Real-Time Energy Consumption Sensing System in SMT Intelligent Workshop

for future researchers.


Introduction
China has the most quantities of the manufacturing systems and mechanical equipment all around the world, and the total energy consumption is the largest.While the energy utilization is lower than 30%.In 2019, the carbon dioxide generated by China reaches to 10 billion tons, making up to 30% in the whole world (33.1 billion tons).What's worse, it still maintains an annual growth of about 1.5%.The US has an annual emission of 4.8 billion tons, and the rate continues to decline both in US and EU.On the other hand, from carbon peaking to carbon neutrality, it is 43 years from 2007 to 2050 planed by US, and it is 71 years for EU.While it is only 30 years for China [1].So, ensure the "carbon peaking and carbon neutrality" goal is a crucial part for China with good performance, low cost and limited time.The nature of energy consumption sensing and energy efficiency controlling provide great insights about concept of innovation, coordination, greenness, openness and sharing [2], helping to prioritize efforts for reducing the carbon dioxide emissions per unit of GDP [3].
The primary idea of energy consumption [4][5] and energy efficiency is mainly divided into two levels.The workshop-level mainly includes the modeling, the optimization and the evaluation of energy consumption and energy efficiency in the manufacturing systems.For example, Liu [6] put forward four characteristics of energy consumption state and energy efficiency evaluation: multienergy source characteristics and hierarchical distribution characteristics, energy consumption process and transient energy efficiency complex variability, and product life cycle process characteristics.He [7] analyzed the energy consumption characteristics of the processing task scheduling oriented to the flexible process route of the mechanical workshop based on the influence characteristics of the flexible process route of the mechanical workshop task on the energy consumption of the mechanical workshop scheduling, processing completion time, machine tool load as the goal of energy-saving scheduling model, and further proposed an improved Q-learning algorithm to solve the model to obtain its Pareto solution, thereby improving the energy efficiency of the machine shop.Zhang [8] established an energy efficient flexible job shop scheduling problem model by considering the transportation time in response to the energy-saving operation requirements of flexible job shops, and used genetic algorithms to solve the model.It also reflects the method of realizing energysaving operation of workshop manufacturing through machine tool selection and process sequencing.He [9][10] formed a preheating-quenching model and a preheatingtempering model for the forging preheating and heating workshop.The optimization model with the capacity difference as the main function was carried out for the typical paradigm of energy saving and consumption reduction in the forging workshop.
Additionally, equipment-level energy-saving design is another major approach [11]: The Fraunhofer Institute of Machine Tool and Forming Technology in Germany proposes an energy efficiency control model with energy autonomy and zero emissions.Japan Mori Seiki modeled the multi-source energy consumption of machine tools and proposed a method for processing energy consumption.Massachusetts Institute of Technology analyzed the motor energy flow [12] to reveal device-level energy distribution.Zhang [13] proposed a new bond graph method for modelling the dynamics characteristics and energy consumption of the spindle system of a CNC machine tool to comprehensively consider the energy consumption and dynamics characteristics of the mechanical transmission of the spindle system.He [14] estimated the machining feature energy consumption under flexible machining configurations.This method can help manufacturers to select optimal energy-efficient machining configurations for performing workpieces and provide designers with knowledge of energy consumption to conduct energy-saving design of features.
After the above literature review, due to the complex types of equipment and large energy nodes, the energy consumption research mostly stays at the factory or workshop level and facing the following problems: 1) Multi-level, multi-source energy consumption increase the modeling complexity [15].2) The real-time responsiveness of the energy efficiency sensing network is challenged by the large number of energy sensing nodes and the group control performance requirements of logistics /distribution equipment.
3) The delayed transmission data cannot form an effective online energy efficiency evaluation and optimization control.These limitations greatly hinder the fine control of energy efficiency in intelligent workshops.
Therefore, the primary arm of this research is to investigate how an energy consumption architecture im-plement the device-level energy consumption sensing and workshop-level energy consumption modeling in the SMT production lines.This paper presents a more detailed enhancement of energy sensing and modeling method.The rest of the paper is organized as follows: In second section, the overall architecture of energy consumption sensing and modelling are described.The third section introduces the detailed implements of the device -level real-time energy consumption sensing technology.The fourth section presents the workshop-level energy consumption modeling in a general way and the fifth section summaries the paper and gives some consideration about the future work.

The overall architecture of the real-time energy consumption system
According to energy consumption has some important fields of application and it is widely discussed.This section describes an overall architecture of the realtime energy consumption system in a general way and argues for its characteristic when applied in SMT production lines.

The SMT production lines
The SMT production lines have the characteristics of high production efficiency and strong flexibility.And the SMT orders have the characteristics of small batch, high frequency, and large total amount.Therefore, the SMT production lines often show cluster characteristics.For example, AKM Electronic Technology Group, one of the leading enterprises in the SMT industry, has formed a cluster of nearly 40 production lines.This cluster has a high degree of customization and the non-standard process is about 30, involving more than 100 sets of heterogeneous equipment, which is a typical equipment-intensive production line cluster.This study takes the SMT production line cluster as an example.A detailed single SMT production line is shown in Fig. 1.From the beginning, the flexible printed circuit (FPC) is transported by the AGV to the feeding machine and combines with the fixtures.The solder paste is printed in the printing machine and the checked by the solder paste inspection machine (SPI).After the inspection of the thickness & area and other quality parameters, the FPC is delivered to the chip mounter to weld with the NX and some other electronic devices.Then, it is reflowed soldering process.The automated optical inspection (AOI) is introduced to inspect the pose position and the welding quality.Most of the production lines are U-shape, where AGV transmission is carried out to the glue dispenser to package the NX pin.And then, the ultraviolet curing machine (UV) and vacuum baking machine is introduced to accelerate the solidification process.The Xray inspection station is carried out to indicate the air bubbles, pin deformation, etc.Finally, the manually check station is used to ensure the outlooking and conductivity.Among them, the failed piece of each inspection will be picked out and marked as NG, and then repaired by workers.The cluster process is similar to the above process.

The overall architecture of the real-time energy consumption system in SMT production lines
The overall architecture is divided into two categories.The former is the device-level energy consumption sensing technology with the complex working condition machine whose power is unknown, such as robotic or AGVs.The power is unknown under multi working states.The energy consumption sensing technology based on mutual inductance is discussed, and the data will be transmitted to the data center over RS485.The second type is the equipment with certain power.The working state can be divided into several states such as shutdown, preheating, and working.The power in different states is marked in the nameplate.A multi-granularity production line energy consumption calculation paradigm (3-matrix method) is put forward.The paradigm is shown in Fig. 2.

Device-level energy consumption real-time sensing technology based on mutual inductance
Section 3 demonstrates the detailed implementation of the energy consumption real-time sensing technology based on mutual inductance in the device-level for the multi working condition machines whose power is unknown.

Robot energy flow analysis
In order to understand the efficiency of a robot when performing a series of standard operations, the energy flow is analyzed as shown in Figs. 3 and 4.
The energy consumption PA is generated by the robot for each command, part of which is the mechanical energy consumption Pout and the other part Ploss is the lost in the friction.The total energy consumption is Pall: .all out loss The efficiency of each stage in the palletizing op-eration is shown below: where: i part E is energy consumption, for performing a certain phase of the palletizing operation; out E is the total output energy consumption, J of the palletizing operation.
The energy consumption of each stage depends on the power of this stage.Since the current change in real time, the power will also change in real time with the current.So where: ET the total energy consumption of actual loop, J; ts the starting time of the actual loop, s; te the ending time of the actual loop, s.

Real-time energy consumption sensing
Firstly, the standard loop and the actual loop of the robot are built up.
In a standard loop, the real-time energy consumption ES can be divided into the operation energy consumption and the segment energy consumption: In actual loop, the real-time energy consumption ET can be divided into the operation energy consumption and the segment energy consumption also: where:    In this research, the operation energy consumption is Ework.The segment energy consumption Epart includes the startup energy consumption Estart, the preheating energy consumption Epreheat and the waiting energy con-sumption Ewait, the auxiliary energy consumption is Ea, and the abnormal energy consumption EF is not considered in the actual loop.So: In the palletizing operation, and the energy consumption in the startup phase and the preheating phase are shown in Table 1.
Table 1 The current in startup and preheating phase The sum of the startup energy consumption and preheating energy consumption is 3723.54J.
The working energy consumption in standard loop is shown in Table 2.And the total is 4642.74J.As shown in Table 3, in the actual working loop, the total is 7452.65J.The auxiliary energy is 1320 J.In this study, a gripper is installed on the AUBO-i5 robot.As shown in Fig. 4.
1) Nominal value of operation energy consumption: ten standard loops are carried out and the average result is calculated.
3) Operation energy efficiency coefficient: where: Δ the operation energy efficiency coefficient; nT in the actual loop, the number of tasks completed by the robot; η the pass rate of the robot task in the actual loop.

Gripper WIP Robot
Fig. 4 The AUBO-i5 According to the nominal value of operation energy consumption, 10 results are shown in Table 4.
The number of tasks completed by the robot T n (100%), and the pass rate η (100%) are taken in, the operation energy efficiency coefficient: 100% 62.95%.

Workshop-level energy consumption modelling
Section 4 devoted the basic aspects of the pro-posed multi-granularity production line energy consumption modeling and summarized the implementation elements of 3-matrix (the attribute attributes, power attributes and cumulative timing attributes).The 3-matrix is used to calculate the SMT equipment, process, production line, order and workshop energy consumption.First of all, an attribute attribution matrix (represented as [a] matrix) is constructed, which is a [total×3] matrix, and has a total-dimensions.The 3-dimensional column vector of the matrix represents the three states of the equipment as the Eq. ( 13).SMTn i starting a − is a Boolean of the SMT production line n device i in the starting phase, "0" indicates not working, and "1" in working.The [p] is a static matrix, which records the power of each equipment in each production line in each state.The [t] matrix represents the time to maintain this state and is the cumulative result.
After that, the standard sensing time t is set as the production takt.Therefore, the [t] is: The 3-matric is: The ""  means dot product, [E] means the energy consumption.
Finally, a multi-granularity production line energy consumption calculation paradigm can be summarized, as shown in Table 5.
A case study is adapted to make it clearly.It is found that SMT line 1, 2, and 5 have roughly same configuration.The [p], [t] and [a] matrix is shown in Tables 6-8.

Index
Formula Description Equipment energy consumption A single device real-time energy consumption, KW Tstandard sensing time; Xthe SMT production line x; jthe jth device; Xnowthe current working state A single device singlestate energy consumption, KW X  [starting, dormant, working]; According the formulas ( 17)-( 28), the equipment, process and production line energy consumption are shown in Tables 9-11.

Conclusion
In this paper, for the intelligent workshop, the energy consumption real-time sensing technology based on mutual inductance & a multi-granularity production line energy consumption modeling, do most of the work implement the energy consumption architecture.In devicelevel an AUBO-i5 is introduced to show the details of realtime sensing technology based on mutual inductance in standard loop and actual loop.Some key parameters, like nominal value of operation energy consumption, the percentage of the auxiliary energy consumption, and the operation energy efficiency coefficient, are obtained.Additionally, a multi-granularity production line energy consumption modeling based on SMT production lines is summarized for the workshop-level.Case studies are presented separately at the end of each section.Results show that the architecture can effectively sense and model the energy consumption of the devices and the workshop, which provides an available method for the fine management and control of energy consumption and energy efficiency.
As the first step in the whole project of the group, this study expects higher levels of workshop control such as the energy efficiency evaluation and the energy conservation optimization in the future study.While during the implementation of this study, it was concluded that: 1) In the workshop / production line, the energy consumption has a significant correlation with the operations.The frequent downtime and excessive waiting can both increase ineffective energy consumption, which can account for more than 30% of the total waste.Only by ensuring continuous and efficient scheduling of the production process can energy efficiency be effectively improved.The Advanced Planning and Scheduling (APS) and some evaluation index systems are needed.
2) The equipment energy consumption shows strong correlation with the control mode.The fundamental reason why the energy efficiency of same device under different control procedures given by different controller often varies greatly is that the existing control logic only aims to get a simple "set function" rather than "efficient, reliable, and low energy consumption function".In order to improve the energy efficiency of the equipment, the energy consumption constraints of control algorithm are essential.
This research has been partly applied in the AKM Co. Ltd. and in the near future, the authors hope it can be applied in some other discrete manufacturing industries or production line clusters.The group want to study the scenario application of blockchain in the energy consumption tracing and the digital twin in the energy saving control.

Fig. 1
Fig. 1 SMT production line and basic process

Fig. 3
Fig. 2 Robot energy flow consumption, J; n the number of operations; tans the starting time, s of a operation n; tane the ending time, s corresponding to tans.
consumption, J; m the number of segments; tams the starting time of segment m, s; tame the ending time, s corresponding to tams.

1
energy consumption, J; j number of auxiliaries; tajs the starting time of a auxiliary j, s; taje the ending time, s corresponding to tajs.
the number of robot operations; tsns the starting time, of the operation n, s; tsne the ending time, s corresponding to tsns; The segment energy consumption Epart includes the startup energy consumption Estart, the preheating energy consumption Epreheat the waiting energy consumption Ewait.While the auxiliary energy consumption EA and the abnormal energy consumption EF are not considered in the standard loop.So: .
=  the segmented energy consumption, J; m number of segments; tsms the starting time, s of a segment m; tsme the ending time, s corresponding to tsms.In this research, the operation energy consumption is Ework.

Table 2
The current in working phase of the standard loop Time, s Current, mA Time, s Current, mA Time, S Current, MA

Table 3
The current in working phase of the actual loop

Table 9
Equipment energy consumption of each state of the SMT-x production line