Is predictive maintenance the shortest path for edge computing and artificial intelligence to be implemented in industry?

My previous article, "Once considered the "killer" application of the Industrial Internet, why has predictive maintenance not developed as expected?", has aroused widespread discussion in the industry. From different perspectives, including operational technology OT, information technology IT, data technology DT and other dimensions, everyone analyzed the prospects and challenges of predictive maintenance and offered suggestions together.

[[261062]]

Predictive maintenance (PdM) is an important application of the Industrial Internet, but it is not new. It was already tried in the field of aircraft engines in the 1990s. In recent years, with the gradual advancement and maturity of industrial artificial intelligence technology and edge computing technology, predictive maintenance, which was previously only used in high-end equipment, can now be used by ordinary people, and has the economic premise for large-scale application.

It should be emphasized that predictive maintenance is different from preventive maintenance. Wikipedia clearly distinguishes the two, which is worth learning from.

Predictive maintenance (PdM) evaluates the time for maintenance based on the condition of in-service equipment to prevent unexpected equipment failures. This method is more cost-effective than preventive maintenance, and the key to achieving it is "providing the right information at the right time."

Simply put, predictive maintenance relies on the actual condition of the equipment, rather than average or expected life statistics, to predict when maintenance is needed.

Going a step further, fault prediction and health management PHM should also be able to predict the remaining service life of the equipment.

There is no doubt that predictive maintenance is the future.

How can we truly leverage the predictive maintenance market and tap into new business opportunities? In the previous article, many interactions and comments were valuable, so this article will be used to continue and expand on them.

In this article you will see:

• Looking closely at the differences between China and abroad, what concerns do domestic end users have when using predictive maintenance?
• In the predictive maintenance arms race, which routes are different companies taking?
• Industrial artificial intelligence and edge computing are introducing new vectors for predictive maintenance.
• What are some of the startups worth paying attention to in China?
• A fascinating review worth watching.

There is a lot of content involved, and I have purified it to a high concentration.

1. End-user concerns

Predictive maintenance is not suitable for all objects.

Here we refer to the classification of the Intelligent Maintenance System (IMS) Center of the United States. Refer to the figure below. The vertical axis represents the frequency of failure, and the horizontal axis represents the impact after the failure. Predictive maintenance is suitable for failures that do not occur frequently but have a great impact once they occur.

• Low frequency, high impact: predictive maintenance
• Low frequency, low impact: traditional maintenance method
• High frequency and great impact: There is a problem with the system design and the design needs to be improved
• High frequency, low impact: prepare more spare parts

Effectively predicting low-frequency, high-impact faults is what many end users expect. The difficulties and risks faced by predictive maintenance in the implementation of the program are slightly different in China and abroad.

Domestic companies are more concerned about risks at the system integration level.

Common concerns among end users include:

(1) Lack of technical experts and need to improve model accuracy

The mechanism model refers to an accurate mathematical model established based on the internal mechanism of an object or production process or the transfer mechanism of material flow. This model is one of the foundations for predictive maintenance, but data thinking in the industry has just begun to be established, and experts who understand the mechanism are few after all.

A deep understanding of the mechanism of equipment requires more than ten years or even decades of long-term experience. For example, a heavy equipment in a factory always deviates from the axis, and there has been a long-term debate on whether it is a problem with the equipment or the components. It was not until an experienced old professor was found that the problem was finally discovered. It turned out that it was not a problem with the equipment itself, but a problem with the soft foundation, which caused the drive shaft to always fail.

What data to collect in predictive maintenance, how to install sensors, how to choose the frequency and cycle of collection... all need to be based on mastering the mechanism model and industry know-how. It can be said that the technical threshold of predictive maintenance is very high, especially the threshold of industry know-how knowledge is high. If you only know the technology in the IT field, you will find that you have eighteen martial arts skills, but they are completely useless.

(2) Poor data portability

You may ask why data needs to be portable? Because it is the foundation of data management strategy. Truly digital industrial enterprises must have a comprehensive understanding of the data they have, master complete and continuously updated data visibility, and migrate data on demand according to business needs.

At the same time, the ownership and usage rights of industrial data have always been sensitive topics.

Because industrial data has low value density and strong corporate attributes, it is unlikely to generate direct data transactions. However, industrial data is very valuable, and all parties hope to obtain the right to use it: end users can improve production and operations based on data; equipment service providers can provide better equipment maintenance and services based on data; equipment manufacturers can improve machine parameters and optimize equipment indicators based on data...

From the perspective of categories, industrial data can be roughly divided into two types: equipment data and operating data. Operating data involves internal business information of the enterprise, and end users do not want to share data.

The higher the accuracy of predictive maintenance, the greater the value, but also the greater the difficulty. Since data itself is an asset, end users are becoming more and more aware of data security and privacy protection. If predictive maintenance wants to improve accuracy, a large amount of data is indispensable for algorithm training. Is the data available, usable, and reusable? There are too many difficulties.

(3) Suppliers present greater risks

After predictive maintenance was touted as a "killer" application, many companies have been developing businesses in this field. Among them, there are many who use "stethoscopes" to do predictive maintenance, which makes professionals who use "CT machines" to make careful diagnoses have no choice but to smile bitterly.

In the process of advancing predictive maintenance projects, unnecessary disputes may also arise. IoT companies that provide predictive services often encounter a lot of doubts.

For example, if a problem is predicted for a device, the end user may question the result of the judgment.

If the end user is convinced of the diagnosis results and goes to the equipment manufacturer for a solution, the equipment manufacturer may believe that the failure was caused by the destruction and intrusion of the equipment during the predictive maintenance process, using embedded sensors.

If embedded sensors are not used and replaced with wearable sensors, the accuracy of prediction will be affected...

(4) Potential transaction risks and switching risks

Whether end users or equipment manufacturers, when looking for a technical partner for predictive maintenance, they will face protocol conversion and technical trade-offs, and may be bound by technical service providers in the process.

The example of Uptake is a good example. In March 2015, Caterpillar began to cooperate with Uptake in technology and data and made an equity investment in Uptake. Through the cooperation, Caterpillar effectively combines its world-leading product engineering and design capabilities with Uptake's software, application and data analysis skills to provide customers with more efficient solutions.

At the end of 2017, Caterpillar announced the termination of its investment in Uptake, no longer holding Uptake shares, and readjusted the areas of cooperation with Uptake. Caterpillar realized that if it continued to invest in Uptake, it would weaken its competitiveness, and hoped to take back the predictive maintenance capabilities within its own "body" to further promote the transformation from equipment manufacturing to manufacturing as a service (MaaS).

• Some people hold different opinions on this, believing that it may be a false proposition for equipment manufacturers to transform their business models through predictive maintenance. (@郭云)
• However, predictive maintenance still needs to be done, but it cannot be done this way... It should be combined with the current status of industrial equipment management and industrial services, from the diversified combination of diagnostic forms and the sharing of human capabilities, and from the industrial service industrialization ecology... (@杨明波-维修合作伙伴制+工业服务研究)
• At present, the application scenarios of IoT in industry are mainly monitoring and diagnosis, and the accuracy of prediction and business model still need to be further explored. (@史磊)

2. Two entry paths

The elements of industrial scenarios include people, machines, materials, methods, and environment. Predictive maintenance is mainly linked to the "machine". Just like the value chain of a car includes car owners, 4S stores, car factories, and suppliers of auto parts at all levels, the value chain of the "machine" includes:

• Final application enterprise (end user)
• Equipment service providers (agents, integrators)
• Equipment Manufacturers
• Upstream industrial automation software and hardware manufacturers

[Path 1]

A common approach for innovative IoT companies is to explore value around end users.

End users are the main body of the industry, and all upstream companies conduct business around end users, so users are also the core destination of various services.

What IoT companies do is to add sensors and industrial gateways around devices without interfering with the data and communication protocols of the original controller. They analyze and feedback the sensor data on the edge and the cloud to "sense the pulse of the device."

The process values ​​(flow, temperature, pressure, etc.) collected by the sensors are transmitted to the edge cloud for analysis, which can display a variety of evaluation charts and combine with pre-defined alarm mechanisms to ensure continuous monitoring and analysis of process values.

Where is the sensor installed? What signal is collected? What are the characteristics of the machine? What principle is analyzed? Detailed analysis is required in the early stage to ensure that the collected data is valuable. The deployment and customized analysis of each sensor takes time and money. It only makes sense to start with the most important fault point. This is a long process of discussion, trial and verification.

Many projects are not typical predictive maintenance. Simple data presentation and report analysis, status detection and remote monitoring can already meet the needs of end users.

[Path 2]

Traditional industrial automation giants, because of their advantages in private protocols and existing equipment, choose to start from the source, directly connect new hardware from the control layer, and provide predictive maintenance services from the shallow to the deep.

Let me give you two examples.

For example, Siemens launched new hardware for edge applications at the end of 2018 as part of the Industrial Edge concept. This compact edge device is based on the embedded industrial computer SIMATIC IPC227E and can directly read and process production data from the production end.

This approach is equivalent to adding an industrial computer to the traditional controller PLC, which directly reads and analyzes the control layer data, and works with the industrial Internet platform MindSphere to provide predictive maintenance and other capabilities, improving the level of on-site management in an edge-cloud collaborative manner. Moreover, when the underlying framework conditions of the industrial application program change, the application on the edge device can be adjusted synchronously to keep the device functionality updated in real time.

Just this weekend, on March 29, Siemens also signed an agreement with Volkswagen Group of Germany. Volkswagen of Germany announced that it would adopt Siemens MindSphere Industrial Internet platform, covering Volkswagen's 122 factories and 1,500 suppliers. The ripples caused by this cooperation are bound to promote the adoption of industrial cloud platforms by multinational companies and play a certain role in promoting the implementation of predictive maintenance.

In addition to direct intervention from the control level, motor and drive manufacturers are also unwilling to lag behind. In early March 2019, Mitsubishi Electric announced the launch of a new generation of universal servo drive system J5, which is not only the world's first servo product using the next-generation industrial network technology TSN, but also integrates the intelligent control technology Maisart.

Maisart is the abbreviation of Mitsubishi Electric's AI creates the State-of-the- ART in Technology. The integration of Maisart and the servo drive system J5 will directly realize the detection, diagnosis and predictive maintenance of mechanical transmission components (ball screws, belts, gears...) and drives in an embedded manner.

Innovative IoT companies and even traditional industrial automation giants have begun an arms race around predictive maintenance, or more accurately, around the huge market space created by predictive maintenance.

Whether it is predictive maintenance, quality control, remote monitoring, or asset tracking, the IoT technology behind it is the same. Therefore, predictive maintenance is an application with horizontal integration capabilities.

The equipment data collected during the predictive maintenance process includes indicators such as process, quality, performance, and efficiency, which can be extended from the equipment level to the production line level.

The more data points predictive maintenance collects, the greater the value of the data, the more thorough the understanding of the mechanism model, the richer the experience accumulated, and the stronger the ability of horizontal integration, which in turn helps enterprises solve the problems of reducing costs and improving efficiency in new ways and means.

3. New vectors for PdM

Most of the existing predictive maintenance solutions on the market are at the cloud computing or fog computing level. With the improvement of edge computing power and the development of industrial artificial intelligence, it has become more economically feasible to complete predictive maintenance at the edge.

Some companies have developed the latest generation of edge controllers equipped with industrial artificial intelligence algorithms to meet users' needs for predictive maintenance.

This type of edge controller is often equipped with "weak artificial intelligence" and integrates artificial intelligence with the control process in real time to diagnose and predict equipment abnormalities.

Simply put, "strong" AI refers to AI systems that have intelligence close to or even beyond that of humans; "weak" AI focuses on solving specific application problems based on mathematical and computer science methods, and the systems developed in this way can self-optimize. Weak AI is more suitable for solving practical problems in the industrial field.

In terms of the anomaly detection algorithm, this edge controller, which is now on the market, was developed using the "Isolation Forest" machine learning engine.

The algorithm has low memory requirements and fast processing speed. Its time complexity is also linear, which is very suitable for high-speed real-time processing. It can improve detection precision and accuracy through fine-tuning. At the same time, the algorithm is also suitable for multimodal data and can be used for highly mixed production lines that require two or more operating modes.

Based on this set of algorithms, the edge controller collects and records the data of the target object, and based on the preset application model, analyzes the historical trend of equipment behavior, predicts possible abnormal conditions of the machine, and prompts and guides the user to take reasonable and effective diagnosis and maintenance measures for the electromechanical system at the appropriate time, and even JIT (Just-in-time) maintenance.

Currently, the predictive maintenance capabilities of the edge controller are mainly aimed at mechanical transmission components such as cylinders, ball screws, conveyor belts, and synchronous wheels.

It can be predicted that the performance of such a preventive maintenance program will largely depend on two key factors:

• Data collection: helps effectively understand machine behavior;
• Apply algorithms: Create and optimize learning models and make accurate judgments.

Based on industrial artificial intelligence algorithms, edge controllers have the potential to identify abnormal conditions that cannot be detected by programs or distinguished by the human eye, thereby achieving a more comprehensive and accurate learning and understanding of device behavior.

4. Typical domestic enterprises

In the field of predictive maintenance, there are several companies in China that are worth paying attention to. Here are two examples:

(1) Tianze Zhiyun

Tianze Zhiyun is an industrial intelligence company. Its industrial big data platform is mainly based on analytical algorithms, engineering expert domain knowledge, and software products. It provides industrial enterprises with customized solutions that integrate various industrial big data application services such as integrated data collection and analysis, equipment health management, fault prediction and diagnosis, maintenance decision-making and optimization.

The company has set a goal to achieve 100 industrial scenarios with worry-free production and worry-free operations within 10 years.

At present, Tianze Zhiyun mainly focuses on the fields of energy, rail transit, CNC machining, industrial robots, equipment manufacturing, etc., providing customers with industrial big data application consulting, data collection and analysis, equipment status assessment and other big data analysis solutions.

(2) Xi'an Yinlian

Xi'an Inlink provides complete data access integration solutions based on a deep understanding of equipment operation scenarios and equipment mechanisms. It excels in combining analysis of vibration signals with sensing, IoT technology, data algorithms and other technologies to provide industrial users with more efficient and reliable predictive equipment maintenance solutions.

Inlink's intelligent equipment operation and maintenance solution can identify early equipment failures three months in advance, with a fault identification accuracy rate of 90%, helping companies to prevent major safety accidents, reduce daily downtime and maintenance time by more than 30%, save more than 30% of maintenance costs, and help companies effectively improve production efficiency. Inlink has created a machine intelligent operation and maintenance SaaS platform and a machine health big data platform product solution, which has been successfully applied by customers in the steel, chemical, electric power, cement, papermaking, automobile, food and beverage and other industries.

Finally, before you enjoy the wonderful comments, I would like to sincerely thank Wang Yu, Marketing Director of Advantech Technology, and Zheng Yuming, founder of the Smart Manufacturing Column, for their strong support during the writing process.

5. Excellent comments

(1) Tsinghua alumni:

What I know is that the awareness of predictive maintenance in China is still in its infancy, but the awareness and willingness of large enterprises are gradually increasing. China needs to emphasize the value of early warning rather than predictive maintenance. The national conditions are that safe production is more important than cost reduction.

The international industrial community has a relatively high definition of predictive maintenance because they already do a relatively detailed job of equipment management. Domestic companies need to improve their foundation first. They can use the idea of ​​predictive maintenance to pay attention to equipment data analysis.

(2) Xue Chenyu:

The value of IoT is to reshape the industrial chain through data, such as launching supply chain finance and insurance business through predictive maintenance, firstly to reduce capital occupation, and secondly to pay for the results and solve customer pain points. Current customers may be hesitant about the value of this transformation, but some catfish will emerge to disrupt the existing value chain, such as entrepreneurs in the sharing economy or leasing economy. Change often comes from the outside, and few people have the courage to revolutionize themselves! If you stay in the existing value chain, predictive maintenance cannot be realized. The sharing economy is actually an application scenario of predictive maintenance.

(3) mcrazy:

Predictive maintenance has been a very mature application in many fields, such as wind power, railways... In my opinion, the reason why many people link the Internet with this technology and give it extremely high expectations is because of the possible economic benefits that can help it popularize in more fields. My suggestion is to study various product solutions and see how they can be transformed into Internet-based solutions. Then look at the actual demands of corporate users, and you will roughly understand why there is such a long implementation cycle; the things are good, but it still takes some patience to face reality... The key lies in: what to predict vs. what to maintain... The former is an uncertain risk that determines the required investment, and the latter is a certain application scenario that determines the required investment budget.

(4) Zhao Min:

The so-called "killer applications", the so-called explosive and phenomenal applications, are imagined to appear in the industrial field, which is purely the YY of some people.

(5) Tang Dong:

The fruit of predictive maintenance is good, but it hangs high, so pick the fruit below first.

(6) Chris Gao:

It requires the promotion of social environment and industry planning, the business model determines the investment in technological innovation, the equipment model and mechanism knowledge base and real-time dynamic closed-loop analysis achieve predictive maintenance, and the industrial information transformation is the cognitive process of human beings to eliminate uncertainty risks.

(7) Mr. Wang:

Two or three years ago, I studied predictive maintenance. The idea was to use data to build a finite state machine model of system operation. Then, preventive measures were taken based on the deviation between the actual data and the prediction of the state machine. Later, I found several problems. The first was that the data volume and dimensions were too large. Basically, the data was segmented in a vector space of at least dozens of dimensions. The amount of calculation was too large. The second was that the error in the actual data greatly affected the results. The third was that the data volume of many devices was insufficient. However, I still believe that the application of predictive analysis on industrial equipment is promising. Because the operating conditions of industrial equipment are relatively stable. After all, it is suitable for predictive maintenance.

(8) Xiao Hu:

The return on investment is difficult to calculate: Industrial Internet is first industrial and then network. Industrial means many types, low industry concentration and complicated details.

In terms of business model, saving money does not necessarily mean making money. The effect depends on whether it is before or after the event, and whether it is active or passive.

The foundation is not solid, the amount of data is insufficient, and intelligent services mean that data flows automatically throughout the entire life cycle. How far an enterprise's value chain can extend and how to judge it remain to be discussed.

(9) Guo Zhaohui:

Ever since this concept came into being, I have never been optimistic about it and have repeatedly criticized this idea for being unrealistic and unreliable for most applications.

(10) Birdman:

It is better to do self-repair at a deeper level than to do preventive maintenance. It is not so easy to prevent equipment failure. For example, the equipment in an electronics factory has thousands of parts and tens of thousands of wires. How to monitor them? Some chips are programmed, and many wirings are hidden. How to determine the location of wear and aging? The core of equipment prevention is regular maintenance, but many companies do not pay much attention to it. The possible direction of predictive maintenance is to be applied only in a few standardized industries and standardized equipment, rather than on a large scale.

(11) Lester:

Predictive maintenance also requires the support of physical models, rather than relying solely on big data analysis. At the same time, data must be a combination of sensor data and process data, otherwise the conclusions obtained are just blind men touching an elephant.

(12) Little Stonemason · Old Chang @ Distribution Network:

Equipment safety issues must be a comprehensive solution. It is impossible to artificially elevate any single solution in the maintenance plan to a level that is detached from objective needs.

Some Internet wizards are addicted to subversion, cutting across or overtaking on the curve. We have been engaged in equipment safety and maintenance solutions since 1998. We know the problems of condition-based maintenance or predictive maintenance. It is not that easy. If we could solve the problem, we old birds would have done it long ago. Would you guys have the chance to cut across?

As early as 2010, I went to the State Grid Corporation of China to talk to experts about transformer fault prediction. We still understand some failure modes. As a result, the experts told me that it is still possible to make a few major indicators, but there are too many types of transformer failures, the mechanisms are very complex, and the fault model cannot be refined.

Nowadays, the electrical complexity of processes and equipment in the process chain is getting higher and higher, and the operating speed is getting higher and higher. Fault models are too difficult to deal with. Without fault models, how can we predict faults? Can we still use the threshold judgment of the past?

Of course, two years ago I suggested to the beautiful expert from the former Huaneng Yangluo Power Plant that she should master the equipment's failure model through statistical methods, which on the surface looks very similar to today's very fashionable artificial intelligence iterative learning.

The problem is that the same equipment is used in different working conditions and environments, and the environment is very complex. Who knows how effective it is? Moreover, artificial intelligence iterative learning belongs to experimental science, and there is no theory with strong constraints. It is very likely that it will be reliable this month and completely unreliable next month. Enterprise security is a big deal. Who will be responsible for security at that time?

Although it is experimental science, I have recently been suggesting to my boss that he should make up his mind to give it a try and do an experiment with our high-voltage cabinet.

(13) Peng Guohua - Industrial Service Alliance:

For a few equipment such as aircraft engines and rail transit, where unplanned downtime must be controlled within 0%, predictive maintenance is a must, but manufacturers have already done it. For equipment where unplanned downtime is allowed within 5-10%, predictive maintenance is only an option, and most of them are very uneconomical options, with a low cost-effectiveness compared to other management measures, so end users are not active.

(14) Roger:

Service providers with industry attributes will have obvious opportunities and advantages. General service providers that provide services on a project basis not only have a long preparation cycle and high risk of customized development, but will also be squeezed from both users and equipment manufacturers.

(15) The Great Demon King Luo:

Equipment operation status detection requires in-depth research in the sensor field, but most domestic IoT companies focus on communication technology. Moreover, how to establish an equipment status evaluation model is also a technical difficulty.

The cost of fully integrating these three technologies and incubating mature solutions is not affordable for ordinary small and medium-sized enterprises. In addition, the replicability of sensors and evaluation models is not high, and a solution can only be expanded in one industry. Therefore, it takes more patience to develop a predictive maintenance solution that is applicable to all industries.

(16) SHOUT:

The analysis is very thorough! Using traditional methods (sensors, expert models) is difficult to solve the problems of high cost and difficulty in replication, while using pure IT methods (access to things, machine learning) will lead to inaccurate predictions and difficult evaluations. A breakthrough technology must be developed to achieve a low-cost, easy-to-replicate, relatively accurate and evaluable system before a large number of applications can be realized. At present, some innovative models are running in a low-key manner, and we look forward to the day when they will explode!

(17) Andrew, evangelist of the Industrial Internet:

In 2008, I brought APM back to China. At that time, Sinopec and PetroChina were still doing RBI verification. Ten years have passed in a flash.

In fact, predictive maintenance of equipment is not a panacea. It has been hyped up by a group of IT laymen in recent years.

Taking petrochemical enterprises as an example, only the vibration model is at the mechanism level, while others such as corrosion and lubrication are far from the theoretical results.

So, is it possible to determine the cause of failure through big data?

Haha, this is interesting.

Good device, imported equipment is OK.

Why?

Because reliability issues are taken into consideration in their designs (design review), and there is complete reliability experimental data (not made up), such as stress distribution in key parts, statistics on corrosion rates at different locations of the pipeline, etc.

Through "digital handover", the process, equipment data and process package are completely unified and visually handed over to the owner. The owner can trust the process data. These data can be used to fit the theoretical distribution and then perform predictive regression...

Even so, in many cases, due to construction deviations and schedule requirements, the delivered system is no longer the same as the original design, and the owner is too lazy to do CA, find the bad actor, and is even more reluctant to do RCM. They just dream of passing PI, IP21, and letting the engineers' brains replace the expensive system.

These Chinese clever tricks have harmed the intelligentization process of Chinese companies.

(18) Mu Ming:

There is still a market for predictive maintenance. Don't try to cut corners, just do it steadily.

(19) Guo Ce:

Big data is not feasible from the perspective of cost or technology. Using traditional small data for detection and modeling is low-cost and has decades of operation and theoretical basis, which is more practical. If the main equipment and main auxiliary equipment can be predicted for a week or month, it will have management significance.

You can refer to the model data when making quarterly equipment maintenance plans and monthly spare parts procurement plans.

(20) Wu Yifei-Shanghai Yuzheng:

The gap between time and expected effect. The essence of "predictiveness" is to abstract mathematical calculation models based on sufficient data sample training, through the understanding of known phenomena by senior industry personnel and the collision between computer experts. Sufficient basic data, human intervention in the process, and verification and optimization of the model all need to be based on time. If we can rationally treat the gap between expected value and actual effect, and rationally recognize AI, AI is currently just a tool to improve efficiency. I believe that "killer" applications will eventually appear.

(21) Anyway:

The focus of predictive maintenance should be the mutual influence of human-machine interaction, such as pilots and drivers, to provide evaluation of the effectiveness of manual operation. In addition to the reliability theory of raw material failure, its existence is based on the working environment of the predicted object, especially to identify the potential external stress of the equipment, such as thermal stress, structural stress, lubrication wear and other factors. At the same time, it also provides quantitative data support for the optimization of operating procedures.

We should stand in the position of the prediction object and discuss the customized design of the prediction method. When a general prediction model is implemented, it is necessary to consider introducing middleware or plug-in technology to provide an extensible interface for secondary development...

Predictive maintenance must be a business that covers the six elements of man, machine, material, method, environment and measurement, based on the experience and common sense of reliability engineering (not to mention professional or specific technology)...

Summary:

• Predictive maintenance is suitable for failures that occur infrequently but have a significant impact once they occur.
• Whether it is predictive maintenance, quality control, remote monitoring, or asset tracking, the IoT technology behind it is the same. Therefore, predictive maintenance is an application with horizontal integration capabilities.
• Most of the existing predictive maintenance solutions on the market are at the cloud computing or fog computing level. With the improvement of edge computing power and the development of industrial artificial intelligence, it has become more economically feasible to complete predictive maintenance at the edge.