Industry is generally divided into process industry and discrete industry. The biggest difference between them lies in the automation of production, the availability of data and the complexity of industry. The biggest commonality lies in that each scenario needs different requirements. Entering any sub-area requires deep enough industry knowledge and downstream resource integration capabilities.

Intelligence can be understood as digitalization and AI built on it. Starting with production line automation, multi-source heterogeneous industrial data are collected, circulated, analyzed and helped to form decision-making and control. End-to-end solutions form a typical portrait of the current industry player.

Why Industrial Intelligence?

Blue ocean

The total GDP of industry, especially manufacturing, is much higher than that of retail, financial and construction industries. In fact, the amount of effective data produced by industry is no less than that produced by Internet companies such as BAT. A large-scale factory can produce billions to tens of billions of data per day.

barrier

Although industrial scenarios generate high-frequency and massive data every day, a large number of raw data itself is not directly meaningful, and may cause large-scale delay and occupy a large amount of bandwidth. We not only need real-time monitoring and analysis in some scenarios, but also need to collect more data to the cloud for more peacekeeping and longer-term economic benefit and value analysis, which is the value of cloud computing. Cloud computing + edge computing, which is finer granularity and more complex architecture than traditional consumer Internet, also means higher barriers.。

Inflection point

A logic of the Internet is called "Copy to China" and "Copy to Industry" is the same. Large-scale data application and platform architecture have experienced sufficient verification and evolution in finance, telecommunications and other industries, and the catalytic role of Made-in-China 2025 on the policy side constitutes the prerequisite for the establishment of inflection point.

Industrial Intelligence Player Portrait

Users at this stage need not a single product, but an end-to-end solution as a whole. A qualified industrial intelligence company should have the ability to construct the overall solution.

First of all, user needs are always the first. Technologies that do not meet the needs are false propositions. In addition, a good solution starts with a perfect architecture. For industrial scenarios, from the integration of internal and external multi-source data, to the platform architecture of cloud + end, the establishment of knowledge base, the selection of appropriate models, and then to the reverse decision-making and control, the closed-loop can only be formed through a complete connection.

Overall, industrial intelligence presents a horizontal (overall architecture) +N vertical (multiple sub-industries) pattern.

Path Selection of Industrial Intelligence

For the big B customers in the industrial field, what they need at this stage is not a single product, but an end-to-end overall solution. Although this is the status quo, it is also the ultimate goal of industrial entrepreneurs. However, path selection is very important.

Regarding the development path, the mainstream in the industry believes that automation - (data) - informatization - intelligence is the reasonable order for industrial users to advance, and the former stage is the necessary condition for the beginning of the latter stage. Therefore, domestic enterprises in the field of industrial intelligence only focus on the opportunities in the field of automation for a long time, and even equate industrial intelligence with "robots" or "industrial automation". From a large number of user site practices, there is a significant sequence of these stages, but at the same time cross-penetration, iteration.

Specifically, most customers in the discrete manufacturing industry are not fully automated, so priority should be given to the completion of production line automation. Some manufacturers use industrial Ethernet and board cards to interconnect equipment, get through equipment-level data, feed back to the platform layer through MES, and realize preliminary material link without replacing the original industrial control equipment. The user acceptance is very high, the performance doubles every year, and the trend is very obvious. This kind of mode can be called "System Integration with M2M Device Material Link as the Core".

Further demand comes from the super-large head customers in discrete manufacturing industry and the vast majority of customers in process manufacturing industry. Due to the high degree of automation of production line itself, we observe that such customers have a higher degree of acceptance of information technology itself.

In addition, there is a kind of vendor that can directly cut into the top-level design, in the platform layer to serve users with large industrial data platform or scenario AI model, and solve business problems in real time. On the other hand, in the data acquisition layer, sensors are installed in some areas where the data is not perfect, intelligent detection equipment is installed, and even the production line integration of small segments is done. User acceptance of this type of model is often higher, which means that the premium of the project is often higher. We can call it "system integration with data application as the core".

Therefore, we can see three development paths, facing different customers, different scenarios, different stages of development, there are different path choices:

1. System integration with production line automation as the core;

2. System integration with M2M equipment material link as the core;

3. System integration with data application as the core.

Of course, the end result is to provide users with overall solutions to meet the needs of users as the core.

Industrial Big Data of Industrial Intelligence

First, where is the data?

▲One is management data: structured SQL data is mainly used, such as product attributes, process, production, procurement, orders, services and so on. This kind of data generally comes from ERP, SCM, PLM and even MES systems of enterprises. The amount of data itself is small, but it has great mining value.

▲The other is the data of machine operation and IoT: most of them are unstructured and streaming data, such as equipment working conditions (pressure, temperature, vibration, stress, etc.), audio and video, log text, etc. This kind of data is generally collected from the equipment PLC, SCADA and some external sensors, which have a large amount of data, high collection frequency and need to be knotted. The edge calculation is done locally.

Generally speaking, due to the fragmentation and decentralization of scenarios, industrial data itself has the characteristics of large volume, multi-source, heterogeneous and high real-time requirements. With the gradual access of 28 billion devices in the future, these characteristics will be further strengthened, which is one of the core difficulties in industrial large data services, and Internet large data is not only magnitude. Different, different structures, different applications.

Secondly, based on these industrial data, what services should the platform layer provide?

▲Complete protocol analysis: Data acquisition must first complete the industrial agreement. Take the application layer protocol as an example, the market share of EtherNet/IP and PROFINET is the largest, followed byEtherCAT、Modbus-TCP和EtherNetPOWERLINK；

▲Standardized data integration: Data collected should be managed by a unified master data management. The first step is to establish standards. Generally speaking, it is very important that we first use ISO or other industry standards to formulate a unified coding, structure, flow mode and attributes to ensure data consistency.

Standardized data integration: Data collected should be managed by a unified master data management. The first step is to establish standards. Generally speaking, it is very important that we first use ISO or other industry standards to formulate a unified coding, structure, flow mode and attributes to ensure data consistency.

Powerful PaaS support: Because of the particularity of industrial data itself, the platform must have strong mid-level support capability. We take the time series database as an example. It is a typical variety of equipment working conditions and sensor data. This kind of data has high frequency and large volume. It needs to pull all the values out every time when it is processed by traditional relational database. The throughput is very high and the performance is very poor. Therefore, a high-compression, high-performance sequential database is one of the essential capabilities of the platform layer.

Finally, what applications should we do?

▲Equipment level: quality control. In the era of industrial intelligence, if we can collect appropriate real-time data and combine the mechanism model of the equipment, it is possible to mine the correlation or causality between product quality and key data by machine learning method, and it is also possible to realize real-time online quality control and fault warning. According to the frequency can form a perfect envelope for the process, we may also achieve maximum efficiency improvement.

▲Plant level: planned production schedule. The ultimate goal of industrial intelligence is to achieve large-scale personalized customization, namely C2M. The objective of this problem is to achieve the optimal local production capacity at that time. The constraints come from the production line equipment, personnel, product attributes, supply chain data and so on. Through the learning and training of historical data, it is not difficult to form a better prediction model.

This model can be dynamically adjusted according to real-time data of production lines and factories, so as to help enterprises achieve accurate control and maximize economic benefits.

In the foreseeable future, with the increasing integrity and reliability of data and the increasing richness of scenarios, there will be a considerable number of priority enterprises at the data application level, which can help industrial users reduce costs, improve efficiency and solve real business problems.