Development of an Advanced Cloud for Industry based on AVM Technology

Min-Hsiung Hung*1, Senior Member, IEEE, Yu-Chuan Lin2, Student Member, IEEE, Hsien-Cheng Huang2, Min-Hsuan Hsieh2, Haw-Ching Yang3, Member, IEEE, and Fan-Tien Cheng2, Fellow, IEEE

 Abstract—In this paper, a cloud manufacturing platform, provide dynamically scalable and virtualized resources as called AMC (Advanced Manufacturing Cloud), is introduced. services over the in a pay-by-use manner, the The AMC provides several manufacturing-related cloud manufacturing enterprises adopting cloud-computing services to facilitate the users to conduct supporting activities technologies and services can not only save the expensive for machine tools and offers intelligent devices which can be costs of creating and maintaining information hardware by plugged in various prediction models for performing predictive themselves, but also create new business models to applications on machine tools. We first describe the architecture design of the AMC and then present methodologies to how to effectively increase their business benefits. build cloud services and key mechanisms in the AMC. Finally, A great number of cloud-computing-related papers have we construct a paradigm AMC and practically deploy its cloud been published. However, most of these papers focus on services and intelligent devices in a public cloud platform and a developing cloud computing technologies or addressing factory of our cooperative machine tool company, respectively. issues in cloud computing, such as [3] [4] [5] . Although We also conduct integrated tests to illustrate the efficacy and several papers of cloud computing can be found to relate to merits of the AMC. manufacturing, such as [6] , their contents are less related to I. INTRODUCTION practical implementation of a cloud computing system for manufacturing. An equipment monitoring system based on With the advancement of manufacturing technologies, cloud computing is presented in [7] , but only the aspects of machine tools are becoming more sophisticated. However, system architecture and GUI design were addressed. because machine tools are often used in long-term operations, In this paper, we leverage several technologies (including some of their components may become aged or broken, which Automatic Virtual Metrology (AVM) [8] [9] , Ontology, will reduce the quality of processed products or workpieces. Virtual Machine Tool (VMT), and Cloud Computing) to Therefore, creating diagnostics and prognostics capabilities, develop a cloud manufacturing platform, called AMC such as fault detection, manufacturing precision conjecture, (Advanced Manufacturing Cloud), for the machine tool and remaining useful life prediction, for machine tools to industry. We design the AMC’s cloud services to offer ensure their reliability and production quality has become an several manufacturing-related functions applicable to important topic for the industry. By using these intelligent machine tools, including collecting data from machine tools, capabilities, we can know certain failures of machine tools in creating and managing prediction models for machine tools, advance and then take actions to reduce the chance of their recommending machine tools and cutting tools for machining occurrence. This can increase the machine tools’ availability tasks, together with conducting virtual machine tool and result in a significant saving of maintenance cost and time machining evaluations and simulations before real machining. of machine tools. Thus, adding such kinds of intelligence to Also, we design a computer-based virtual machine which can machine tools fits the trend towards smarter machines and be plugged in various prediction models for performing manufacturing systems [1] and is a key factor for the machine intelligent applications (such as precision prediction of tool companies to increase their competiveness. workpieces) on machine tools. Finally, we show the practical With the rapid development of information and network deployment of the AMC in a public cloud and a machine tool technology, cloud computing [2] has become an emerging manufacturing factory, as well as the integrated testing results, trend of Internet applications. Because cloud computing can to validate the effectiveness and performance of the AMC.

This work was supported by the National Science Council of Republic of II. ARCHITECTURE DESIGN OF ADVANCED China under Contract No: NSC 101-2221-E-034-023, NSC MANUFACTURING CLOUD 101-2218-E-006-022, and NSC 102-2622-E-006-022-CC2. This work was The architecture of the proposed AMC is designed as also financially supported by the Ministry of Education of ROC with Project shown in Fig. 1, which consists of three parts: the cloud side, AIM-HI. 1M.-H. Hung is with the Department of Computer Science and the factory side, and the client side. Information Engineering, Chinese Culture University, Taipei, Taiwan, R.O.C. 2.1 The Cloud Side (*Dr. Hung is the corresponding author; phone: +886-2-28610511 ext. 33511; The cloud side of the AMC contains several e-mail: [email protected]). 2Y.-C. Lin, H.-C. Huang, M.-H. Hsieh, and F.-T. Cheng are with the manufacturing-related functional components, including Institute of Manufacturing Information and Systems, National Cheng Kung Data Acquisition, Model Creation, Model Management, University, Tainan, Taiwan, R.O.C. Ontology Inference, Virtual Machine Tool. Furthermore, 3H.-C. Yang is with the Institute of System Information and Control, these functional components are wrapped by Web Services to National Kaohsiung First University of Science and Technology, Kaohsiung, form the AMC cloud services so that other systems or users Taiwan, R.O.C.

can easily access these cloud services through the Internet. pre-processing module that can process and filter the raw data The cloud side also holds the databases (such as Data of sensors installed on machine tools, generating the Acquisition DB, Central DB, and Ontology DB) and storage corresponding indicators. Then, these indicator data can be (such as Model Repository) needed by the AMC cloud uploaded to the v-Supplier for creating prediction models or services. In addition, a cloud Web server is needed on the fed to the pluggable algorithm module (PAM) for on-line cloud side to host a variety of Web GUIs for the users to computing virtual metrology (VM) values in various operate the AMC cloud services. Since the cloud side real-time prediction applications, such as conjecturing the provides services, storage spaces, and GUIs to the users process quality, detecting equipment faults, estimating the through the Internet, it can serve as a virtual provider remaining useful life of equipment, and so on. The GDAD (v-Supplier). can collect data from various types of equipment. The Local DB is used to store the indicator data and the VM results. On the factory side, a Service Broker is designed to act as Central DB Ontology DB the communication agent and security guard between the

Model AMC cloud services and v-Machines. All communications Repository between the cloud services and v-Machines need to pass Data Acquisition Cloud Web DB Server through the Service Broker. The Service Broker is used to manage v-Machines as well. A local Web server is also needed to host a variety of Web GUIs for the local users to operate the v-Machines. 2.3 The Client Side The client side can be anywhere having Internet access. GCI GCI

VMK VMK The Web GUIs [7] of the AMC are built using Silverlight,

PAM PAM GDAD GDAD GCI: Generic Communication Interface one of popular RIA (Rich Internet Application) technologies. VMK: Virtual Machine Kernel GDAD: Generic Data Acquisition Driver PAM: Pluggable Algorithm Module Once the user utilizes a browser to download the Web GUIs, Precisions Precisions Local DB: Local Database Sensors Sensors Variables Variables the GUIs can run as a standalone application in the user’s browser, needing no Web server’s resources. According to Product Product Process Metrology Process Metrology purposes, the GUIs of the AMC are divided into the following Machine A Equipment A Machine B Equipment B categories [7] : Data Acquisition, Model Creation, Model Fig. 1. Architecture of the advanced Manufacturing cloud. Management, Historical Data Search, Service Management, The Data Acquisition cloud service, Model Creation and User Management. cloud service, and Model Management cloud service are the core of Automatic Virtual Metrology. They are responsible III. DESIGN OF KEY CLOUD SERVICES AND MECHANISMS for collecting machine tools’ data from factories, creating 3.1 Design of Data Acquisition Service prediction models, and managing the prediction models, The Data Acquisition service is designed to collect respectively. The Ontology Inference cloud service is built process data of machine tools and metrology data of using Ontology technology and the Protégé tool. It can draw workpieces from factories and store them in cloud databases. inferences from machine-tool knowledge in Ontology DB These collected process and metrology data will subsequently and recommend proper machine tools and cutting tools for be used by the Model Creation service to create prediction machining tasks. The Virtual Machine Tool cloud service can models for machine tools. In this study, process data refers to provide functions like preliminary evaluation, VMT assembly, the indicators’ values of the raw data of sensors installed on NC post-processor, virtual machining, cutting simulations, machine tools, and metrology data refers to the measurement and so on. values of some precision items of machined workpieces. 2.2 The Factory Side To flexibly specify which data items should be collected On the factory side, we design a computer-based virtual from which factory, we design data collection plans (DCPs) machine, called v-Machine, which can monitor various kinds using XML. Fig. 2 shows a sample data collection plan in of equipment. The v-Machine consists of five parts: Generic XML. The schema of the DCP XML file is designed as Communication Interface (GCI), Pluggable Algorithm follows. “DCP” is the root element and has an attribute “ID” Module (PAM), Virtual Machine Kernel (VMK), Generic with a unique value as the identity of the data collection plan. Data Acquisition Driver (GDAD), and Local Database (Local “DCP” has three child elements: ”DCProperty,” ”Factory,” DB). and “WorkPiece.” The GCI are built using WCF (Windows Communication The “DCProperty” element is to express the properties Foundation) and can allow the v-Machine to communicate of the data collection and contains the following four with other systems via multiple protocols (including SOAP attributes. (1) “CollectionMode” indicates the modes of data and REST). The PAM can host various prediction models for collection, and its value is “Immediate” in this study, meaning on-line monitoring the connected machine tools or predicting that once the data collection is completed, return the data machining precision of workpieces. The VMK is responsible collection report immediately. Other modes (such as for handling commands and messages, as well as setting the “regular”) of data collection can be defined as well. (2) configuration of v-Machine. The VMK also contains a data “ConjectureType” indicates the types of prediction models to

be created, and its value can only be “MQM,” “PPP,” or process data of the CNC machine tool “CNC_001” of type “KDP” in this study, standing for machine-tool quality “QP2040-L” and metrology data of the workpiece maintenance, product-precision prediction, and “CellphoneBackPlane_1” of type “CellphoneBackPlane.” key-component diagnosis and prognosis, respectively. (3) The plan needs to collect three metrology data items. The first “StartTime” and “EndTime” indicate the start time and the one is “Parallelism_1” of type “Parallelism,” which relates to end time of the data collection, respectively, with intent to machining actions 1 and 3. The second one is collect data records whose timestamp is within the start time “Straightness_1” of type “Straightness,” which relates to and the end time. machining action 1. The third one is “Straitghtness_2” of type “Straightness,” which relates to machining action 3. Two process data items need to be collected. They are indicators “Average” and “Crest_Factor,” both of which are from “Vibration” sensor installed on the first spindle. Once the v-Machine “vMachine_001” completes the data collection, it immediately returns the data collection report in XML to the Data Acquisition service, which will then save these collected data in cloud databases. 3.2 Design of Model Creation Service The Model Creation service is designed for creating various prediction models using intelligent algorithms. It totally contains the following eleven algorithmic modules (the first seven modules for data preprocessing and the rest of modules for model creation): (1) Data Transfer: This module is to pair metrology data and process data for creating models. It also transfers data into the

Fig. 2. A sample data collection plan in XML. format used by the following modules. (2) MDFR (Metrology Data Filter Rule): This module provides a mechanism to The “Factory” element is to express the information of allow the user to filter out abnormal metrology data. (3) factory and machine tools. Each factory may have several DQIy Pattern: This module utilizes ART2 (Adaptive Service Brokers, each Service Broker may manage several Resonance Theory 2) technology, an unsupervised neural v-Machines, and each v-Machine can connect with 1~4 CNC network, to generate clusters of process data for getting rid of machine tools. Thus, we express the elements “Factory,” abnormal associated metrology data. (4) KSS (Keep Sample “ServiceBroker,” “v-Machine,” and “CNC” in a hierarchical Scheme): This module implements a scheme to keep relationship and let each of these elements occur proper times important data samples. (5) KVS (Key Variable Selection): according to the factory’s practical deployment configuration. This module provides a method for selecting key variables Also, every element has an attribute “Name” to indicate its from a high-dimension variable set. (6) DQIX (Process Data name. Lastly, “CNC” has another attribute “Type” to indicate Quality Index): This module is to compute the quality index the type of the CNC machine. of process data. If the computed DQIX is less than the The “Workpiece” element is to express the information of threshold, the process data is abnormal and should be workpiece, process data items, and metrology data items. excluded. (7) DQIy (Metrology Data Quality Index): This “Workpiece” has two attributes “Name” and “Type” to module is to compute the quality index of metrology data. If indicate the name and type of the workpiece whose data will the computed DQIy is less than the threshold, the metrology be collected in this plan. “Workpiece” also has two child data is abnormal and should be excluded. (8) BPNN elements “Y_Data” and “X_Data” to include the metrology (Back-Propagation Neural Network): This module is to create data items and the process data items to be collected, prediction models using back-propagation neural network. (9) respectively. Each metrology data item is expressed as a child MR (Multiple Regression): This module is to create reference element “Y_Item” of “Y_Data.” Each “Y_Item” has three models using multiple regression methods. (10) RI (Reliance attributes “Name,” “Type,” and “Block” to indicate the name, Index): This module is to compute the reliance index of the type, and related machining actions of a metrology data item, prediction results of a prediction model. The value of RI respectively. Similarly, each process data item is expressed as ranges from 0 to 1, and having a higher RI means that the a child element “X_Item” of “X_Data.” Each “X_Item” has prediction results have a higher reliance. (11) GSI (Global three attributes “Name,” “SensorType,” and “Position” to Similarity Index): This module is to compute the GSI value. If indicate the name, sensor type, and sensor position of a the GSI of an input set of process data is greater than the process data item, respectively. threshold of the GSI, it indicates that some process data may According the above-mentioned XML schema design, the have deviated. For detailed development and usage of these sample data collection plan in Fig. 3 is intended to be algorithmic modules, please refer to [8] [9] . downloaded from the Data Acquisition service to the A complete model-creation procedure of the Model v-Machine “vMachine_001” in the “Chevalier” factory Creation service consists of sequentially executing these through the Service Broker “Chevalier_SB” to collect eleven cloud functions one by one, the Data Transfer function

being executed first and the RI function being executed last. from the Queue in a first-come-first-get manner, and a Furthermore, the intelligent algorithms in these cloud Worker VM executes only one command at a time. By this functions are built using MATLAB, and MATLAB does not design, many users can be served simultaneously. The Table support multi-core or multi-threading programming in is used to store the statuses of model-creation operations for general. Consequently, if multiple users simultaneously users. The Blob is used to store temporary model files access the Model Creation service, then only one user at a generated by model-creation functions, as well as the files of time can execute this service, and the rest of users need to final created prediction models. wait until the service is available, leading to unsatisfying user Fig. 4 illustrates the operational flow of creating experiences. To tackle this problem, we design the prediction models. The GUI sends a Model-Creation architecture of the Model Creation service as shown in Fig. 3, command to the MCS Kernel, which then dispatches the which can effectively support multiple users’ simultaneous command to the Queue. Next, an available Worker VM gets access. the command and subsequently downloads the required temporary model files from the Blob. Afterward, the Worker VM executes the command and then uploads its generated temporary model files to the Blob, as well as save the computational status in the Table. The MCS Kernel checks the computational status in the Table and returns the MC results corresponding to this MC command to the GUI.

Fig. 3. The architecture of the Model Creation service for supporting multiple users’ simultaneous access. The architecture is composed of four layers: GUI, MCS Fig. 4. Operational flow of creating prediction models. Kernel, Storage, and Worker VM. The GUI layer contains 3.4 Design of Model Management Service various Web GUIs based on Silverlight [7] . The user can use The Model Management service is designed for a browser to download the Web GUIs from the cloud Web managing prediction models created in the cloud. This cloud server, and then is able to directly interact with the Model service provides the following three functions: Creation service via these RIA-based Web GUIs running on (1) Storing and recording models: This function can save the the browser. Notably, unlike traditional Web technology, prediction models created by the Model Creation service which needs to work with Web server’s resources, RIA in the cloud storage and record their metadata, such as technology enables the Web GUIs to run as standalone model size, model creation time, model creator, applicable applications, needing no Web server’s resources. workpieces, etc. The MCS Kernel layer is the front tier of the Model (2) Searching models: This function allows the user to query Creation service, i.e. providing the service’s interfaces. It is historical prediction models in the cloud storage. responsible for receiving various model-creation commands (3) Downloading models: This function enables the user to from the users’ Web GUIs and dispatching them to Queue select a set of prediction models in the cloud storage and (depicted later), as well as responding results to the users’ download them to a target v-Machine in a factory. Web GUIs. As shown in Fig. 4, we can use a virtual machine to host the MCS Kernel for a proper number of users. We are IV. CASE STUDY AND TESTING RESULTS also able to lease multiple instances of the MCS-Kernel 4.1 Experimental Setup virtual machine on demand to handle a tremendous users. Based on the designs in Sections II and III, we have The Worker VM layer consists of multiple instances of constructed a prototype of advanced manufacturing cloud for Worker VM (Virtual Machine), each hosting the eleven the machine tool industry. We deploy the cloud services and model-creation functions. We can alter the number of functions, together with the cloud database and storage, in Worker-VM instances on demand according to the number of Windows Azure [7] , Microsoft’s public cloud platform, in users. Hong Kong. Specifically, the manufacturing cloud services The Storage layer hosts three types of cloud storage: and the cloud Web server are deployed in Web Roles of Small Queue, Table, and Blob. The Queue possesses the class. The model-creation functions with the MCR First-In-Fist-Out characteristic. Thus, the MCS Kernel can (MATLAB Compiler Runtime) are deployed in Worker dispatch various model-creation commands from many users Roles of Extra Small class. The cloud databases are deployed to the Queue in order. Then, the available Worker VMs will in SQL Azure, and the model repository is deployed in competitively retrieve these queued commands one by one Azure’s Blob storage.

On the factory side, we deploy a v-Machine, a Service Scenario B: Conducting the PPP service on the QP2040-L Broker, and a local Web server in a factory of our cooperative CNC machine tool using a v-Machine equipped company, FALCON Machine Tools Corp., in Changhua, with prediction models: Taiwan. The v-Machine is connected with a three-axes CNC B.0: The user logins the GUI of v-Machine. machine tool of type QP2040-L. The workpieces include B.1: The user installs the workpiece in the CNC tool and ISO10791 standard testing workpieces and cellphone downloads the G Code to the controller before starting backplanes. Fig. 5 shows the photos of the QP2040-L CNC machining. machine tool, a standard testing workpiece, and a cellphone B.2: The user starts machining the workpiece. backplane. The client side can be anywhere having Internet B.3: The v-Machine on-line collects and preprocesses the access. The user can download the RIA-based Web GUIs signal data of sensors installed on the CNC tool, from the cloud Web server and begin to operate the AMC. generating a sample of process data. B.4: The v-Machine computes the virtual metrology values (representing the predicted product precision) of the machined workpiece by feeding the sample of process data into the prediction models in the v-Machine. B.5: The v-Machine stores the virtual metrology values in its (b) internal database and sends them to the GUI. B.6: The GUI displays the virtual metrology values in data tables and graphic charts for the user to check the product precision of the machined workpiece. 4.3 Testing Results and Performance Evaluation (c) We have conducted integrated tests on the AMC following the testing scenarios planned in Section 4.2. (a) Fig. 5. Photos of (a) the QP2040-L CNC machine tool, (b) a standard testing Testing results show that the AMC works successfully to workpiece, and (c) a cellphone backplane. fully support these testing scenarios in many demo cases. Two Web GUIs of the AMC are shown Fig. 6 and Fig. 7. 4.2 Testing Scenarios In order to verify the effectiveness of the proposed AMC, various operational scenarios are designed to test the system, including (1) recommending proper machine tools for machining via the Ontology Inference cloud service, (2) recommending cutting tools for machining via the Ontology Inference cloud service and the VMT cloud service, (3) collecting the machine tool’s data in the factory via the Data Acquisition cloud service, (4) creating prediction models via the Model Creation cloud service, (5) downloading prediction models to a v-Machine in the factory via the Model Management cloud service, and (6) conducting PPP (Product Precision Prediction) service on the QP2040-L CNC machine tool using a v-Machine. Due to space limitation, the following only depicts two testing scenarios: Scenario A: Recommending cutting tools for machining via Fig. 6. Snapshot of the Model Creation GUI right after finishing model creation. the Ontology Inference (OI) cloud service and the VMT cloud service: A.0: The user downloads Web GUIs from the cloud Web server and logins the AMC to access the OI service. A.1: The OI service sends the cutting-tool information to the GUI for the user to select an alternative cutting tool. A.2: The OI service sends the data of the selected cutting tool to the VMT service to perform a cutting simulation. A.3: The VMT service sends the results of the cutting simulation to the GUI through the OI service. A.4: If any over-cutting occurs, repeat Steps A.1~A.3 for recommending a proper cutting tool having no over-cutting. A.5: The OI service sends the data of the recommended cutting tool to the v-Machine A for machining. Fig. 7. Snapshot of the v-Machine GUI right after finishing the machining of a workpiece.

Fig. 6 shows a snapshot of the Model Creation GUI right V. CONCLUSION after finishing model creation. As shown in the figure, the By leveraging the advantages of cloud computing and model creation involves eight steps in sequence. By checking realizing the merits of cloud manufacturing, we develop an the displayed charts of virtual metrology values together with advanced manufacturing cloud (called AMC) for the machine their RI and GSI values, the user can verify whether the tool industry. In addition to cloud computing, the AMC prediction model is reliable or not. The left hand side of the integrates several intelligent technologies, such as Automatic GUI also shows the values (in um) of MAE (Maximum Virtual Metrology, Ontology, Virtual Machine Tool. The Average Error) and Max Error of the created NN (Neural AMC provides several cloud services to facilitate the users to Network) model and MR (Multiple Regression) model. In conduct supporting activities for machine tools, including this test, the values of MAE and Max Error of both models are remotely collecting data from machine tools, creating small (about 1.5 times those of the real metrology) and within prediction models for machine tools, recommending cutting the industry’s acceptable specifications. tools for machining tasks, together with conducting virtual Fig. 7 shows a snapshot of the v-Machine GUI displaying machine tool machining simulations before real machining. on-line virtual metrology values in data tables and graphic The AMC also offers intelligent devices (called v-Machines) charts right after finishing the machining of a workpiece. The which can be plugged in various prediction models for figure are displaying the virtual metrology values of the performing predictive applications (such as predicting metrology item “MaxDistance.” In this test, when the machining prediction of workpieces) on machine tools. In machining of a workpiece is finished, the virtual metrology this paper, we describe the architecture design of the AMC values of this workpiece can be computed and displayed and present methodologies to how to build cloud services and within 10 seconds, which is rather fast. key mechanisms in the AMC. Finally, we construct a We also test the performance of the AMC in supporting paradigm AMC and practically deploy its cloud services and multiple users’ access to the Model-Creation service. In the v-Machines in the Windows Azure public cloud platform and first experiment, we turn off multi-user support and conduct a factory of our cooperative machine tool company, tests on five cases with 1, 5, 10, 15, and 20 users, respectively. respectively. Integrated testing results show that the AMC In the second experiment, we activate multi-user support and can achieve merits of small prediction errors (about 1.5 times conduct tests on five cases with 1, 5, 10, 15, and 20 users, those of the real metrology), fast prediction (within 10 respectively, each of which sets the users/instances ratio to seconds after the machining of a workpiece is finished), and one, where instances refer to the number of Worker VMs being able to simultaneously support many machine tools and used in the Model Creation service. Fig. 8 shows the effect of many factories. This paper can be a useful reference for activating multi-user support. Without multi-user support in researchers and practitioners who would like to investigate or the Model Creation service, the model-creation times of the construct manufacturing cloud systems. five testing cases in the first experiment are about proportional to the number of users. By contrast, when the REFERENCES multi-user support is activated, the model-creation times of [1] J. Lee, M. Ghaffari, and S. Elmeligy, “Self-Maintenance and the five testing cases in the second experiment are almost the Engineering Immune Systems: Towards Smarter Machines and Manufacturing Systems,” Annual Reviews in Control, vol. 35, pp. same, about 300 seconds. These results validate the efficacy 111-122, 2011. of the designed multi-user support mechanism by paying for [2] P. Mell and T. Grance, “The NIST Definition of Cloud Computing,” vol. more cloud computing resources as the number of users 53, Issue 6, National Institute of Standards and Technology, Oct. 2009. increases, leveraging the advantage of the pay-by-use [3] A. Berl, E. Gelenbe, M. D. Girolamo, G. Giuliani, H. D. Meer, M. Q. Dang and K. Pentikousis, "Energy-Efficient Cloud Computing," The property. Computer Journal, Vol. 53 pp. 1045-1051, 2010. [4] H. Kim, et al, "Transparently Bridging Semantic Gap in CPU Management for Virtualized Environments," J. Parallel Distrib. Comput., vol. 71, pp. 758–773, 2011. [5] S. Subashini and V. Kavitha, “A Survey on Security Issues in Service Delivery Models of Cloud Computing,” Journal of Network and Computer Applications, vol. 34, pp. 1-11, 2011. [6] M. Wang, J. Zhou, and S. Jing, “Cloud Manufacturing: Needs, Concept and Architecture,” the 2012 IEEE International Conference on Computer Supported Cooperative Work in Design, pp. 321-327, 2012. [7] M.-H. Hung, Y.-C. Lin, T. Q. Huy, H.-C. Yang, and F.-T. Cheng, “Development of a Cloud-Computing-based Equipment Monitoring System for Machine Tool Industry,” Proceedings of the 8th annual IEEE Conference on Automation Science and Engineering (CASE 2012), Seoul, Korea, pp. 958-963, August 20-24, 2012. [8] Y.-T. Huang and F.-T. Cheng, “Automatic data quality evaluation for the AVM system,” IEEE Trans. on Semiconductor Manufacturing, vol. 24, no. 3, pp. 445–454, Aug. 2011. [9] M.-H. Hung, C.-F. Chen, H.-C. Huang, H.-C. Yang, and F.-T. Cheng, “Development of an AVM System Implementation Framework,” IEEE Fig. 8. The performance curves of the Model Creation service by activating Trans. on Semi. Manufacturing, vol. 25, no. 4, pp. 598-613, Nov. 2012. multi-users support.