System for maintaining and analyzing manufacturing equipment and method thereof

• US 8,306,928 B2
• Filed: 01/27/2009
• Issued: 11/06/2012
• Est. Priority Date: 12/16/2008
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Chu, Chun-I, Chen, Ching-Shun, Chou, Tze-Chin
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Bharadwaj, Kalpana
• Application Number:
  US12/360,234
• Publication Number:
  US 20100152878A1
• Time in Patent Office: 1,379 Days
• Field of Search: 706/14
• US Class Current: 706/14
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G05B 23/0243 model based detection method, e.g. first-principles knowledge model
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A system and a method are used for maintaining and analyzing manufacturing equipment. The system includes an embedded forecast device (EFD) configured in the manufacturing equipment and a server in communication with the EFD. The EFD is built in with a feature extraction (FE) algorithm and a forecast model (FM), so as to obtain a real time data of the manufacturing equipment and carry out forecasts on the manufacturing equipment to generate a forecast result. The server has various types of FE algorithms and modeling algorithms, which are selected and combined by a user for training and building a required FM and setting related parameters. The embedded forecast device also has a novelty detection model (NDM), which is capable of informing the server to retrain the models when the manufacturing equipment fails and then to update the FM and the NDM of the embedded forecast device.

Claims

1. A system for maintaining and analyzing a manufacturing equipment, adapted to manage and maintain a forecast model of the manufacturing equipment, the system comprising:

an embedded forecast device, directly configured in the manufacturing equipment, the embedded forecast device comprising the forecast model and a feature extraction (FE) algorithm corresponding to the forecast model therein for obtaining a real time data of the manufacturing equipment and generating a forecast result, and embedded forecast device further comprising a novelty detection model corresponding to the forecast model for generating a model retraining notification message when the manufacturing equipment fails; and

a server, comprising;

a communication interface, for communicating with the embedded forecast device;

a plurality of different FE algorithms and a plurality of different modeling algorithms;

an operating interface, for selecting one of the FE algorithms and one of the modeling algorithms and for combining the selected FE algorithm and the selected modeling algorithm, so as to train and build the forecast model and the novelty detection model executed by the embedded forecast device according to the selected FE algorithm and the selected modeling algorithm; and

a parameter setting table, for recording setting parameters related to the training and building of the forecast model and the novelty detection model,

wherein the server retrains the forecast model and the novelty detection model according to the model retraining notification message, and updates the forecast model and the novelty detection model of the embedded forecast device according to the retrained forecast model and novelty detection model.

2. The system for maintaining and analyzing a manufacturing equipment according to claim 1, wherein the embedded forecast device comprises an embedded microprocessor, a communication module, a data storage unit, and an embedded operating system.

3. The system for maintaining and analyzing a manufacturing equipment according to claim 1, wherein the FE algorithm is one selected from a group consisting of kernel principal component analysis (KPCA), principal component analysis (PCA), and wavelet analysis algorithm, or any combination thereof.

4. The system for maintaining and analyzing a manufacturing equipment according to claim 1, wherein the modeling algorithm is one selected from a group consisting of support vector machine (SVM), support vector regression model (SVR Model), support vector data description (SVDD), radial basis function (RBF), and back-propagation neural network (BPN) algorithm, or any combination thereof.

5. The system for maintaining and analyzing a manufacturing equipment according to claim 4, wherein the novelty detection model is built according to the SVDD algorithm.

6. The system for maintaining and analyzing a manufacturing equipment according to claim 1, wherein a content of the parameter setting table comprises:

an identifier (ID) of the embedded forecast device, a communication address of the embedded forecast device, a data source of the manufacturing equipment, the adopted FE algorithm, parameter settings of the FE algorithm, an adopted training modeling algorithm, parameter settings of the training modeling algorithm, a parameter “

whether an automatic retraining mechanism is performed or not”

(Auto re-Train), a parameter “

whether a retraining is needed or not”

(Need re-Train), the forecast model, the novelty detection model, support vectors of the novelty detection model, and time for building the forecast model.

7. The system for maintaining and analyzing a manufacturing equipment according to claim 6, wherein the communication address of the embedded forecast device is an Internet protocol (IP) address.

8. A method for maintaining and analyzing a manufacturing equipment, adapted to manage and maintain models of the manufacturing equipments and automatically update the models, the method comprising:

providing an embedded forecast device, wherein the embedded forecast device is pre-loaded with a feature extraction (FE) algorithm, a novelty detection model, and a forecast model, and wherein the embedded forecast device is mounted in the manufacturing equipment for obtaining real time data of the manufacturing equipment, and then the FE algorithm is used to extract a feature data from the obtained real time data, so as to generate a forecast result through the forecast model;

providing a server, wherein the server is pre-loaded with various FE algorithms and various modeling algorithms, and an operating interface of the server is provided for a user to select one of the FE algorithms and one of the modeling algorithms, and to combine the selected FE algorithm and the selected modeling algorithm for training and building the forecast model and novelty detection model according to the selected FE algorithm and the selected modeling algorithm and for setting related parameters to be stored in a parameter table;

performing a novelty detection on the real time data of the manufacturing equipment, and sending a model retraining notification message when a new feature data is found;

retraining the models, wherein after receiving the model retraining notification message, the server retrains the forecast model and the novelty detection model according to the new feature data and support vectors of an old model; and

updating the forecast model and the novelty detection model, wherein the server uploads the retrained forecast model and novelty detection model to the embedded forecast device of the manufacturing equipment for updating.

9. The method for maintaining and analyzing a manufacturing equipment according to claim 8, wherein the FE algorithm is one selected from a group consisting of kernel principal component analysis (KPCA), principal component analysis (PCA), and wavelet analysis algorithm, or any combination thereof.

10. The method for maintaining and analyzing a manufacturing equipment according to claim 8, wherein the modeling algorithm is one selected from a group consisting of support vector machine (SVM), support vector regression model (SVR Model), support vector data description (SVDD), radial basis function (RBF), and back-propagation neural network (BPN) algorithm, or any combination thereof.

11. The method for maintaining and analyzing a manufacturing equipment according to claim 10, wherein the novelty detection model is built according to the SVDD algorithm.

12. The method for maintaining and analyzing a manufacturing equipment according to claim 8, further comprising setting a parameter “

whether an automatic retraining mechanism is performed or not”

(Auto re-Train) in the parameter table, wherein once receiving the model retraining notification message, the server first checks the Auto re-Train parameter, and retrains the forecast model and the novelty detection model when the Auto re-Train parameter is set as “

True”

.

13. The method for maintaining and analyzing a manufacturing equipment according to claim 8, further comprising setting a parameter “

whether a retraining is need or not”

(Need re-Train) in the parameter table, wherein when a new feature data is found, the novelty detection model sets the Need re-Train parameter as “

True”

, and sends the model retraining notification message.

14. The method for maintaining and analyzing a manufacturing equipment according to claim 13, further comprising setting the Need re-Train parameter as “

False”

after the server uploads the retrained forecast model and novelty detection model to the embedded forecast device for updating.