Directly hit the Huawei Cloud IoT smart manufacturing behind the Huawei Mate 40 production line

Abstract: digital twin? Find the equipment of the physical world in the digital world!

This article is shared from the HUAWEI cloud community " [Cloud-based co-creation] HUAWEI CLOUD IoT intelligent manufacturing directly hit by the Huawei Mate 40 production line ", the original author: Qiming.

Part 1: Digital Twin in the Era of Intelligent Industry 4.0

1. Industry 4.0, the era of intelligence has come

Looking back on human history, we have successfully experienced three industrial revolutions together:

The first time was the steam engine era, which created an era where machines replaced manual labor; the second time was the electrical era, where the development of natural science and industry were closely integrated, and science played a more important role in promoting productivity; the third time was In the information age, the speed at which science and technology are transformed into direct productivity is rapidly accelerating.

Today, we have ushered in the fourth revolution, namely, Industry 4.0: the era of intelligence. "The essence of Industry 4.0 is to use data flow automation technology to shift from economies of scale to economies of scope, and to build a heterogeneous and customized industry at the cost of homogenization and scale. This is essential for the reform of the industrial structure. effect."

As a new round of industrial revolution, the core feature of Industry 4.0 is interconnection. Industry 4.0 represents the intelligent production of "Internet + manufacturing", bred a large number of new business models, and can truly help realize the "C2B2C" business model.

Second, the current pains and difficulties of digital transformation of factories

At present, everyone is still in the groping stage of "Industry 4.0". A large number of factories have begun their own path of intelligent transformation, such as visualization of collected data by building applications, and maximizing the value of data. However, in the course of this practice, problems continue to emerge, such as:

2.1 Data/information islands with many chimneys

A factory, at different stages, because of different projects, it is possible to find different suppliers to undertake. Segmented project suppliers lead to different system applications. To put it in perspective, if multiple systems are not interoperable, they are like independent "chimneys". Each "chimney" has "smoke", but they are not interoperable. In the industry 4.0 stage, not interoperable, which means information islands, which means that the company's digital assets are scattered, maintenance costs are high, and use efficiency is low;

2.2 Slow application launch, time-consuming and labor-intensive

As mentioned in the first point, the non-interoperability between different systems results in the "repetitive wheels" when new applications are launched: each application is online, there is a lot of repetitive work, waste of manpower and material resources, and time-consuming. More importantly, the data processing problems brought about by new applications: Due to the lack of unified modeling, each application needs to process the original data repeatedly. Two "duplications" make the already high cost even more "worse";

2.3 The threshold of data analysis is high

Factories, or companies, have a desire to reduce costs and increase efficiency. For example, they want to find patterns by analyzing existing data to optimize processes, but they are discouraged because of the high threshold of data analysis. The most critical reason for this is that the business scenario is not clear, and a good data platform has not been found.

3. Finding the right platform is half the battle

The above pain points and difficulties are encountered by most manufacturers in the industrial field in the process of exploring "Industry 4.0", and "applications" run through them. In other words, software developers did not achieve sufficient layered decoupling is one of the important reasons for the above problems . Based on "application", the factory has experienced three modes in three periods:

3.1 Mode 1: "Chimney" application

Before Industry 4.0, due to the lack of application and practice, most manufacturers' applications, as mentioned above, were "chimney-style":

This leads to a lack of overall planning, independent deployment of each application, and separate collection and use of data based on business needs; second, low efficiency, such as repeated data collection, which has a greater impact on production.

3.2 Mode 2: Platform Decoupling-Unified Data Acquisition Platform

After the concept of "platform" was put forward, factory managers gradually realized that perhaps, between the production line and the application, a "platform" is needed. Such decoupling can enable the application and the production line, and the interconnection between applications and applications. . And this is a basic model of Industry 4.0.

The emergence of the second mode allows the professional data collection team to complete as much data collection as possible, and it is centralized and open, so that the overall efficiency is improved. But we can find that even so, the use of data is still independent, and no real integration has been achieved. The data obtained in the production line or production equipment is still metadata. After the application obtains the data, the data still needs to be used for secondary processing separately, resulting in a lot of duplication of data processing between applications.

3.3 Mode 3: Data Processing-Unified Twin Model

Huawei IoT has its own methods to solve the problems of "application decoupling" and "data unified processing" simultaneously.

In the field of Internet of Things, there is a concept of "twin". Through the "twin body", the perception of the device and the cognition of the device are processed in a unified manner. Also take the factory as an example. There are a lot of production equipment, production lines, and various other physical equipment in the factory. Can we help the factory to integrate all these physical equipment through unified modeling? These devices are processed one by one, abstracted into a digital image?

The answer is yes. By digitizing the physical objects one by one, the interaction between the application and the physical device is transformed into the interaction between the application and the digital twin. Compared with the first two models, the development method of this model has a very big change: we can ignore the physical devices or physical interfaces at the bottom and transfer the data modeling part to the IoT's "unified twin model". Layer" is complete.

The concept of "twin" means that when we are modeling, we first need to have a clear understanding of the model, that is, a wide range of data acquisition capabilities. After all, in the factory, there will be a variety of equipment, these equipment At the same time, there are a variety of protocols; secondly, you need to have a very high abstraction ability. You need to abstract the equipment in the physical world into a model in the digital world and be able to interact.

Data acquisition and abstraction capabilities are two very critical capabilities in the current development of IoT applications.

Based on HUAWEI CLOUD IoT brings a new development model to help users quickly build a basic platform for digital transformation.

Next, take Huawei's own factory as an example to briefly explain how Huawei Cloud IoT uses a new development model to help the factory's digital transformation.

As everyone knows, Huawei itself is also a manufacturing factory, and the capabilities of Huawei Cloud IoT are first implemented in its own factories. We take the southern factory, which is the Huawei Mate 40 production factory, as an example, and build a digital production line twin in the digital world through the mobile phone patching process of the factory through data acquisition and modeling.

Based on the capabilities of Huawei Cloud IoT, a unified twin is completed in the south direction, and visual and intelligent applications are built on the upper layer. The specific architecture diagram is shown below:

In the actual digitization process of the southern factory, there are the following challenges:

• The production line equipment manufacturers/types/models are diverse, involving more than 30+ different application layer protocols that need to be connected, which is difficult to collect;
• There are more than thousands of measurement points on a production line, and the lack of data modeling methods has led to poor data processing.

So, how to stand at the developer's point of view, to save time and effort to complete the digitalization? HUAWEI CLOUD IoT officially debuted.

4. Build a digital twin with a multi-dimensional model as the core

Behind the practical application of a digital twin, there are many models, such as production line models, equipment models, quality defect models, and so on. In the modeling process, from the perspective of looking at the physical objects in the physical world of a factory, factory twins can be divided into two categories: manufactured digital twins and product digital twins.

makes a digital twin:

• positioning: digitally mirrors the manufacturing links of the factory, which can reflect the manufacturing process of the factory in real time; after a unified abstraction of the manufacturing process, different applications can interact based on the same semantics;
• modeling content: production equipment, production lines, production process, quality defects, physical structure, etc.;

product digital twin:

• positioning: organizes the various data generated during the production process from the dimension of the factory's products in production, and reserves data for the product design phase and product maintenance phase through the ability to connect with the digital main line;
• modeling content: product various attributes, production process data, quality data, etc.
  

The above are the two very important digital abstract dimensions of the factory digital twin. By making the production process transparent, the production is orderly and controllable, and the application launch time has been reduced from 6-9 months to 3 months; at the same time, twin modeling + intelligent analysis, using data to drive intelligent production , So that the efficiency of data development can be increased by 70%. Through Huawei Cloud IoT, we can , and by building a factory digital twin model, we can greatly improve the efficiency of data utilization.

Part 2: Digital twin practice based on the southern factory

Back to our topic. The southern plant is the production line for the Huawei Mate 40. The rapid increase in the production of mobile phones has made the digital demand of the production line imminent. Through the digitization of the entire production line, the production process can be improved, the management of manufacturing engineering vendors, and the management of quality control can be optimized, so that the efficiency of the production line can be greatly improved, and the cost of operation can be reduced at the same time.

The picture above is a multi-dimensional model of a factory twin. We can see that in the product model, there are equipment models and production line models. There are also process capability models, quality defect models, equipment physical/structural models, and equipment failure prediction models at the upper level.

By applying the modeling and analysis capabilities of Huawei Cloud IoT data analysis services, electronic engineering production lines and equipment twins can be quickly constructed. So in this article we will introduce how to build a data analysis service modeling.

1. Introduction to basic concepts

(1) Introduction to OEE concept

Before modeling and explaining, we first popularize a basic concept. OEE stands for Overall Equipment Effectiveness (Overall Equipment Effectiveness). Generally speaking, every production equipment has its own theoretical capacity. To realize this theoretical capacity, it must be guaranteed that there is no interference and quality loss. OEE is used to express the ratio of the production capacity of the equipment to the theoretical capacity.

When calculating OEE, three dimensions are involved:

1. time utilization rate: time utilization rate = Σ actual running time / Σ planned start-up time * 100%. Used to evaluate the loss caused by the shutdown, including any event that caused the planned production shutdown, such as equipment failure, shortage of raw materials, and changes in production methods;
   2. performance utilization rate: performance utilization rate = Σ[output quantity cycle time for a product to be processed under the condition of the equipment]/Σ actual running time 100%. Used to evaluate the loss in production speed. Including any factors that cause production to fail to run at maximum speed, such as equipment wear, unqualified materials, and operator errors;
   3. pass rate: pass rate=[quantity of qualified output]/[quantity of output]*100%. Used to evaluate the loss of quality, it is used to reflect the products that do not meet the quality requirements (including reworked products);

Then the final calculation formula is: OEE=[Time Utilization] [Performance Utilization] [Qualification Rate]*100%, which is a key indicator to measure the overall operating efficiency of equipment, and it is also a key indicator for many electronic manufacturing plants and other similar plants A key indicator in the

Generally speaking, the value of OEE of domestic manufacturers is not too high, generally only 70%, or 80%, or even only about 40%.

(2) Modeling analysis effect diagram of factory twin production line and equipment

The factory twin production line and equipment modeling analysis can be viewed through some visual management backends. The following are the renderings of three different functions.

Picture 1: There are 3 production lines on the picture, which can be dragged and dropped appropriately. You can see the OEE value of each device in the graph. Through asset modeling and analysis capabilities, the OEE of production lines and equipment can be calculated in real time, key indicators of each equipment can be monitored in real time, and historical data can be viewed at the same time.

Picture 2: Equipment modeling diagram. Through the combination of equipment reporting fault messages and equipment models, real-time monitoring of equipment operating status.

Picture 3: Asset analysis chart. Through the asset model analysis capabilities, real-time analysis and monitoring of reported device data are possible for abnormalities. For example, the humidity is 45%~63% under normal conditions. If the reported data is not within this range, it is considered abnormal data. The interface will display a yellow dot, indicating that the data reported by the device is abnormal. It can be seen that data analysis can be calculated and monitored in real time, and if there are some serious abnormalities, it can even be pushed to operation and maintenance personnel.

(3) Data processing and analysis process of the factory digital twin Demo

To achieve the above renderings, we need to go through the following steps (because it is not a real factory, it is a simulated device):

• device simulator: based on the standard object model. The simulator automatically reports device attribute data through the MQTT protocol at a fixed time of 5 seconds. It can simulate a manual start to report a message, such as a device out of service message.
• IoT device access service: sends device attribute data and device messages to IoTA (data analysis) service by configuring device data forwarding rules.
• IoT data analysis service: receives equipment data based on the data pipeline, and uses asset modeling and calculation analysis capabilities to calculate and generate real-time production line and equipment OEE-related data to determine whether there is abnormal information in the data.
• 3D application: obtains data by calling IoTA's API, and displays the production line and equipment in 3D. You can view the production line and equipment OEE, equipment key indicators, run-out and other fault information, and you can also find relevant historical data. This is the rendering of the second part.
  

(4) The analysis process in IoT data analysis

Next, let's focus on the internal circulation process of "IoT Data Analysis Service".

The first step is the data pipeline. We connect the data through the data pipeline, and at the same time, we will back up the data locally;

The second step is to model the equipment;

The third step is to establish equipment assets;

The fourth step is to complete real-time calculation-related analysis tasks through the calculation engine of the equipment asset analysis of the equipment after the model is instantiated and the data poured in;

The fifth step is to store the data inside the IoT;

The sixth step is to make this data available to third parties through the API.

See the figure below for details:

In this process, we need to explain in detail how the second and third steps are operated, that is, how do we create models and assets?

(5) Introduction to the basic concepts of IoT digital twins

Before proceeding with the creation of models and asset explanations, let's first introduce the basic concept of "IoT digital twin".

We believe that the physical world has a real-time and accurate mapping in the digital world. It can organize the actual device data and some other data to form a jason model and become a carrier.

The picture above is a conceptual picture of our digital twin.

First, the data twin can be divided into two parts: model and asset. The model is equivalent to a Java class in the development process, which represents a template of a class. Generating an asset after instantiation is equivalent to a new class, and then an object is also generated. One object corresponds to one asset.

At the same time, the model is divided into two types. The first one is attribute. The attribute can also be divided into three types in the past:

first type of 160c09830abfd0 is static configuration attributes, such attributes do not need to be reported by the device, and will not change much, such as product model, device type, etc.;

is the measurement data attribute, and the measurement data attribute needs to be reported by the device. In layman's terms, data analysis cannot be obtained by itself, and data provided to the system by others is needed. Including the attributes reported by the device, and possibly the attributes read from the third-party business system, which are considered by the system to be a measurement attribute;

third type of 160c09830abff7 is to analyze task attributes, such attributes need to be further calculated after the data is reported.

For the last task analysis attribute, there are corresponding tasks to configure and calculate. In this process, it is equivalent to the loading and configuration of the algorithm: first analyze the data, and then the background computing engine loads the configured business logic. There are currently three types of analysis task attributes:

first type is conversion calculation: For a simple example, suppose it contains two attributes, a and b, when it is created, and we require that in this process, a+b=c, then this is a conversion calculation. The conversion attribute requirements are real-time, and the data timestamps of the two values of ab are the same;

The second type of aggregation is a time dimension calculation. Assuming that an average temperature in the past five minutes is required, if the device reports data every five seconds, then it needs to do all the data reported within five minutes An average is equivalent to doing aggregation operations in the time dimension;

third type of 160c09830ac043 is flow calculation: flow calculation is mainly used in more complex scenarios, and when the logic cannot be expressed by a simple if /else, it is necessary to use flow calculation. For example, when the asset reports many parameters, the system needs to calculate a result based on these parameters, and then return the asset, then the role of flow calculation in it is equivalent to a calculator. The function of stream computing is very powerful. In the digital model of the factory, most of the scenarios can be realized, such as sliding windows, data filtering, adding attributes, etc., which is a relatively common capability.

The above is an overall modeling concept. Based on the above concepts, we can better understand the following content.

2. Asset modeling practice

(1) Equipment modeling: SMT production line printing press equipment

When building a digital asset model for things in the physical world, you must first define the asset model, and then create the asset. Generally speaking, a production line has 7 types of equipment. Let's take the printing press as an example to explain how the equipment is modeled.

First of all, it is the configuration of attributes. aimed at printing presses, our three attributes are:

Static configuration attributes: product ideal printing time, equipment model

Measurement data attributes: printing speed, demolding speed, printing height

Analyze task attributes: time utilization, performance utilization, pass rate, OEE

and the analysis task attribute also has the following calculation configuration:

Conversion calculation: calculating time utilization, calculating performance utilization, calculating OEE and judging temperature status

Aggregate calculation: calculate the actual working time, calculate the actual working time, calculate the pass rate

Stream computing: SMT scene is not used yet

The following figure shows the page for property editing, including static configuration, measurement data, and analysis tasks for reference.

The figure below is a complete sample after all parameters are equipped. There are about 70 attributes that can be seen in it, and these attributes are simulating some attributes of the real industry. All the data in the figure below, including samples and formats, are from the actual production data of the southern factory, so they are relatively real.

Through the screenshot below, we hope to explain how the analysis task of the printing press is configured. Taking "conversion calculation" as an example, you only need to read the reported temperature value and make an expression judgment. For example, if the temperature is greater than 25 and less than 35, then it is considered normal temperature. Copy the result of the judgment to the application, and the application can use the result directly.

The figure below shows the configured analysis task. It can be seen that we have currently configured 11 analysis tasks, including the calculation of capital interest rate, performance utilization, qualification rate, OEE, and various status judgments, etc. The types mentioned above.

(2) Production line modeling: SMT production line

Having finished "equipment modeling", let's explain about "production line modeling".

Production line modeling is actually the same concept as equipment modeling, and the model is similar. However, static attributes and measurement data attributes are temporarily not configured, because the production line is relatively simple, mainly to find the value of OEE, that is, to analyze task attributes, including four indicators related to OEE, as well as conversion calculation, aggregation calculation and flow calculation.

The configuration of the analysis task attribute is consistent with the equipment production line, so the explanation will not be repeated.

The following figure is an example of the equipment asset configuration diagram of a printing press:

Next, let's take a look at how production line assets are constructed. As shown in the figure below, the production line assets are divided into three layers:

The first layer is the factory (parent asset);

The second layer is the production line (sub-assets);

The third layer is equipment (sub-assets).

Production lines and equipment also have models, and the three-tier model constitutes a "parent-child relationship" of assets. Assets come from the model and are instantiated from the model. At the same time, when the model is instantiated as an asset, the hierarchical relationship can be specified according to the business scenario, and the assets are independent of each other.

The figure below is the constructed asset tree. Compared with the logic diagram in the previous picture, this is an example diagram. The figure shows that an electronics factory has three SMT production lines, and each production line has 7 SMT equipment

(Three), OEE related index configuration (equipment & production line)

Let's take a look at how each indicator of the device is calculated, as shown in the figure below. Let's take the "product qualification rate" (the gray part in the figure below) as an example.

As mentioned earlier, the pass rate = [quantity of qualified output]/[quantity of output]*100%. "TS_Sum" in the table represents time series summation, that is, the output can be summed within a time range, for example, the sum of the output within five minutes. The calculation methods of other indicators are similar to the pass rate, so I won’t repeat them one by one.

The indicator calculation process for production line and equipment is similar, the difference lies in the source of the data. The data of the production line comes from sub-assets, not the production line itself. Because the data "between father and son" of assets can be referenced to each other, and the production line itself does not report any data.

(4) Asset operation monitoring

After all the product creation and attribute configuration are completed, you can click "Publish" to publish and run the model. When the model is defined, it is a static process, and once it is released, it will be activated. According to the task analysis logic defined in the preamble, the system will automatically calculate and obtain real-time results for reporting. All the data can be observed in the figure below.

In addition to the above-mentioned data display mode, you can also display the data as a line graph, heat map, curve graph, etc. according to the needs of the business, a graphical display method that is easier to analyze, and get the results you want. The sample diagram is as follows.

To experience the process of establishing a production line model, you can go to Huawei Cloud IoT Data Analysis Service ( https://www.huaweicloud.com/product/iotanalytics-platform.html) for in-depth experience. Follow the instructions on the "Overview" page step by step.

Three, summary

From the above explanation and introduction, we can draw the following conclusions:

• The real-time and accurate mapping of objects in the physical world in the digital world, the carrier of organizing data & models, is the digital twin of the IoT field;
• Organize data & models around a specific physical object, and define a digital model, which is the digital modeling process in the IoT field;
• The equipment digital twin model is composed of two parts: attribute and task analysis;
• When building an asset model for things in the physical world, you must first define the model and then create the asset
• The benefits of object-oriented modeling ideas: encapsulation, inheritance, composition, and improve reuse efficiency and scalability.

Based on the IoT asset model, Huawei Cloud IoT data analysis service integrates IoT data integration, cleaning, storage, analysis, and visualization to provide one-stop services for IoT data developers, lowering development thresholds, shortening development cycles, and quickly realizing the value of IoT data , Making the digital transformation and upgrading of the factory "at your fingertips."

Come and experience it together~

Click to follow, and get to know the fresh technology of