“Edge AI” is one of the major purposes of edge computing. This edge AI is one of the very important key technologies in the realization of IoT, and its spread is expanding. Here, we will introduce what edge AI is, what is different from cloud AI, why edge AI is attracting attention, and some concrete examples of edge AI.
What is Edge AI
Edge AI is to put artificial intelligence on an edge server. It can be said that it is a technology to balance the advantages of speed and the advantages of artificial intelligence, which are the advantages of edge computing.
The advantage of artificial intelligence is that it can automatically perform tasks that cannot be standardized once it is learned. In other words, it enables processing that cannot be done with a normal computer (Von Neumann computer). This kind of work that cannot be standardized can be said to be a characteristic of the work that humans have done so far. In the old days, there are several types of artificial intelligence, such as expert systems, but in recent years most of them have been deep learning of about 14 to 20 layers using neural networks (deep learning).
It is believed that by making a typical work artificial intelligence, it is possible to make up for the shortage of skilled engineers. From a different point of view, it can be said that the judgment and experience of skilled engineers are transferred to artificial intelligence. Specific examples include “predictive maintenance” that predicts equipment failures and the timing of parts replacement in advance, and “image processing” that inspects products.
Skilled technicians have conducted “learning” over a long period of experience, and have judged whether the product is good or bad, when to replace parts, etc. based on the intuition cultivated through experience. It can be said that one of the ultimate goals of smart factories and IoT is to save labor without degrading the quality of inspection and maintenance by transferring those experiences and judgments to artificial intelligence.
Differences between Edge AI and Cloud AI
On the other hand, cloud AI is to put artificial intelligence on a cloud server. The disadvantage of cloud AI is that the processing speed is slow because communication is always performed between the edge server and the cloud server via the network.
On the other hand, if you put AI on the edge server, the processing will be completed only in the edge server, and you only have to send the analysis result and prediction result to the cloud server. For this reason, it has the advantage of faster processing speed than cloud AI, which requires communication for each process.
On the other hand, cloud AI, which can handle a huge amount of information, will be easier to handle when learning work is easier and new events occur. In other words, cloud AI is more flexible. Also, it should be noted that the resources of the edge server are limited, and the size of the artificial intelligence of edge AI (the scale of the neural network) is also limited.
From the above, it can be said that the best system configuration is basically to prepare both edge AI and cloud AI, perform learning with cloud AI, and send the learning results to edge AI. Specifically, cloud AI first learns using a huge amount of information, and when the learning progresses to some extent, the learning results are sent to edge AI. And the edge AI that got the learning result performs the actual processing. Then, the analysis result judged by Edge AI to be defective is sent to Cloud AI, which is used for re-learning with Cloud AI, and the learning result is returned to Edge AI. By repeating this series of processes, it is possible to improve the detection rate of defective products as the product is inspected.
Why Edge AI is drawing attention
With the introduction of Edge AI, “discovery on site and analysis on site” will be possible at high speed. In other words, abnormality occurrence / detection / analysis / notification can be performed at high speed without delay compared to cloud AI. This is the biggest reason why edge AI is attracting attention. This is a particularly effective technology in fields that require processing in milliseconds, such as automobiles (automated driving) and production systems.
In the field of autonomous driving, it is conceivable to transfer the judgment and experience of experienced drivers to the judgment and experience of skilled engineers in the field of production systems, and to Edge AI. In addition, edge AI will be advantageous in terms of processing speed when a problem occurs in production equipment and an emergency stop is performed. There is a shortage of experts in both areas, and edge AI is seen as an effective means of solving problems.
Specific examples of edge AI
Then, what are some concrete examples of using edge AI?
Taking quality inspection as an example, in the past inspections by humans, it was common to take the flow of “visual inspection of the product”-> “judgment of the visually-visualized image”-> “good / bad judgment”. In this “judgment” → “pass / fail judgment” part, a skilled technician makes a judgment using the “brain”. On the other hand, in the quality inspection using the automatic inspection technology in recent years, the flow of “shooting the product with a camera”-> “judging the shot image”-> “good / bad judgment” is taken. Then, it is conceivable to use artificial intelligence instead of the “brain” of a skilled technician in the “judgment of captured image” → “good / bad judgment” part.
The decision to put artificial intelligence on a cloud server or an edge server should be made based on whether the inspection requires processing speed. For example, if mass production does not allow time for quality inspection of each product, Edge AI may be suitable. On the other hand, if quality inspection takes time and you just want to save labor, cloud AI will suffice instead of edge AI.
Next, let’s take a look at failure prediction (predictive maintenance). Here, a skilled technician infers a sign of failure from the flow of, for example, “feeling the sound or vibration of the device” → “noticing those changes” → “considering the time since the previous failure” → “judgment”. It is thought that you are doing it. Therefore, if the sensor installed in the device captures changes in sound and vibration, considers the time from a timer, etc., and makes a comprehensive judgment using artificial intelligence, it is theoretically possible to replace it with artificial intelligence. (In reality, there are problems of “learning” and various small problems. Solving these problems can be said to be a showcase of the skill of the production engineering department.)
In such cases, it is easier to make a subsequent response plan (this is called a “maintenance plan”) if the time from predicting a failure to issuing an alarm is as short as possible. Therefore, it may be possible to introduce edge AI, which is advantageous in terms of speed, to gain time for making a maintenance plan.
In addition, “learning” is required to introduce artificial intelligence, but this “learning” often requires the experience of a skilled engineer. Therefore, it is likely that the final job of a skilled technician in the field at retirement age will be more and more to entrust his experience to artificial intelligence.
Edge AI is one of the effective means to deal with the shortage of engineers
So far, we have outlined Edge AI. Edge AI is a technology designed to realize the benefits of artificial intelligence as fast as possible. In the old-fashioned production line, production engineers were always on the production line, and they maintained the line based on their experience and intuition. Now that these production engineers are in short supply, edge AI can be one of the effective solutions to the labor shortage problem if judgment and experience can be successfully transferred to artificial intelligence.