COMPUTER-IMPLEMENTED METHOD AND SYSTEM FOR AUTOMATICALLY MONITORING AND DETERMINING THE STATUS OF ENTIRE PROCESS SEGMENTS IN A PROCESS UNIT

• US 20170082988A1
• Filed: 12/05/2014
• Published: 03/23/2017
• Est. Priority Date: 12/05/2013
• Status: Active Grant

• Current Assignee: Bayer AG
• Original Assignee: Bayer Technology Services GmbH (Bayer AG)
• Inventors: Ross, Thomas
• Granted Patent: US 10,261,480 B2
• Time in Patent Office: Days
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 19/41885 characterised by modeling, simulation of the manufacturing system
• CPC Class Codes: G05B 2219/24015 Monitoring
• CPC Class Codes: G05B 2219/33037 Learn parameters of network offline, not while controlling system
• CPC Class Codes: G05B 2219/33322 Failure driven learning
• CPC Class Codes: G05B 23/0216 Human interface functionality, e.g. monitoring system providing help to the user in the selection of tests or in its configuration
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
• CPC Class Codes: Y02P 90/80 Management or planning

Abstract

The invention relates to a method and a computer-implemented system for automatically monitoring and determining the status of entire process sections in a process unit in a computer-implemented manner.

Claims

1. A computer-implemented method for monitoring installation parts which are connected by process engineering and comprise one or more process units, having the following steps:

in step a), M process parameters to be monitored and their measurement range limits are input at a user interface and are forwarded to a module for defining the process unit,

in step b), learning data phases consisting of potential learning vectors are imported to a database module,

in step c), in the database module, N learning vectors from the learning phases from b) are alternatively automatically selected via the user interface using the measurement range limits from step a),

in step d), the learning vectors from step c) are transmitted from the database module to a model module connected to the latter;

in the model module, a model based on a neural network is generated by assigning each learning vector to a best matching unit defined by its M process parameters and adapting the best matching units to the learning vectors in a self-organizing neural map,

in step e), the following features of the model are automatically calculated;

for each of the N learning vectors, a deviation error as the distance between each process parameter value of a learning vector and the corresponding process parameter value of the best matching unit assigned to it,

for each of the N learning vectors, a total deviation error as the sum of the deviation errors of its M process parameters,

in step f), the total deviation error for each learning vector from step e) is transmitted to a module for analysing the deviation errors and a minimum and a maximum value of the calculated N total deviation errors are automatically determined there,

in step g), in order to monitor the process units, the M process parameters monitored online are automatically transmitted, as the monitoring vector of the process unit over time, to the database module at a time stamp t followed by the transmission of the monitoring vector to the model module,

in step h), each monitoring vector from step g) is automatically assigned to the best matching units of the neural map (SOM) from step c) in the model module by comparing the process parameter values of the monitoring vector with the process parameter values of each neuron in the neural map and making a selection on the basis of the shortest distance,

in step i), the deviation errors and the total deviation errors of the monitoring vector are calculated,

in step j), the total deviation error of the monitoring vectors is transmitted to the module for analysing the deviation errors, is assigned in comparison with the minimum and maximum values of the calculated N total deviation errors from f), and the assignment is displayed via the user interface.

2. The computer-implemented method according to claim 1, wherein the range of values determined by the minimum and maximum values in step f) is subdivided into 10 to 100 equidistant segments, the learning vectors are sorted into the segments using their total deviation errors, and at least one threshold is automatically set using the following steps, threshold=first segment supporting point, for which ≧

5 to 15% of the highest total deviation errors are within the segments starting from the maximum value, and the threshold is in the range of the uppermost 7/10 of the segments starting from the maximum value.

3. The computer-implemented method according to claim 2, wherein the calculated threshold is multiplied by a tolerance range in the form of an integer positive factor of usually 2-4.

4. A computer program or software for carrying out the computer-implemented method according to claim 1.

5. Software for carrying out the computer-implemented method according to one of claim 1.

7. The computer system according to claim 4, wherein the module for analysing the deviation errors can divide the determined range into segments, can sort the learning and monitoring vectors into the range segments and can automatically determine a threshold on the basis of the distribution of the learning vectors in the segments.

6. A computer system for monitoring installation parts which are connected by process engineering and comprise one or more process units, the computer system comprising the following modules:

a) a user interface for defining i) M process parameters to be monitored and ii) input of the measurement range limits of the M process parameters of one or more process units to be monitored,

b) a module for defining the process unit for storing the inputs i) and ii), connected to the user interface,

c) a database module for importing and storing learning data phases, learning vectors and monitoring vectors via a data interface, wherein the database module is connected to the module for defining the process unit and to the user interface,

d) a model module which is based on a neural network and is intended to train a model based on a neural network in a fully automatic manner in a self-organizing neural map by automatically calculating;

a deviation error as the distance between each process parameter value of a learning vector and the corresponding process parameter value of the best matching unit assigned to it,

for each of the N learning vectors, a total deviation error as the sum of the deviation errors of its M process parameters,

and to assign a monitoring vector to one of the best matching units of the self-organizing neural map and calculate;

a deviation error as the distance between each process parameter value of the monitoring vector and the corresponding process parameter value of the best matching unit assigned to it,

for each of the monitoring vectors, a total deviation error as the sum of the deviation errors of its M process parameters,

e) a module for analysing the deviation errors by automatically determining a range of the calculated total deviation errors, sorting the learning and monitoring vectors using their total deviation errors in the range, wherein the module for analysing the deviation errors is connected to the model module, and

wherein the module for analysing the deviation errors is connected to the user interface for displaying the sorting of the monitoring vectors in the range of the calculated total deviation errors.