US 11,989,010 B2
System and method for detecting and measuring anomalies in signaling originating from components used in industrial processes
Alison Michan, Lausanne (CH)
Assigned to BÜHLER AG, Uzwil (CH)
Appl. No. 17/425,942
Filed by Bühler AG, Uzwil (CH)
PCT Filed Jan. 30, 2020, PCT No. PCT/EP2020/052330
§ 371(c)(1), (2) Date Jul. 26, 2021,
PCT Pub. No. WO2020/157220, PCT Pub. Date Aug. 6, 2020.
Claims priority of application No. 19154618 (EP), filed on Jan. 30, 2019.
Prior Publication US 2022/0163947 A1, May 26, 2022
Int. Cl. G05B 19/418 (2006.01); G06N 20/00 (2019.01)
CPC G05B 19/4184 (2013.01) [G05B 19/4185 (2013.01); G05B 19/41865 (2013.01); G05B 19/4188 (2013.01); G06N 20/00 (2019.01)] 11 Claims
OG exemplary drawing
 
1. A method for detecting anomalies or early indications of equipment failure in industrial equipment or production plants by monitoring measuring data and/or process parameters originating from components used in an industrial process, the method comprising:
measuring and/or monitoring the measuring data by monitoring the process parameters of components used in the industrial process by measuring devices or sensors and identifying equal sized time frames in the measuring and/or process parameters for time frames where the components used in the industrial process are functioning normally, the measuring and/or process parameters comprising parameter values for a plurality of measuring/sensory parameters and/or process variables;
converting, using circuitry, the parameter values of the plurality of measuring/sensory parameters and/or process variables into observable binary processing codes for each of the identified, equal-sized time frames and assigning the binary processing codes to a sequence of storable Markov chain states;
generating, using the circuitry, a multi-dimensional data structure comprising a definable number of variable hidden Markov model parameter values, wherein the variable model parameters of the multi-dimensional data structure are determined by a machine-learning module implemented in the circuitry applied to the sequence of the storable Markov chain states with assigned binary processing codes, and wherein the variable hidden Markov model parameters of the multi-dimensional data structure are varied and trained by learning a normal state frequency of occurring alarm events based on the measuring data and/or the process parameters of the identified, equal-sized time frames;
initializing and storing, using the circuitry, a plurality of probability state values by applying the trained multi-dimensional data structure with the variable hidden Markov model parameter values to presampled binary processing codes having a same equal-sized time frame as the parameter values of the plurality of measuring/sensory parameters and/or process variables;
determining, using the circuitry, a logarithmic threshold value of an anomaly score by ordering logarithmic result values of the stored probability state values; and
deploying, using the circuitry, said trained multi-dimensional data structure with the variable hidden Markov model parameter values to monitor newly measured respectively determined the measuring data and/or the process parameters from industrial equipment or plants using the threshold value of the anomaly score to detect anomalous sensor data values that could be indicative of an impending system failure, wherein, for triggering at the anomalous sensor data values, a logarithmic result value of the probability state value of the newly measured respectively determined measuring data and/or process parameters is generated and compared to the stored probability state values based on said logarithmic threshold value of the anomaly score.