Miwen Power CTO Su Jun: Edge Ai Enable Factory Wisdom Upgrade

Hello everyone, I am very happy to participate in the "Edge Calculation Frontal Practice" series of rice, this series of public classes focuses on the application cases of rice dynamics in edge computing practice, and prospects to the edge computing technology route.

This issue will focus on the application of edge calculation in smart plants. My name is Su Jun, is the cto of rice.

01, what is the utilization of rice?

As a video-assisted AIOT edge solution provider, rice power is committed to solving all problems encountered during the batch landing process for edge computing scene users.

HardwareMilin power provides a variety of low-power, different powerful hardware computational products for different scene requirements.

Software aspectTo provide hardware users with a variety of free value-added services such as remote system upgrades, bulk burning tools.

At the same time,Part of the sceneMiwen’s power also provides a complete solution of clouds, clearing obstacles to customers, solving the final artificial intelligence scenes.

Therefore, if you want to solve a edge problem, your questions will be answered in rice.

Today, I will combine the actual application scenarios, and everyone to explore the application of the edge AI in the intelligence of the factory.

02, the development trend of smart plants

The concept of smart factory, the earliest industry 4.0 proposed by Germany, generally equipped with the following characteristics:

The first is the digitalization of the factory, can automatically monitor the device, the management of information resources, and information services to the user;

Secondly, the equipment status is monitored by Internet of Things technology, which combines Internet network technology, information management and information service technology;

Based on these technologies, information-based energy saving and consumption consumption is realized, and the purpose of achieving green intelligence.

03, pain points applied in smart plants

For the goal of factory wisdom, it will always fall to the essence of factory production, such as:

How to shorten the order production time?
How to improve product quality, reduce manufacturing costs?
How to reduce inventory?
How to control the human cost of production and management?
How to monitor production site?
How to make production management transparency, visualization, can be quantified?
How to provide complete, accurate manufacturing data?
How to effectively control product quality?

04, the edge application inside the wisdom factory

Of course, there are many solutions to these problems, and in the edge computing application of rice dynamics, it will mainly involve the following aspects:

Industrial Quality Inspection: Find out the defects caused by the product in the production process, classify the product itself, main purpose is to strengthen the control of product quality, improve quality inspection efficiency, transparent, visualization, and quantification of quality inspection Management.

Safety production: refers to the production activities, in order to avoid the accident prevention and control measures to avoid personal injury and property loss accidents, ensure the personal safety and health, equipment and facilities of employees, and the environment is protected from damage. .

Wisdom operations: including safety production, but broader, the purpose is to make the information related information transparent and visualize.

existIndustrial quality inspectionAs shown in the figure, it is a feather grading application in partners. In this system, there will be 21 different levels of different grades of feathers.Its accuracy exceeds 98%, and the operational efficiency is 100 times that of manual detection.

This type of application scenario has highly requested the efficiency, real-time and response speed, and is usually high temperature, high humidity in industrial quality inspection. It is, with the company’s industrial specifications and powerful computing power of rice power EVO Xavier, it can achieve such high efficiency.

In terms of safety productionUsing the factory’s existing monitoring camera, whether the employee wears a hard hat, whether it is detected in violations such as smoking in the factory, timely warning fireworks, gas leaks and other safety accidents.

Smart operations often contain safe production, but the range is wider.

In the wisdom operationIt is necessary to confirm that the personnel entering the warehouse, performs perimeter inspection on the important area of ??the factory, and inspects whether the workers work is qualified, and the employees are departed, and the monitoring of sleeping posts. These are all could not data or quantifically, and now you can provide more dimensions for operational decisions by dataization.

05,Edge VS cloud

Maybe someone will ask, these things seem to be made through the cloud, what is the benefit of using the edge of the edge?

A comparison is carried out for the dressing (such as safety cap detection, hand suit compliance, etc.) to simultaneously access 15 channels.

Divide costs into hardware costs and communication costs

In terms of hardware costs, it is also 30T computing power, and it takes 28,000 / year’s high cost in the cloud. At the edge of the edge, only 12 million / year, while enjoying the three-year warranty,Cost difference reached 7 times.

In terms of communication costs, it is assumed that the data path is free. One minute needs to upload a 1M data to the cloud.The communication cost that needs to spend approximately 100 yuan.. And the cloud program, due to the cost of video traffic, it takes 6,000 yuan.


06, embedded edge VS desktop GPU

Of course, the cost of clouds and ends has long been self-evident.

However, in many applications, users use the GPU of the desktop as the edge computing equipment, the embedded edge devices provided by the desktop GPU and rice power supply?

Taking industrial vision as an example, comparison from both power costs and downtime probability.

Petitude is electrified: 0.725 yuan per degree, 12 hours of full load. Due to power differences, in terms of electricity cost, use Jetson Xavier embedded and using desktop GPUs, the gap is large, the year is 600 yuan, but for an expensive testing equipment, 600 yuan is completely what.

Industrial use scenarios: We found that workers’ switchgear operations will not be carried out in accordance with the manual, press the power button to use the most habitual use of them. This habit has caused the maintenance probability to make both gaps more than 20 times.

For equipment manufacturers, this is an additional maintenance cost; for factories, it will face the loss caused by stopping.. This is where we think is the largest difference in embedded edge devices and desktop GPUs.

We will look at the difference between desktops and embedded GPUs in terms of integrity:

In this figure, we can see that the left half of the Y-axis is an embedded edge device, that is, the Jetson series, the right half of the Y-axis is usually the graphics card.

It can be seen that so far, even the strongest embedded edge device – Jetson AGX Xavier, its strength is just a brother relative to the graphics card on the right, is only equivalent to the previous 1050Ti graphics card.

The poor of the force is one of the reasons why the embedded edge equipment is currently used in industrial scenarios. However, everyone can see,ORIN’s emergence has brought very huge changes to the entire pattern..

Although its computational power is still not as follows, the power consumption of 3080Ti is 350 watts, while ORIN is just an embedded device.

In this case, the force problem of embedded edge devices is no longer a big problem, butHigh reliability, strong energy savingThe advantages will continue to keep it.

Therefore, the application of embedded edge devices in the future industry will be more extensive.

07, the exploration of rice power in the smart factory scene

In order to make the edge calculation in industrial scenes, it is more likely to land.End users who have business needs, rice power and partners jointly launched a series of tools and products that are closer to application practices..

In terms of industrial quality inspection, there are very many small parts in industrial requirement, and Miwen Power Joint Partners Commentary provides a one-stop solution for industrial quality inspection. The standardization machine is launched.

For parts such as seals such as seals, screws, lamp holders, buttons, this standardized machine can be used to effectively detect, and only a small transformation is required, and it can be docked with the current production lines, and a set of " Fool "training tools to help users perform rapid tuning.

Similarly, in the factory scene, due to the very sensitive production process data, a large number of customers want themselves to hold this data to ensure data security.

Based on this consideration, we provide the following models with our partners:

Firstly, the common model is preset, enabling users to perform fast verification.

At the same time, support users to perform scenario selection and model design, allowing users to perform quick validation without giving data.

In the industrial sector, how to use a small amount of data to achieve the best effect is still a technical challenge, and the user does not exclude the user in the case of using our model, it is still unsatisfactory. Based on this requirement, rice dynamics also provides custom algorithm services to provide users with algorithms.

In a variety of modes, the user data security is not only protected, but also meet the performance requirements of different levels of users.

These tools are for users who ultimately have business needs.

Provide a series of easy-to-use software tools for developer user needs

Demand 1: Remote operation of the dispersed edge device

The application scenario of the edge calculation is usually very dispersed. For development types, customers who serve their customers are often far away from the city, relatively dispersed. When remote maintenance is required, they have to pay more high travel and human costs, so they want to be able to pass remotely as much as possible. True question.

These pain points and demand for users,Miwen’s power independent research and development cloud management platformThis platform is managed by enterprises. The information of different enterprises is isolated from each other and can simultaneously supervise different projects and various devices in each project.When the user needs remote maintenance, open the remote access function (green icon), click to jump to the device interface, you can view the device usage statusYou can also use the web terminal to make a more in-depth command line debug.

It is worth noting that in addition toThe user has permission settings, view, modify the username and password of the device, anyone includes merit power, there is no need to set up, check the accountIt greatly guarantees the security of equipment data.

In addition to remote operation and maintenance,The platform can also remotely monitor the CPU usage, hard disk usage, temperature, etc.. The hard drive is often the main reason for the problem of users. If the hard disk is full, the system cannot start, the remote monitoring function can effectively view the abnormal state, timely alarm, and make the corresponding countermeasure. As shown above, both the hard disk and the CPU have a sharp concave, indicating that the device is restarted at this point in time.

at the same time,Big screen display function is also availableIt is convenient for users to control the equipment in the project.

Demand 2: Application and model deployment, update

Milin motivationProvide graphical batch burn toolsIt can simultaneously support multiple devices to burn and mirror clones, and users can issue themselves from the volume of the containers and models.

Demand 3: Need a large number of engineers to customize development

Take the remote monitoring production site as an example, develop this function requires embedded engineers, front-end engineers, backend engineers, etc. Users who have limited resources have more hope that they can focus on the part that is really good, this part may be business functions, or the algorithm performance.

The video is an example of the production management system used in our company, showing rice powerBox, ready-to-use platform functionHelp users better integrate artificial intelligence with their business systems, speed up the speed of the project.

This production management system can remotely detect intrusion, fireworks, temperature and humidity sensor data, etc., system background reception, record, analysis of abnormal data, and further manage the library. In addition to providing a similar set of outlookable features in addition to different users’ needs, Milin Dynamics can also be passed directly to users, or allow users to upload their own algorithms themselves to quickly verify their algorithms.

08, the future calculation


Let’s discuss the future of the edge calculation.

As we all know,Next year NVIDIA will release the Embedded GPU of Orin.

For Xavier, in the CPU, ORIN’s CPU core is 50% more Xavier, and the single core performance will have further improvement. The CUDA core of GPU is 4 times that of Xavier, and the force reaches 254T, close to 3080ti and 3090. The level, although the memory bandwidth is not as good as Apple’s latest IMAX (400GB / s), but also reached an amazing 200Gb / s, it is possible to play the GPU computational performance very well.

Under the bonding of this chip, more industrial scenes will solve faster, and that is the top peak of embedded GPU?

No, let’s take another picture.


Atlan is the chip that NVIDIA announces that the chip is being developed. The power will reach 1000T, and 1000T is 4 times the most powerful 3090 desktop video card. Therefore, from ORIN, the performance of the edge calculation will be improved. At the same time, in various applications of the edge, it is no longer using SOC designed specifically for the data center, but has a strong powerful new chip specifically suitable for edge scenarios.

This change will push the edge application to a higher dimension, in the scene calculated,Use chips and programs designed specifically for edges to the future of the edge.

My sharing content is over, thank you for your listening.

Sitting in more than 3,500 scientists engineers, three years have lost 24 billion, Shang Dynasty is not going to market?

Professor and doctoral gather, plus thousands of scientists and engineers, if such a dream lineup is used to build a company, will it be a pair of scenes?

As you say, you may not believe that such companies really exist.

Not long ago, Ai, a business Tang Technology, passed the HKCO listed hearing.

The unicorn in this AI is gathers more than 40 professors, more than 250 doctors. In addition, more than 5,000 employees, two-thirds are scientists and engineers.

What is even more cow has been founded in 2014, and Shang Tang has issued more than 600 papers in various journals, and there are more than 8,000 patents.

Describe with a sentence, that is, the scientific and technological content burst.

More unexpectedly, there is so many cow people, Shang Tang three years is still over 24 billion. . .

And today, the bad reviews are also preparing to smash the small hotspots listed on the market, and spend the story of the company, this company has the story of "AI Si Xiaolong’s head".

First of all, I said that I had to mention Tang Xiao ‘s big cow.

He graduated from the 1990s, he chose to go to the United States, and pursued a doctorate in Mirace, and he was in this period. He touched his face recognition algorithm.

After getting a doctoral degree, the big 佬 continued in research on computer visual algorithms at the Hong Kong University of Chinese University and Microsoft Asia.

He has also formed a multimedia laboratory of Hong Kong Chinese University, and Shang Dynasty Technology is also initially created here.

At that time, the field of face recognition has always been a bottleneck, which is how to improve the recognition accuracy of the algorithm.

Simply put, it is to improve the ability to judge that two photos are not the same. The accuracy of 97.53% of the human eye is considered to be a ceiling of the algorithm.

However, Tang Xiaou did not agree. In 2014, the Tang Xiao Team used the original GaussianFace algorithm to transcend his eyes, and the "monster" of 98.52% of the accuracy was created.

Subsequently, the high-yield Tang Xiao Team immediately released a more variatory DeepID algorithm, directly increased the experimental score to more than 99%.

The data of the anti-day is out, IDG’s investors also smell the wind and plan to invest tens of millions of dollars, helping the business of people’s face recognition technology, and Shang Tang Technology has also been formally established.

As for the company, I was named "Shangshuo", and their official explanation is that Shang Dynasty is currently the era of Chinese civilization, and the emergence of text has brought about human civilization. They think that technology is also the same. The founding emperor of Shang Dynasty was just calling soup.

Just, everyone is more likely to explain, that is, Tang Xiao will turn it from the professor to the business, and the professional soup, refer to business soup.

Shang Tang Technology has always been the fragrance of the capital market.

From 2015, the company has experienced 12 rounds of financing, with a total of $ 5.2 billion, and the business of Shang Dynasties is also greater.

In the computer visual field of its home, it is said that there is more than 400 million mobile phones in more than 400 million mobile phones in more than 400 million mobile phones.

The Chengdu Shuangliu Airport, Guangzhou Baiyun Airport, Zhengzhou’s subway station also uses brush face ban on Shang Sang Technology.

In other aspects, Shang Tang also established a universal artificial intelligence infrastructure based on his own old, and helping a AI for each field.

Take the smart city as an example, the merchant soup can use visual analysis, detect the displacement of the well cover, the fence guardrail, grab the image of illegal parking.

It is also possible to estimate traffic congestion, predict the occurrence of fires.

In this way, the technology of Shang Tang is used in life, and the corners of the city.

According to the last year’s data, Shang Dynasty also became China’s largest computer vision software company, Asia’s largest artificial intelligence software company.

And in comparison with clouds, according to the map, 视, etc., there are nearly 10 billion months of Shangjun, which has become the first of AI Si Xiaolong.

When I saw this, I would say that there is a friend who will say, since Shang Dynasty technology is so top, after the listing, I just can only rush to Rapuha, directly selling the business soup stock!

The bad review may be said to stop, be impulsive, because Shang Dynasty technology seems to be full of fruits, it is actually the money.

Shang Tang Technology can also be said to be a perfect representative of the AI ??industry, behind the light and bright, actually a huge loss. . .

Just say this, although the revenue of Shangshui Technology in the past three years is 1.85 billion, 30.3 billion and 3450 million. But after the various costs of research and development, in fact, Shangshui is a loss every year.

Among them, the light is more than 24 billion due to the three years.

Why do you burn this?

The bad review is simple to give an example, and the Shang Tang Guang is in order to build their own Ai superior centers, and the merchant soup has invested 5 billion.

But the superior center is in this place, but it can be eternal once after it is finished, because it is a truly anti-Gold beast.

The researchers of Shang Tang have been slightly estimated. Press the super-budget center that labeled Run (running) button, the iteration of the data training is at least 500,000 yuan.

So, selling hundreds of hot pots, it can be not enough.

So many scientists and engineers inside Shang Tang, every day, relying on "run" to train their own AI models.

And, this hasn’t counted the salary of scientists, and there are hundreds of millions of costs that have added the purchase of GPU clusters every year. Yes, AI is such a game of burning money.

This is not an example of Shang Dynasties. Other AI San Xiaolong’s investment in R & D is not small. It does not move to billions or even billion R & D.

However, it doesn’t necessarily have a result, and there is no need to commercialize. If you don’t invest money, it is afraid of being eliminated. The AI ??industry has fallen into such a strange circle.

Plus the construction cycle of the smart city project such as subway, airport, and the return period is also long; and many mobile phone manufacturers have implemented the AI ??algorithm, you don’t have money to make new stings, they will not pay again. you.

Loss, that is, it has become a homewinner in the AI ??industry.

These two years have attached importance to personal information, especially human face data, but also make simple face recognition data more difficult, so, "Shang Tang", who rely on this family, really encountered some bottlenecks. If this time is, the establishment is also a wave of direction adjustment.

The listed plan of this business soup is ready to raise 60% of the IPO, and then invested in Ai R & D.

So, as a research institution, these explosions and crazy money "Shang Tang" are really excellent.

But as a company, profitability and commercialization is their purpose. In this regard, "Shangjou" is still not a matter of time.

In addition, the story of the AI ??industry also reminds me of an innovative field a theory called "Valley of Death".

Simply put, technology innovation has been proposed, and there is still a long way to go between the technology innovation. Most of these projects will die on this road, and this divide from scientific research to industrialization is also called "Valley of Death".

In the past few years, the AI ??frenzies, after the PPT, the era of increasing, today’s AI industry seems to be in this deep valley.

Although AI has been applied in many fields, 97% to 99% accuracy of the leap, may take hundreds of millions of real gold and silver investment, but to tell the truth to the people of bad reviews, I feel true. not very big.

And from now on the performance of the AI ??industry, losses are likely to become a homemade.

Finally, the profitability of these AI industries may have doubts, but for the future of AI, the bad review is still very optimistic.

Just like a lot of money, in the 1950s, the "modern communication theory", UNIX operating system, mobile phone technology Bell laboratory, did not expect that these technologies can be decades later Root root germination, even change the world.

The AI ??industry may also be the same, I believe that all the way to burn money all the way to the current "Shangjun", will also plant a seeds, and they are also likely to change the fruits of the world.

A Novelty story: action, climax & the end

In 1958, the Bharatiya Jana Sangh, the precursor to the BJP, lost the Delhi civic polls it contested. The evening of the results, its top leaders, L K Advani and Atal Bihari Vajpayee, friends and confidants, watched Raj Kapoor’s Phir Subah Hogi at Paharganj’s Imperial cinema.

Advani had narrated this at an event in 2011. “We lost. To drown our sorrows, we went to a movie together at Paharganj’s Imperial cinema. Incidentally, the Raj Kapoor movie was titled Phir Subah Hogi… and we finally came to power at the Centre.”

Such was the enduring charm of cinema. But that was another era, when cinema held centre stage, before plush seats and popcorn in tubs changed the movie-watching experience. Those were the days of wooden seats and 35mm projectors, of groundnuts in paper cones and hooting crowds.

The demolition last year of another single-screen hall, Old Delhi’s Novelty cinema, a few kilometres from Imperial and one of the oldest and earliest cinemas of the Capital, further brought the curtains down on that era. Recently, the North Delhi Municipal Corporation, which owns the property, leased out the 1,157-square-metre plot to a private player for Rs 35 crore for 99 years.

The Corporation, which owns the property, is said to have been struggling to redevelop it since 2000, when it took over the premises from Vijay Narain Seth.

Seth, now 75, says his father, Jagat Narain Seth, bought Novelty (then called Elphinstone Picture Palace) from the East India Trading Company in 1933. The first movie to be shown here was Night of Love in 1935. “Initially, only English movies used to be screened. Later, my father started screening Hindi films, including Ratan (which ran for 50 weeks in the 1940s), and Kadambari,” he says. There were to be many more — Dilip Kumar’s Andaz, Jugnu and Mughal-e-Azam. Ramesh Sippy’s Sholay ran for three years.

Seth also recalls how Novelty screened Gemini Studios’ Chandralekha, “the first ever south Indian film to be screened in the north”.

“We would show only three-four films a year, while other theatres would change the shows more frequently,” he adds. The last film that had a jubilee run, Seth says, was Sridevi’s Nigahen, in the late 1980s. In 2000, the family had to down the shutters on their six-decade enterprise.

The family-owned two more movie halls in Old Delhi — Jagat near Jama Masjid and Ritz at Kashmere Gate. “All the cinemas at that time were located inside the Walled City, but a little away from the residential areas,” he says.

Initially, when the civic body took over, it wanted to revamp it as a cinema-cum-commercial complex. In 2012, there was even a proposal to convert the Novelty cinema building into a spice market.

However, both the plans failed to take off.

What is left of Novelty is an open ground with rubble and strewn with plastic bags and water bottles and beer cans.

Old-timers from the Fatehpuri area, where Novelty once stood, recall its two waiting halls, where the crowds waited for the next show, “gorging on cream rolls, pastries and cold drinks” in the meantime.

This is the third cinema in the area to be razed. In the 1990s, Jubliee near Chandni Chowk was demolished, and a few years later, New Amar was razed to make space for the Chawri Bazaar Metro Station. In the heydays of single screens, Old Delhi alone had 12. Of those, only four remain operational — Delite, Abhishek, Moti and Ritz.