WO2017163153A1

WO2017163153A1 - System and method for predictive condition monitoring and controlling of machines

Info

Publication number: WO2017163153A1
Application number: PCT/IB2017/051525
Authority: WO
Inventors: Praveen JAMBHOLKAR
Original assignee: Jambholkar Praveen
Priority date: 2016-03-21
Filing date: 2017-03-16
Publication date: 2017-09-28

Abstract

The present invention relates to predicting the possibility of failure of a machine over definite time intervals. It employs estimation and prediction models in combination with various sensors, signal processing, micro- controller or Digital Signal Processing (DSP), in order to identify the possibility of failure of a machine, before the occurrence of failure and to rectify the system autonomously as and when the system fails to perform its desired functions.

Description

SYSTEM AND METHOD FOR PREDICTIVE CONDITION MONITORING AND

CONTROLLING OF MACHINES

TECHNICAL FIELD

The present invention relates to a technique for predicting the possibility of failure of a machine and rectifying damages in the machine as it occurs. Specifically, the invention pertains to artificial intelligence based prediction of failures in the machine and preemptive control action taken to avoid the same.

BACKGROUND

Various prediction-based techniques for identifying damages in machines are available. One of the known approaches are Data-driven approaches that use information from observed data to recognize the properties of the progress of damage and accordingly predict the future state without using any specific physical model. Instead, mathematical models and/or weight parameters are employed, which are determined based on training data that are obtained under various usage conditions. Since the data- driven approaches usually depend on trend of data, which often show a unique property near conditions which may lead to a fault conditions, they are powerful in predicting near-future behaviors.

The data-driven approaches are generally classified into two categories: the artificial intelligence based approaches that include Neural Networks (NN) and fuzzy logic and the statistical approaches that include the Gaussian process (GP) regression, relevance/support vector machine, least squares regression, the gamma process, the Wiener processes, hidden Markov model and many more.

The most common Neural Network model amongst all the existing neural network paradigms is the multilayer perceptron (MLP) trained by back propagation (BP) algorithm. Back propagation is known as a supervised network because it requires a desired output in order to learn. However, back propagation algorithm requires a huge amount of time as it requires to iterate many number of iterations and the algorithm only requires a single pass for training, which is a limitation when compared with the class of Fuzzy Neural Networks.

In certain other conventional algorithms like Na^'ive Bayes algorithm, k-means clustering etc., rule extraction is needed in some algorithms and in some algorithms, clustering is performed which does not exactly classify the boundary points where as this algorithm gives accurate and non-linear classification.

This algorithm supports both clustering and classification problem. Hence, it is known as hybrid learning. This learning includes both supervised and unsupervised methods.

Another point of concern arises when the systems are installed in geographically distributed regions and in large numbers. At such places, the preventive maintenance or reactive maintenance is quite costly and causes abnormal delays in rectifying the damages.

SUMMARY

In order to alleviate at least the aforementioned drawbacks of the known art, the present invention pertains to predicting the possibility of failure in a machine over a definite time interval. The invention involves prediction models to identify the possibility of failure of a machine, before the occurrence of failure and takes preemptive control action or advises on corrective measures to rectify the system autonomously even before the system fails.

According to an embodiment of the invention, a system for real-time monitoring and correcting failure of a machine is disclosed. The system includes an input unit for receiving one or more input parameters of a machine, a control unit connected to the input unit for performing identification of one or more fault conditions in the machine based on the input parameters, wherein the control unit is configured to define corrective action for the identified fault condition. According to an embodiment of the invention, the system is a hyperbox based neuro- fuzzy system.

According to a further embodiment, the control unit is configured to perform identification of fault conditions with digital signal processing.

According to another embodiment of the invention, the system is operable on a plurality of machines.

According to yet another embodiment, the input parameters in a Variable Frequency Drive (VFD) are selected from at least Vdc, Idc, Irms, Vrms, frequency, over temperature, fan on/off, ambient temperature, ground noise, time running, heat generated.

According to still another embodiment, the control unit identifies one or more fault conditions are selected from frequent jamming, fan off, heat sink improper, open circuit, dirty panel, SMPS fault, IGBT burnt.

According to another embodiment of the present invention, the control unit selects corrective action from sending technician to the site, changing drive board, changing SMPS, changing fan, cleaning panel.

According to an embodiment of the present invention, the system further comprises hybrid learning mechanism which includes training performed dynamically by the system and unsupervised.

According to an embodiment of the present invention, further comprises cognitive analytical mechanism including time series based data.

According to another embodiment of the invention, a method for real-time monitoring and correcting failure of a machine is disclosed. The method comprises receiving one or more input parameters, identifying one or more fault conditions of the machine based on the input parameters, and defining corrective action for the identified fault condition.

BRIEF DESCRIPTION OF FIGURES The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and components.

Figure 1 illustrates a block diagram of the overall system architecture, according to an embodiment of the present disclosure.

Figure 2 illustrates generic block diagram of artificial Neuro-fuzzy model at individual machine according to the embodiment of the present disclosure.

Figure 3 illustrates machine side AI Neuro fuzzy model-I to identify various conditions according to an embodiment of the present disclosure.

Figure 4 illustrates machine side AI Neuro fuzzy model-II to identify various conditions according to an embodiment of the present disclosure.

Figure 5 illustrates a flow diagram for predictive monitoring for Variable Frequency Drive on machine side according to an embodiment of the present disclosure.

Figure 6 Illustrates Z- transform based on Taylor's series, according to an embodiment oi present disclosure.

Figure 7 Illustrates a basic Cognitive analytical model used to identify the fault on c according to an embodiment of the present disclosure.

Figure 8 Illustrates the Neuro-fuzzy network model on cloud for identification of shadow and

MPPT problems according to an embodiment of the present disclosure.

Figure 9 Illustrates the fuzzy neural network model on cloud for identification of Dust problem according to an embodiment of the present disclosure.

Figure 10 Illustrates the fuzzy neural network model on cloud for identification of p degradation Problem according to an embodiment of the present disclosure.

Figure 11 Illustrates a flow diagram of remote monitoring and auditing of machines on server according to an embodiment of the present disclosure. Figure 12 Illustrates a power graph on sunny day according to an embodiment of the prs disclosure.

Figure 13 Illustrates a power graph for morning shade on sunny day according to an embodir of the present disclosure.

Figure 14 Illustrates a power graph for Improper MPPT where voltage at maximum peak p( condition is constant on sunny day according to an embodiment of the pre disclosure.

DETAILED DESCRIPTION

Exemplary embodiments will now be described with reference to the accompanying drawings. The disclosure may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey its scope to those skilled in the art. The terminology used in the detailed description of the particular exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting. In the drawings, like numbers refer to like elements.

The specification may refer to "an", "one" or "some" embodiment(s) in several locations. This does not necessarily imply that each such reference is to the same embodiment(s), or that the feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments.

As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless expressly stated otherwise. It will be further understood that the terms "includes", "comprises", "including" and/or "comprising" when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may be present. Furthermore, "connected" or "coupled" as used herein may include operatively connected or coupled. As used herein, the term "and/or" includes any and all combinations and arrangements of one or more of the associated listed items.

Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

The figures depict a simplified structure only showing some elements and functional entities, all being logical units whose implementation may differ from what is shown. The connections shown are logical connections; the actual physical connections may be different. It is apparent to a person skilled in the art that the structure may also comprise other functions and structures.

The present invention disclosure aims to alleviate at least the aforementioned drawbacks by providing a capability of predicting the possibility of failure of a machine over a definite time interval. It employs estimation and prediction models to identify the possibility of failure of a machine, even before the occurrence of failure and to rectify the system autonomously as and when the system fails. Besides the prediction models, it comprises a combination of current and voltage sensors, signal processing, microcontroller or Digital Signal Processing (DSP).

The present invention particularly discloses a system for real-time monitoring and correcting failure of a machine, comprising an input unit for receiving one or more input parameters of a machine, a control unit connected to the input unit for performing identification of one or more fault conditions in the machine based on the input parameters, wherein the control unit is configured to define corrective action for the identified fault condition.

An Artificial Intelligence (AI) Engine is provided in the form of a two- tier structure capable of unsupervised learning on new field conditions to identify various modes of failure. In the first tier, i.e. at the remote machine end, a preventive action is attempted in real time before any damage is caused to a machine.

In the second tier, i.e. at the server side which functions as a cloud, another Artificial Intelligence self- learning engine is configured to collect all exception conditions (event driven) from multiple remote machines, identifying root causes and the corrective measures to be performed. This analytics takes advantage of the knowledge of events over a time period for the same machine, as well as events occurring over a plurality of machines at the same time. This knowledge may not be available with an individual machine.

Figure 1 illustrates the overall system structure according to an embodiment of the present disclosure. Inputs are captured using Analog to digital conversion, Digital Signal Processing from a machine. Signal processing may be performed to elicit specific data such as amplitude, frequency, time combination. According to the type of signals, their repetitions and persistence, different filtering techniques are considered. If they are steady state signals, stochastic filtering is used and if they are repetitive signals, Fast Fourier Transforms (FFT) are utilized. If the signals are transient, then Wavelet Transform (WT) can be employed.

As illustrated in Figure 1 , the artificial intelligence processing identifies the present conditions, combinations and accordingly takes real time preventive/corrective action at individual machine level from 'n' remote machines referred as 101 . The relevant data is communicated to the server 103 for analytics, thus requiring minimal bandwidth. On the cloud server, analytics are performed to identify the reason for fault condition and corrective actions are recommended.

After processing, the input parameters are passed as inputs to one of the Neuro-fuzzy classifier model 102 which is a variant of an Adaptive Resonance Theory Neural Network. If any fault is identified, then notify the fault and solution to the end user.

Figure 2 illustrates a fuzzy neural network model at individual machine level according to an embodiment of the present disclosure. This model describes the status of the machine whether it is safe or unsafe condition. At individual machine, real time control is performed using Artificial Intelligence.

The inputs X I, X2, Xn referred to as 201, are fed into multiple hyperboxes 202 and based on output from the hyperboxes 202, the condition 203 is determined. An event driven mechanism ensures that all real time values collected from different machines installed in different geographical regions need not be passed on to the cloud and thus, a low bandwidth on internet connection is sufficient. Analytics is also performed at server end using the Neuro-fuzzy model.

The inputs 201 for this model is considered to be input to the Neuro-fuzzy model at a particular machine where due to the specific input values, it goes to an unsafe condition (events) 203 and the minimum and maximum values of an event hyperbox. Event hyperbox 202 is the hyperbox due to which an unsafe condition is identified.

Figure 3 illustrates an exemplary situation of Predictive condition monitoring of Variable Frequency Drive (VFD) for solar water pumping. Each machine (VFD) has current and voltage sensors, GPRS modem to send input values 301 onto the cloud and estimation and predictive models to predict the failure over a definite horizon of time. The parameters are Vdc, Idc, Irms, Vrms, frequency, over temperature, fan on/off, ambient temperature, ground noise, time running, heat generated. The model at machine level is as shown in this figure.

According to an exemplary embodiment, there are four levels namely low, moderate, high and very high. There are conditions that cause danger to the motor such as jamming, dry run, open circuit and short circuit which need immediate action. Hence, special hyperboxes 302 are used to specify those conditions 303.

As illustrated in Figure 4, the analytics at machine side with the input and event hyperbox data 401 for a Variable Frequency Drive (VFD) are depicted. It shows the appropriate condition of the machine, whether it is safe or unsafe condition and input values 402 are then sent to the cloud and saved and analytics is performed based upon that data. It describes the analytics performed on the machine data and displays the corrective measures 403 that need to be taken according to the resultant fault condition.

Figure 5 illustrates a flow diagram for predictive monitoring for Variable Frequency Drive (VFD). The fault conditions or reasons for fault may be jamming frequently, fan off, heat sink improper, open circuit, dirty panel, SMPS fault, IGBT burnt etc. according to the reason the corrective action should be taken like send technician to the site, change drive board, change SMPS, change fan, clean panel, etc. Future failure possibility is predicted at individual machine level and over multiple machines over a definite horizon of time. According to an embodiment, the raw parameters are given as inputs, subsequently signal processing is performed, and then those values are fed to the neural network inputs and status of the machine is identified.

Figure 6 illustrates z- transform based on Taylor's series according to an embodiment of the present disclosure. This method is capable of predicting the next few points of an unknown continuous function. If a new input arrives which is previously not defined by the system, it is trained automatically by keeping track of previous classes. Prediction is also performed by using the Z transform by considering last five input values based on time.

Figure 7 illustrates a cognitive analytical model used to identify the fault on cloud according to the present disclosure. The values of power (kWh) are given as inputs 701 to the cognitive analytical model. The hyperbox 702 is a condition on which minimum and maximum values are set according to the fault condition. The solution is displayed to the user through the classes 703. In this model the power inputs are given to the cognitive analytical model and according to the input given it will notify the condition to the user based on the severity.

Figure 8 illustrates fuzzy neural network model on cloud for identification of shadow and Improper MPPT problems. Basically power values for every half an hour on a day is given as input 801 to cognitive analytical model. For example if a machine is working for 12 hours a day like 6:00 AM to 6:00 PM the input values contains (power at 6:00 AM (P _{tl )}), power at 6:30 AM (P (_t2))... power at 6:00 PM (P ₍tk₎)). After calculation in hyperboxes 802, the problem is identified through classes 803 and notify to the end user whether there is a shade or improper MPPT based on given inputs in 801.

Figure 9 illustrates fuzzy neural network model on cloud for identification of accumulation of dust on panels. Basically every day average power values in a month are captured .This values are given as inputs 901 to cognitive analytical model. For example the input values contains (power on 1^st day of the month (P _(ti ₎), power on 2^nd day of the month (P _(t2)) power on last day of the month (p

_(tk₎)). After calculations done in hyperboxes 902 notify to the end user through 903 whether there is a dust accumulated on the panels based on given inputs in 901.

Figure 10 illustrates fuzzy neural network model on cloud for identification of panel degradation. Basically weekly average power values in a quarter year are captured. These values are given as input 1001 to cognitive analytical model. For example the input values contains (power on 1 ^st week of the month (P _(ti₎), power on 2^nd week of the month (p _(t2)) power on last week of the quarter i.e., the last week of the third month (p _(tk₎)). After calculations done in hyperboxes 1002, notify to the end user through 1003 whether there is panel degradation based on given inputs in 1001.

As illustrated in Figure 11 , a flow diagram of remote monitoring and auditing of machines on cloud. First train the Neuro-fuzzy model based on previous data, now the model is ready by adjusting its minimum and maximum values while training. The dimensions of a hyperbox are set according to the fault condition. If any fault is predicted then notify to the user else we can conclude that the machine is in good condition.

We can identify the fault occurrence from the collected information. Some of the examples are illustrated in Figure 12, figure 13 and figure 14. In the Figure 12, we can see the complete power graph without any fault from morning to evening on a sunny day. If we consider a power graph in Figure 13, we can identify one of the side is curved at the time of 9:00 AM to 1 1 :00 AM hence we can say that it is occurred due to the presence of shade in the morning. Hence wc define it as morning shadow. If we consider a power graph in Figure 14 we can identify the curve at the top is a straight line where the power is constant.

According to an embodiment of the invention, this is a hybrid learning technique. For instance, training data includes panel degradation, dust on panels, Improper MPPT, and Morning shade fault conditions. If a new data regarding evening shade is given as input then it classifies as a new class and training is performed dynamically by the system.

The Artificial Intelligence (AI) engine is succeeded by a prediction engine based on z- transform, performing non-linear prediction of parameters over a finite horizon of time. Finally the z- transform is evaluated by the AI engine for predicting the safe or unsafe operating conditions.

BENEFITS

This methodology is primarily used in operation and maintenance (predictive pre-emptive) of a large number of distributed equipment. This mechanism can be applied to hybrid learning based decision making in a Management information System (MIS). The technique can be used to predict outcome of corrective medication or procedures in healthcare systems, and in production environment for predicting issues in manufacturing or assembly industries.

It will be apparent to those having ordinary skill in the art that various modifications and variations may be made to the embodiments disclosed herein, consistent with the present disclosure, without departing from the spirit and scope of the present disclosure. Other embodiments consistent with the present disclosure will become apparent from consideration of the specification and the practice of the description disclosed herein.

Claims

Claims:

1. A system for real-time monitoring and correcting failure of a machine, comprising: an input unit for receiving one or more input parameters of a machine; a control unit connected to the input unit for performing identification of one or more fault conditions in the machine based on the input parameters; wherein the control unit is configured to define corrective action for the identified fault condition.

2. The system as claimed in claim 1 , wherein the system is a hyperbox based neuro- fuzzy system.

3. The system as claimed in claim 1 , wherein the control unit is configured to perform identification of fault conditions with digital signal processing.

4. The system as claimed in claim 1 , wherein the system is operable on a plurality of machines.

5. The system as claimed in claim 1 , wherein the input parameters in a Variable Frequency Drive (VFD) are selected from at least Vdc, Idc, Inns, Vrms, frequency, over temperature, fan on/off, ambient temperature, ground noise, time running, heat generated.

6. The system as claimed in claims 1 and 4, wherein the control unit identifies one or more fault conditions are selected from frequent jamming, fan off, heat sink improper, open circuit, dirty panel, SMPS fault, IGBT burnt.

7. The system as claimed in claims 1 and 4, wherein the control unit selects corrective action from sending technician to the site, changing drive board, changing SMPS, changing fan, cleaning panel.

8. The system as claimed in claim 1 , further comprising hybrid learning mechanism which includes training performed dynamically by the system and unsupervised.

9. The system as claimed in claim 1 , further comprises cognitive analytical mechanism including time series based data.

10. A method for real-time monitoring and correcting failure of a machine, comprising the steps of: receiving one or more input parameters; identifying one or more fault conditions of the machine based on the input parameters; and defining corrective action for the identified fault condition.