates.