(39 of 1970) & The Patents Rules 2003

COMPLETE SPECIFICATION

(SECTION 10 and Rule 13)

1. Title of the invention:

AN APPARATUS AND METHOD TO ESTIMATE CYCLE TIME FOR PROCESSING A WORKPIECE IN AN INDUSTRY

2. Applicants: a. Name: Robert Bosch Engineering and Business Solutions Private Limited

Nationality: INDIA

Address: 123, Industrial Layout, Hosur Road, Koramangala, Bangalore - 560095, Karnataka, India b. Name: Robert Bosch GmbH Nationality: GERMANY

Address: Feuerbach, Stuttgart, Germany

Complete Specification:

The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed. Field of the invention:

[0001] The present invention relates to an apparatus and method to estimate cycle time for processing a workpiece by a machine in a manufacturing industry.

Background of the invention:

[0002] A digital information is not readily available centrally in Small Medium sized Enterprises (SMEs). The two main concerns of SMEs are cost of ownership and simplification. The manual documentation of production count by operators may often lead to human data entry errors. Also, due to absence of automatic alerts, sometimes when a machine breaks down, it is not known to the industry manager/operator unless they visit the factory floor/machine. Due to idle machinery or faulty machines consuming energy in an erratic manner, energy cost is an issue that is less addressed in older machines. There exists high cost of sensor-based solution in market. The machine utilization insights that come from such complex Programmable Logic Controller (PLC) backed controls equipment require high capital costs, complex programming and masses of wiring. A time lag in report availability is faced more often. Other products in the market analyze machine data on cloud only. Hence, the reports are not real time. In many, sensor installations can be difficult to standardize, difficult to install, and are subject to degradation.

[0003] According to apriorartUS20180299944, a production management method and system using power consumption features. An electric meter is connected to an equipment machine for measuring power consumption data of the equipment machine. The power consumption data during a processing cycle of the equipment machine is used as a power consumption sample. The power consumption sample is uploaded to a cloud server and a plurality of feature points is set. The power consumption data of the equipment machine is uploaded to the cloud server in real time and compared with the power consumption sample through feature matching, to obtain a facility utilization rate, a production efficiency and a product yield of the equipment machine, so as to calculate an overall equipment effectiveness. Brief description of the accompanying drawings:

[0004] An embodiment of the disclosure is described with reference to the following accompanying drawings,

[0005] Fig. 1 illustrates a block diagram of an apparatus to determine cycle time of processing a workpiece by a machine in a manufacturing industry, according to an embodiment of the present invention, and

[0006] Fig. 2 illustrates the method for determining cycle time of processing a workpiece by the machine in the manufacturing industry, according to the present invention.

Detailed description of the embodiments:

[0007] Fig. 1 illustrates a block diagram of an apparatus to determine cycle time of processing a workpiece by a machine in a manufacturing industry, according to an embodiment of the present invention. The apparatus 120 is shown as a part of a system 100. The system 100 comprises a sensing unit 104 clamped to power cables of a work machine 102 in a non-intrusive manner. The sensing unit 104 measures and collects electrical power consumption data (or power signature) of the machine 102 during a working cycle. The working cycle corresponds to a time shift or time zone during which the machine 102 is operated (or is to be operated) to process a workpiece as part of manufacturing a product. The apparatus 120 comprises a controller (not shown), which is configured to, receive power consumption signals, as measured by the sensing unit 104, and generates an apparent power signal. The apparatus 120 is characterized by, the controller configured to filter the apparent power signal and identify prominent repetitive patterns. The controller further extracts active peaks from the apparent power signal having values higher than a production threshold (explained later). The controller then transforms the extracted peaks using wavelet scattering and determines the cycle time based on events identified in the transformed signal using the wavelet scattering The edge device 110 functions on principles of edge computing near the source of data and thus increases the response time. [0008] The controller is equipped with necessary signal detection and acquisition circuits in connection with the sensing unit 104. The controller comprises memory element such as Random Access Memory (RAM) and/or Read Only Memory (ROM), Analog-to-Digital Converter (ADC) and vice-versa Digital-to-Analog Convertor (DAC), clocks, timers and at least one processor (capable of implementing machine learning) connected with the each other and to other components through communication bus channels. The memory element is pre stored with logics or instructions or programs or applications or modules/models and/or threshold values, which is/are accessed by the at least one processor as per the defined routines. The internal components of the controller are not explained for being state of the art, and the same must not be understood in a limiting manner. The controller may also comprise communication units to communicate with the cloud server 118 through wireless or wired means such as Global System for Mobile Communications (GSM), 3G, 4G, 5G, Wi-Fi, Bluetooth, Ethernet, serial networks and the like.

[0009] In accordance to the present invention, the controller processes the signals received by the sensing unit 104 through few modules. A data processing module 106 is provided to process the measured power consumption data and generate the apparent power signal. The data processing module 106 further processes the apparent power signal through a filter and generate a filtered signal for the identification of prominent repetitive patterns. A cycle time estimator 108 is another module which executes the extraction of active peaks, applies wavelet scattering and determines cycle time based on the identified events. The working of the cycle time estimator 108 is explained later.

[0010] The filter used is Median Absolute Deviation with sliding window process and not limited to the same. The filter is a Median Absolute Deviation with Sliding Window Technique (MovMAD or Mov-Mad). The filter reduces noise and identifies prominent repetitive patterns. The filter is customized with automatic threshold by segmentation. The customized filter is generic enough to automatically identify threshold for fdtering and preserves abrupt level changes in power signal which might represent production, by attenuating ‘normal’ observational noise, excludes outliers, and keeps track of monotonic trends in signal within a reduced computation time. The fdter overcomes the difficulties in known methods such as Moving Averages, Running Medians, and other linear filters, as these known methods though track trends and attenuate Gaussian noise efficiently, but are highly vulnerable to outliers and blur level shifts. The Running Medians remove outliers and preserve shifts in a piecewise constant signal but have shortcomings in trend periods. The filter used on the present invention is more efficient for Gaussian noise, removes spikes and preserve steps even better than running medians.

[0011] The production threshold is calculated using K-means clustering, which is pre-determined during training process. The production threshold identifies the actual production activity of the machine 102. According to the present invention, the production threshold aids in identification of the machine 102 operation/activity. The higher amplitude power values are most probably related to the active production. Depending on the type and number of operations involved in the production, the duration and the power fluctuation varies. Thus, identifying a single production can be simplified to identifying any one of the operations. The spikes caused by the operations are identified visually as peaks in the power signal. However, spike could be due to other reasons as well. The active peaks, or the peaks that represent probable production, are identified from the filtered signal. The peaks identified in the filtered signal are clustered into two groups, to separate higher and lower amplitude power samples. The least value of the cluster with highest mean is selected as threshold for identifying active peaks. production threshold = mm clusters_higher_power_samples )

[0012] In accordance to an embodiment of the present invention, the apparatus 120 is at least one selected from a group comprising an edge device 110 and a cloud server 118. In an embodiment, the apparatus 120 is the edge device 110. In another embodiment, the apparatus 120 is the cloud server 118 or a remote server. In yet another embodiment, the apparatus 120 is combination of the edge device 110 and the cloud server 118.

[0013] In an embodiment of the present invention, the apparatus 120 as the cloud server 118 is explained. The cloud server 118 is in communication with the sensing unit 104 through the data processing module 106. The data processing module 106 transmits the fdtered signal to the cloud server 118. The cloud server 118 comprises a database 112 and a cycle time estimator 108 similar to that present in the controller. The cycle time estimator 108 executes the extraction of active peaks, applies wavelet scattering and determines cycle time based on the identified events. The cycle time is then accessible to the interested party such as by the owner/operator.

[0014] In accordance to an embodiment of the present invention, the operation or working of the cycle time estimator 108 is explained. The cycle time estimator 108 selects events from the transformed signal using Principal Component Analysis (PCA). The cycle time estimator 108 then clusters the events with similar characteristics using K-Means clustering or other clustering techniques. Then a distance between two consecutive events within each of the clusters is calculated. Finally, the cycle time is determined/estimated based on comparison of the distances with empirically predetermined ranges and/or rules.

[0015] The working of the cycle time estimator 108 is explained in more detail in stages. In a first stage, events are identified, and feature extraction is done. The active peaks that have higher power values than the production threshold are extracted from the power data consumption signal. The extracted active peaks or the events representing a production are transformed using wavelet scattering. The scattering representations construct invariant, stable, and informative signal representations by cascading wavelet modulus decompositions followed by lowpass filter. The wavelet time scattering representations are insensitive to translations in the input signal without sacrificing class discriminability. The scattering transform builds a signal representation that is invariant to transformations while preserving as much signal information as possible. It is defined as a convolutional network whose filters are fixed to be wavelet and lowpass averaging filters coupled with modulus nonlinearities. In a second stage, feature selection is performed. The important events are selected using Principal Component Analysis (PC A) with two components. The analysis done on over predetermined samples (for example 40 number of machine’s data or samples) showed that more than 95% features is explained by the first two components. In a third stage, clustering is performed. The pool of events selected using PCA may contain events that do not represent production cycle. Similar events amongst the selected events are grouped together using K-Means clustering. Optimal number of groups, ‘K’, is calculated with Gap Statistic automatically. In a fourth stage, cycle time is determined. The cycle time is the time required to produce a single unit. Amongst the clusters generated from the previous stage, the distance between two consecutive events within a cluster represent probable cycle times. The highly likely cycle time of the product is chosen by binning the probable cycle times into ranges and rules generated by analyzing over few samples (for example 80 number of machine’s data or samples) of product data.

[0016] In accordance to an embodiment of the present invention, the system 100 is shown with a second machine 114 and an n ^th machine 116 with respective apparatuses 120. All the apparatuses 120 either work independently or are in communication with the cloud server 118 and operates in the similar manner as that for the machine 102. Further, the data processed by the apparatus 120 is not only the filtered signal, but also the actual raw signals measured by the sensing unit 104.

[0017] Fig. 2 illustrates the method for determining cycle time of processing a workpiece by the machine in the manufacturing industry, according to the present invention. The method comprises plurality of steps, of which a step 202 comprises measuring power consumption data of the machine 102, which processes the workpiece (a single unit), and calculating the apparent power signal. The power consumption data (or power signature) of the machine 102 is collected during the working cycle using the sensing unit 104 clamped to the power cables of the machine 102. The method is characterized by, a step 204 which comprises filtering the apparent power signal for identification of prominent repetitive patterns in the filtered signal. A step 206 comprises extracting active peaks from the apparent power signal having values higher than the production threshold. A step 208 comprises transforming the extracted peaks using wavelet scattering. A step 210 comprises determining the cycle time by processing events in transformed signal obtained from wavelet scattering. In the step 210, processing events further comprises multiple steps, of which a step 212 comprises selecting events from the transformed signal using Principal Component Analysis (PCA). A step 214 comprises clustering the events with similar characteristics using K-Means clustering or other clustering algorithms. A step 216 comprises calculating distance between two consecutive events within each of the clusters. A step 218 comprises determining the cycle time based on comparison of the distances with predetermined ranges and/or rules. The method described above must be read in combination with the Fig. 1 and corresponding description.

[0018] The filtering is performed by a Median Absolute Deviation with sliding window process. The production threshold is calculated using K-means clustering or other clustering techniques or algorithms. The method is executed by at least one apparatus 120 selected from a group comprising the edge device 110 and the cloud server 118. The edge device 110 connected/interfaced between the sensing unit 104 and the cloud server 118.

[0019] According to the present invention, an ensemble based approach for non- intrusive machine and production monitoring in manufacturing industries is provided. Specifically, the present invention provides analytics solutions for improving OEE in manufacturing industries through an Intemet-of-Things (IoT) and Artificial Intelligence (AI) based asset and energy monitoring system. The data processing module 106 further configured to calculate and generate an apparent power signal from the measured consumption data (active power and the reactive power). The approach used for this invention is termed as Ensemble -based Machine Learning and Signal Detection Techniques (EMLSD) for Non-intrusive Machine and Production Monitoring in Manufacturing Industries. In the present invention, the result in terms of OEE is generated using a tree-based approach.

[0020] It should be understood that embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and changes in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.