Sensor-independent mechanical machine fault identification

Abstract

A method for identifying a fault of at least one mechanical machine includes causing a first plurality of sensors coupled to a corresponding first plurality of mechanical machines to acquire a first plurality of groups of signals emitted from the first plurality of mechanical machines, the first plurality of mechanical machines sharing at least one characteristic; supplying at least the first plurality of groups of signals of the first plurality of mechanical machines to a pre-existing fault classifier, the pre-existing fault classifier being pre-trained to automatically identify a fault of a second plurality of mechanical machines based on signals emitted from a second plurality of mechanical machines and pre-acquired by the second plurality of sensors, the second plurality of sensors being of a type different from that of the first plurality of sensors, the second plurality of mechanical machines sharing at least one characteristic; modifying the pre-existing fault classifier by using transfer learning based on at least the first plurality of groups of signals of the first plurality of mechanical machines, thereby providing a modified fault classifier; applying the modified fault classifier to at least one additional group of signals acquired by at least one sensor of the first plurality of sensors and emitted from at least one given mechanical machine sharing at least one characteristic, the modified fault classifier being configured to automatically identify at least one fault of the at least one given mechanical machine based on the at least one additional group of signals; and providing a human-perceptible output by an output device, the human-perceptible output comprising at least an identification of a fault of the at least one given mechanical machine, performing at least one of a repair or maintenance operation based on the human-perceptible output.

Description

[0001] Cross-references

[0002] Reference is hereby made to U.S. Provisional Patent Application No. 62 / 895,247, filed on September 3, 2019, entitled “SYSTEMS AND METHODS FOR MULTI-SENSOR MAPPING FOR AUTOMATED MACHINE DIAGNOSTICS,” the disclosure of which is incorporated herein by reference pursuant to 37 CFR 1.78(a)(4) and (5)(i), and the benefit of priority is hereby claimed. Technical Field

[0003] The present invention relates generally to monitoring mechanical machines and, more particularly, to identifying faults in mechanical machines by monitoring the mechanical machines. Background Art

[0004] Various systems and methods are known in the art for fault identification of mechanical machines. Summary of the invention

[0005] The present invention seeks to provide a novel system and method for using transfer learning between different types of sensors monitoring signals emitted from a mechanical machine for the purpose of fault identification and maintenance of the monitored mechanical machine.

[0006] Therefore, according to a preferred embodiment of the present invention, there is provided a method for identifying a fault of at least one mechanical machine, the method comprising: causing a first plurality of sensors coupled to a corresponding first plurality of mechanical machines to acquire a first plurality of groups of signals emitted from the first plurality of mechanical machines, the first plurality of mechanical machines sharing at least one characteristic; supplying at least the first plurality of groups of signals of the first plurality of mechanical machines to a pre-existing fault classifier, the pre-existing fault classifier being pre-trained to automatically identify faults of the second plurality of mechanical machines based on signals emitted from a second plurality of mechanical machines and pre-acquired by the second plurality of sensors, the second plurality of sensors being of a type different from that of the first plurality of sensors, the second plurality of mechanical machines sharing at least one characteristic; modifying the pre-existing fault classifier by using transfer learning based on at least the first plurality of groups of signals of the first plurality of mechanical machines, thereby providing a modified fault classifier; applying the modified fault classifier to at least one additional group of signals acquired by at least one sensor of the first plurality of sensors and emitted from at least one given mechanical machine sharing at least one characteristic, the modified fault classifier being configured to automatically identify at least one fault of the at least one given mechanical machine based on the at least one additional group of signals; and providing a human-perceptible output by an output device, the human-perceptible output comprising at least an identification of a fault of the at least one given mechanical machine, performing at least one of a repair or maintenance operation based on the human-perceptible output.

[0007] According to a preferred embodiment of the present invention, the method also includes: after causing the first plurality of sensors to acquire the first plurality of groups of signals and before supplying the first plurality of groups of signals to a pre-existing fault classifier: obtaining first plurality of groups of operating condition data for mechanical machines among the first plurality of mechanical machines, each group of operating condition data indicating an operating state of a mechanical machine among the first plurality of mechanical machines, each operating state being related to at least one of the plurality of groups of signals; supplying at least the first plurality of groups of signals to a pre-existing fault classifier also includes supplying the operating condition data of the first plurality of mechanical machines to the pre-existing fault classifier; modifying the pre-existing fault classifier by using transfer learning at least based on the first plurality of groups of signals also includes modifying the pre-existing fault classifier by using transfer learning based on the first plurality of groups of operating condition data of the first plurality of mechanical machines.

[0008] Preferably, identifying a fault includes identifying a specific fault of at least one given mechanical machine and predicting a failure of the at least one given mechanical machine due to the specific fault without performing a recommended maintenance operation on the at least one given mechanical machine, wherein the at least one given mechanical machine will indeed fail without performing the recommended maintenance operation.

[0009] Preferably, the pre-existing fault classifier comprises a neural network, and modifying the pre-existing fault classifier comprises adding at least one mapping layer to the neural network, the neural network not being modified in any other way except for adding the at least one mapping layer.

[0010] Preferably, the neural network comprising a pre-existing fault classifier comprises a data layer and an input layer for receiving data from the data layer, and at least one mapping layer is added between the data layer and the input layer, so that the at least one mapping layer is configured to receive data from the data layer in the modified fault classifier.

[0011] Preferably, the first plurality of sensors have a first frequency response distribution, and the second plurality of sensors have a second frequency response distribution, and the mapping layer is configured to map between the first frequency response distribution and the second frequency response distribution.

[0012] According to a preferred embodiment of the method of the invention, the first plurality of sensors are operated to sense signals of the same type as the signals sensed by the second plurality of sensors.

[0013] Preferably, the same type of signal includes one of a vibration signal, a magnetic flux signal, an electric current, a temperature and a machine internal pressure signal.

[0014] According to another preferred embodiment of the present invention, the first plurality of sensors and the second plurality of sensors are operated to sense mutually different types of signals.

[0015] Preferably, the mutually different types of signals include at least one of a vibration and a magnetic flux signal; a vibration and a current signal; a vibration and a temperature signal; a current and a magnetic flux signal; and a vibration and a machine internal pressure signal.

[0016] Preferably, at least some of the operating states of a mechanical machine of the first plurality of mechanical machines as indicated by the first plurality of sets of operating condition data are faulty operating states.

[0017] In accordance with yet another preferred embodiment of the present invention, the first plurality of sets of signals and the first plurality of sets of operating condition data of the first plurality of mechanical machines include less than 30 sets of faulty operating conditions.

[0018] According to another preferred embodiment of the present invention, a system for identifying a fault of at least one mechanical machine is also provided, the system comprising: a first plurality of sensors, the first plurality of sensors being coupled to corresponding first plurality of mechanical machines and being operated to acquire first plurality of groups of signals emitted from the first plurality of mechanical machines, the first plurality of mechanical machines sharing at least one characteristic; a data processing unit, being operated to: receive the first plurality of groups of signals from the first plurality of mechanical machines, the data processing unit comprising a pre-existing fault classifier, the pre-existing fault classifier being pre-trained to automatically classify the operating states of the second plurality of mechanical machines based on signals emitted from the second plurality of mechanical machines and pre-acquired by the second plurality of sensors, the second plurality of sensors being of a type different from that of the first plurality of sensors, the second plurality of mechanical machines sharing at least one characteristic; and the sensors share at least one characteristic; based at least on the first plurality of groups of signals of the first plurality of mechanical machines, modifying a pre-existing fault classifier by using transfer learning to provide a modified fault classifier, and applying the modified fault classifier to at least one additional group of signals acquired by at least one sensor of the first plurality of sensors and emitted from at least one given mechanical machine sharing the at least one characteristic, the modified fault classifier being configured to automatically identify at least one fault of at least one given mechanical machine based on the at least one additional group of signals, and an output device in communication with the data processing unit and operated to provide a human-perceptible output, the human-perceptible output at least including an identification of a fault of at least one given mechanical machine, and performing at least one of a repair or maintenance operation based on the human-perceptible output.

[0019] According to a preferred embodiment of the present invention, the system also includes a data collection unit, which is operated to obtain first multiple groups of operating condition data for mechanical machines among a first plurality of mechanical machines, each group of operating condition data indicates an operating state of a mechanical machine among the first plurality of mechanical machines, and each operating state is related to at least one of the multiple groups of signals; the data processing unit is operated to receive the operating condition data of the first plurality of mechanical machines, and further modify a pre-existing fault classifier based on the first multiple groups of operating condition data of the first plurality of mechanical machines by using transfer learning.

[0020] Preferably, identifying a fault includes identifying a specific fault of at least one given mechanical machine and predicting a failure of the at least one given mechanical machine due to the specific fault without performing a recommended maintenance operation on the at least one given mechanical machine, wherein the at least one given mechanical machine will indeed fail without performing the recommended maintenance operation.

[0021] Preferably, the pre-existing fault classifier comprises a neural network, and modifying the pre-existing fault classifier comprises adding at least one mapping layer to the neural network, the neural network not being modified in any other way except for adding the at least one mapping layer.

[0022] Preferably, the neural network comprising a pre-existing fault classifier comprises a data layer and an input layer for receiving data from the data layer, and at least one mapping layer is added between the data layer and the input layer, so that the at least one mapping layer is configured to receive data from the data layer in the modified fault classifier.

[0023] Preferably, the first plurality of sensors have a first frequency response distribution, and the second plurality of sensors have a second frequency response distribution, and the mapping layer is configured to map between the first frequency response distribution and the second frequency response distribution.

[0024] According to a preferred embodiment of the system of the invention, the first plurality of sensors are operated to sense signals of the same type as the signals sensed by the second plurality of sensors.

[0025] Preferably, the same type of signal includes one of a vibration signal, a magnetic flux signal, an electric current, a temperature and a machine internal pressure signal.

[0026] According to another preferred embodiment of the system according to the invention, the first plurality of sensors and the second plurality of sensors are operated to sense mutually different types of signals.

[0027] Preferably, the mutually different types of signals include at least one of a vibration and a magnetic flux signal; a vibration and a current signal; a vibration and a temperature signal; a current and a magnetic flux signal; and a vibration and a machine internal pressure signal.

[0028] Preferably, at least some of the operating states of a mechanical machine of the first plurality of mechanical machines as indicated by the first plurality of sets of operating condition data are faulty operating states.

[0029] According to yet another preferred embodiment of the system of the present invention, the first plurality of sets of signals and the first plurality of sets of operating condition data of the first plurality of mechanical machines include less than 30 sets of faulty operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be more fully understood and appreciated from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0031] Figure 1 is a simplified high-level block diagram of a system for mechanical machine fault identification constructed and operative in accordance with a preferred embodiment of the present invention;

[0032] Figure 2is a simplified high-level block diagram of a system for mechanical machine fault identification illustrating the construction and operation of another preferred embodiment of the present invention;

[0033] Figure 3 Is it Figure 1 or Figure 2 A simplified diagram of a modification of a fault classifier performed by a system of the type shown in;

[0034] Figure 4A and Figure 4B is shown for training Figure 1 or Figure 2 used in systems and by Figure 1 or Figure 2 A simplified corresponding block diagram of components of a corresponding system of a modified fault classifier of the system;

[0035] FIG. 5A to FIG. 5D It is used to identify the fault of mechanical machine Figure 1 or Figure 2 A graph showing the performance of a system of the type shown in;

[0036] Figure 6 is a simplified flow chart showing the steps involved in a method of mechanical machine fault identification according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0037] Now refer to Figure 1 , Figure 1 is a simplified high-level block diagram of a system for mechanical machine fault identification constructed and operative in accordance with a preferred embodiment of the present invention.

[0038] like Figure 1 As shown in FIG. 1 , a system 100 for mechanical machine fault identification is provided. The system 100 preferably includes a first plurality of sensors 102 of a first type (represented herein as sensor type S1). All sensors 102 may be of the same type (e.g., sensor type S1 as shown herein). Optionally, the first plurality of sensors 102 may include more than one type (e.g., sensor types S1, S11 to S1N, as shown herein). Figure 2) which will be described in more detail hereinafter. The sensors 102 are preferably coupled to corresponding first plurality of mechanical machines 104. Here, as an example, the first plurality of mechanical machines 104 are shown to include mechanical machine 1, mechanical machine 2 to mechanical machine N, where N can be any number of mechanical machines (such as two or more mechanical machines). Typically, the sensors 102 are coupled to the mechanical machines 104 in a one-to-one corresponding arrangement, where one of the sensors 102 is coupled to a corresponding one of the machines 104. However, other arrangements are also possible, where a single sensor can be arranged to sense signals from more than one mechanical machine 104. The sensor 102 can be in physical contact with the machine 104 (such as mounted directly or indirectly on the machine 104). For example, where the sensor 102 is an optical sensor, the sensor 102 can optionally be physically separated from the machine 104 (such as located at a given distance from the machine 104).

[0039] The sensor 102 may be implemented as any type of sensing device operable to sense signals emitted from the mechanical machine 104. Machines and mechanical systems with moving parts (such as machines or motors including bearings, rotors or shafts, engines, compressors, pumps, fans, gearboxes, coolers, etc.) may generate signals during their operation. The mechanical machine 104 may be any of the aforementioned types of mechanical systems, or may be any of other types of mechanical systems that generate signals during their operation. The sensor 102 is preferably operable to sense such signals. Analysis of the sensed signals may be used to determine the condition of the machine that emitted the sensed signals, and in some cases to determine a fault in the machine that emitted the signals. By way of non-limiting example only, the sensors 102 may all be the same type of vibration sensors (such as all single-axis accelerometers or all multi-axis accelerometers) for sensing vibrations emanating from the machine 104; the sensors 102 may all be the same type of flux sensors for sensing magnetic flux emanating from the machine 104; the sensors 102 may all be the same type of current sensors for sensing changes in current generated by the machine 104; the sensors 102 may all be the same type of temperature sensors for sensing heat generated by the machine 104. It should be understood that the sensors 102 may alternatively all be any other type of sensing device of the same type capable of sensing signals emanating from and generated by the machine 104, including signals related to machine operation such as torque, displacement, input line frequency, etc. The sensors 102 may alternatively include two or more types of sensors (such as flux sensors and vibration sensors, by way of example only).

[0040] Mechanical machines 1 to N, which are members of the first plurality of mechanical machines 104, are preferably characterized by one or more shared characteristics. Mechanical machines 1 to N may or may not be identical machines, as long as mechanical machines 1 to N have at least one shared characteristic in common. For example, the shared characteristic may refer to a type, model, manufacturer, physical characteristics or dimensions, operating characteristics or parameters, or other shared characteristics (the other shared characteristics indicate that the observed behavior of one mechanical machine in the plurality of mechanical machines may be typical of another mechanical machine in the plurality of mechanical machines).

[0041] The first plurality of sensors 102 are preferably operable to acquire a first plurality of sets of signals emitted from the first plurality of mechanical machines 104. The plurality of sets of signals may be "signal snapshots" sensed by an appropriate one of the sensors 102 over a short period of time. For example, the signals may be sensed over a period of a few seconds (such as 1 to 4 seconds). Each set of signals may optionally include a plurality of "signal snapshots" over time (e.g., a four-second "signal snapshot" measured every hour over a period of hours, days, or even months). Alternatively, each set of signals may include signals monitored continuously over a longer period of time or over more than one period of time. For example, the signals may be monitored continuously every millisecond.

[0042] The signals acquired by the sensor 102 may be pre-processed, for example, by analog or digital processing capabilities of the sensor 102 itself or by other hardware and / or software signal processing components 108. It should be understood that although Figure 1 The signal processor 108 is shown in FIG as a separate element from the sensor 102, but the signal processing functionality may be incorporated into one or more sensors 102. Signal preprocessing may include at least one of digitization, compression, feature extraction, and representation in the time or frequency domain of the signal.

[0043] In one embodiment of the invention, the plurality of sets of signals acquired by the first plurality of sensors 102 may be uploaded to a remote server (such as a server in the cloud). Signal processing functionality 108 may be performed at the remote server. Preferably, the plurality of sets of signals acquired by the first plurality of sensors 102 are accumulated as they are acquired from the first plurality of machines 104. For example, the plurality of sets of signals may be accumulated at a server in the cloud.

[0044] The system 100 may optionally include a data collection unit 110. The data collection unit 110 may be operable to receive a first plurality of sets of operating condition data regarding a mechanical machine in the plurality of mechanical machines 104, each set of operating condition data indicating an operating state of a mechanical machine in the first plurality of mechanical machines 104, each operating state being associated with at least one of the plurality of sets of signals acquired by the first plurality of sensors 102.

[0045] The operating state data collected at the data collection unit 110 is preferably in the form of a machine state diagnosis supplied by a human expert such as an engineer. The plurality of sets of signals acquired by the first plurality of sensors 102 are optionally provided to the data collection unit 110 by the sensors 102 directly or indirectly and / or by the signal processor 108. The human expert may optionally analyze the plurality of sets of signals acquired by the first plurality of sensors 102 and label each of the plurality of sets of signals as representing a specific operating state of the corresponding mechanical machine 104 from which the signal was emitted. The human expert may interact with a user interface (e.g., with a user interface of the data collection unit 110, or with a user interface of another device capable of communicating between the human expert and the data collection unit 110) to input the operating state data. The signals and labels applied to the data collection unit 110 may be accumulated and stored in a database in the data collection unit 110. In one embodiment of the present invention, the data collection unit 110 may be located in a remote server such as a server in the cloud.

[0046] Identifying a fault by a human expert may include identifying one or more specific faults of the monitored machine 104. Depending on the specific machine 104 being monitored, the identified specific faults may include bearing wear, mechanical looseness, misalignment, imbalance, electrical faults, or other faults of the rotating machine. Identifying a fault may optionally include identifying that the machine 104 is in a fault state (i.e., an abnormal state relative to identifying a normal, healthy operating state of the machine 104), but not identifying a specific fault. In this case, the human expert fault identification identifies the machine as not operating in a healthy manner, but does not identify what the specific cause of the unhealthy operation is.

[0047] Regardless of whether the specific operating conditions represented by the signals are labeled by a human expert, the multiple sets of signals acquired by the sensor 102 may include multiple sets of signals corresponding to normal, non-faulty operating conditions of the machine 104 and multiple sets of signals corresponding to abnormal, faulty operating conditions of the machine 104.

[0048] Optionally, the plurality of groups of signals acquired by the sensor 102 do not necessarily include a plurality of groups of signals corresponding to an unhealthy faulty operating state of the machine 104. According to this embodiment of the present invention, the machines 104 monitored by the sensor 102 may all be in a healthy operating state. Therefore, the plurality of groups of group signals and optionally the associated machine operating states may all correspond to a healthy state of machine operation.

[0049] The accumulated sets of signals from the first plurality of machines 104 are preferably supplied to a pre-existing fault classifier 114 included in the system 100. Where the system 100 further includes a data collection unit 110 and the signals are labeled, the labeled signals are preferably supplied to the pre-existing fault classifier 114 jointly by the signal pre-processor 108 and the data collection unit 110.

[0050] The pre-existing fault classifier 114 may be an algorithmic classifier. For example, the pre-existing fault classifier 114 may be stored as a remote server. The system 100 may include a non-transitory computer-readable storage medium having computer-executable instructions stored thereon for executing, by a processor, the functions of the pre-existing fault classifier 114. The one or more processors executing the pre-existing fault classifier 114 may be remote processors located, for example, in the cloud, or may be local processors.

[0051] The pre-existing fault classifier 114 is preferably a fault classifier that has been pre-trained to automatically identify faults of the second plurality of mechanical machines based on signals emitted from the second plurality of mechanical machines and pre-acquired by the second plurality of sensors, the second plurality of sensors being of a type different from the type of the first plurality of sensors 102. The second plurality of sensors may all be of the same type as one another, the type of the second plurality of sensors may be different from one or more types of sensors in the first plurality of sensors 102. Alternatively, the second plurality of sensors may all be of the same type as one another, the type of the second plurality of sensors may be different from at least one of the types of sensors in the first plurality of sensors 102. The second plurality of sensors may optionally include more than one type of sensors, which may all be of a type different from one or more types of sensors in the first plurality of sensors. The second plurality of sensors may optionally include more than one type of sensors, which may be of a type different from at least one of the types of sensors in the first plurality of sensors 102. The second plurality of mechanical machines preferably share at least one mechanical characteristic shared by the first plurality of mechanical machines 104.

[0052] It should be understood that the fault classifier 114 is referred to herein as “pre-existing” because it may pre-exist with respect to the multiple sets of signals and optionally operating condition data acquired by the sensors 102 from the machine 104. The fault classifier 114 may be pre-generated at an earlier point in time prior to the generation of the data set including the multiple sets of signals and optionally operating condition data of the machine 104.

[0053] Preferably, the fault classifier 114 is an accurate classifier configured to accurately identify faults in mechanical machines that share at least one common characteristic based on signals acquired by a second plurality of sensors. The fault classifier 114 can be such an accurate classifier because it has been pre-trained using machine learning based on a large data set including signals acquired from a large number of machines and possibly related operating condition data. As is well known to those skilled in the art, the greater the amount of data provided to the machine learning fault classifier for the purpose of training the machine learning fault classifier, the more accurately the classifier can perform, up to a given limit. For example, the fault classifier 114 can be trained using data sensed from over 40,000 individual rotating machines, including motors, pumps, fans, coolers, compressors, and gearboxes. The common characteristic shared by such machines may be the inclusion of bearings therein. The following description Figures 3 to 4B An example of how the fault classifier 114 may be pre-trained is shown in FIG.

[0054] Therefore, the pre-existing fault classifier 114 can be successfully applied to signals emitted from mechanical machines having shared characteristics in the mechanical machines based on the data on which the fault classifier 114 is trained, so as to identify faults thereof. However, it should be noted that the fault classifier 114 is pre-trained based on signals acquired by one or more specific types of sensors (i.e., the second plurality of sensors) of a type different from the type of the first plurality of sensors 102. As a result, the fault classifier 114 is able to most successfully classify and identify faults when applied to signals acquired by specific sensors (based on which the fault classifier 114 is trained). However, in the case where the signals supplied to the fault classifier 114 are acquired by sensors of a type different from the type of the second plurality of sensors (based on which the fault classifier 114 is trained), the fault classifier 114 will not be able to accurately classify and identify faults based on these signals. This is due to the difference in sensor characteristics between the sensors (i.e., the second plurality of sensors) on which the fault classifier 114 is trained and the sensors (e.g., the first plurality of sensors S1 or the first plurality of sensors S1 to S1N) that have acquired the current signals that need to be classified.

[0055] Even if the fault classifier 114 is a highly accurate classifier for identifying faults based on signals obtained by the second plurality of sensors, the fault classifier is not very accurate in identifying faults obtained by sensors of different types (e.g., the first plurality of sensors S1 or S1 to S1N ( Figure 2 If the fault classifier 114 is applied to the signals acquired by the first plurality of sensors, the results will be inaccurate.

[0056] This can be illustrated by referring to the case of two types of vibration sensors (such as a three-axis accelerometer and a uniaxial accelerometer), which sense vibration signals generated by a mechanical machine. The two types of vibration sensors differ from each other in various parameters such as geometry, mass, internal materials, etc., resulting in differences in the moments of inertia and resonant frequencies of the respective sensors. As a result, the sensors have mutually different frequency responses. Due to the inherent differences between the two sensors, a specific signal generated by the monitored mechanical machine will be sensed and recorded differently by the two sensors. Moreover, if the two sensors are mounted at different positions on the monitored machine, this difference will be even further exacerbated due to the fact that the sensors measure along mutually different measurement axes and due to the different vibration levels measured due to the positional differences.

[0057] For example, in the case of a triaxial accelerometer and a uniaxial accelerometer mounted on a cylindrical machine, the two types of accelerometers will measure different vibration levels from each other because the triaxial accelerometer will measure radial vibration along one axis and tangential vibration along two axes, while the uniaxial accelerometer will measure radial (or direct) vibration.

[0058] Therefore, the use of a fault detection classifier trained using signals acquired from one type of sensor (e.g., a vibration signal acquired by a tri-axial accelerometer) will lead to incorrect classification results when applied to the same type of signal acquired by a different type of sensor (e.g., a vibration signal acquired by a uniaxial accelerometer).

[0059] This can be further exemplified by reference to the more extreme case of two types of sensors sensing different types of signals, such as a vibration sensor and a flux sensor respectively sensing vibration and flux signals emanating from a particular mechanical machine. A fault detection classifier trained using data acquired by one type of sensor (e.g., vibration signals acquired by a vibration sensor) will be limited to classifying vibration signals and will provide poor results that are almost irrelevant if applied to identifying faults in, for example, flux signals generated by the same machine.

[0060] In order to provide a fault classifier that can accurately classify signals acquired by different types of sensors as signals based on which the fault classifier was previously trained, it is possible to classify signals acquired by different types of sensors (e.g., a first plurality of sensors S1 or S1, S1 to S1N ( Figure 2)) to train a completely new fault classifier. In this case, the pre-existing fault classifier 114 is not utilized, and a new fault classifier is developed to identify faults of a mechanical machine such as the machine 104. However, in order for this new classifier to provide accurate fault identification, it is necessary to provide this new classifier with signals obtained by different types of sensors (e.g., the first plurality of sensors S1 or the first plurality of sensors S1 to S1N ( Figure 2 )) A large amount of new data is acquired, and a fault classifier must be trained based on the large amount of new data. This process may be tedious, and such a large amount of data may not be available. In addition, due to the many parameters that control the frequency dependence of the machine signal (such as sensor mounting position, orientation, mounting type, etc.), this process is also highly limited in terms of performance and scope of applicability. Moreover, in this method, since the previous fault classifier 114 is not applied at all, the capabilities of the original pre-existing fault classifier 114 are simply wasted rather than utilized.

[0061] The present invention advantageously provides a solution to the problem that fault classifiers trained based on data acquired from a particular type or types of sensors are of limited use, if any, in classifying signals acquired from a different type or types of sensors due to differences in sensor characteristics. Advantageously, the present invention does not require "from scratch" training of new classifiers based on signals from different types of sensors. Instead, the present invention utilizes transfer learning methods (for mapping between the original sensor type on which the pre-existing classifier was trained and the new, different sensor type from which the new data requiring classification is obtained).

[0062] The present invention can utilize a pre-existing classifier 114 by modifying (also referred to as modifying, improving) the pre-existing classifier 114 based on the mapping between different sensor frequency response distributions of corresponding different sensor types, and only requires a small data set from different types of sensors (e.g., the first plurality of sensors 102) in order to perform such mapping and modification. Therefore, in a preferred embodiment of the present invention, a modified classifier can be generated based on the original pre-existing classifier 114 and a small new data set obtained from one or more types of sensors different from the one or more types of sensors on which the pre-existing classifier 114 is based. Despite the small data set supplied to the modified classifier, the modified classifier can be able to accurately identify faults in signals acquired by different types of sensors (e.g., the first plurality of sensors 102). The modified classifier utilizes the original pre-existing classifier 114 and maps it to different types of sensors (e.g., the first plurality of sensors 102) so as to be accurately applicable to signals acquired by different types of sensors.

[0063] However, it should be noted that the present invention can be useful even in situations where large, high-quality data sets from different types of sensors (e.g., the first plurality of sensors 102) are used. While in this case, since a large, high-quality data set is available, a new specialized classifier can be trained to provide adequate results, it is still advantageous to use transfer learning to modify the pre-existing classifier in order to leverage the capabilities of the original pre-existing classifier. Thus, while it is expected that the present invention is most useful in situations where there is insufficient data in terms of quantity and / or quality to train a new classifier, the present invention is also useful in situations where a large, high-quality data set is available.

[0064] In the case of using a pre-existing accurate fault classifier trained on signals acquired by a specific type or types of sensors, the present invention therefore provides a solution for modifying the classifier so as to be able to accurately classify signals acquired by any other type or types of sensors than the specific type or types of sensors on which the fault classifier was trained, wherein these signals emanating from machines have at least one shared characteristic with the signals of those machines on which the classifier was previously trained. This may be referred to as a sensor-independent approach, wherein the classifier may be calibrated to be able to be applied to data acquired by any sensor, regardless of the type and / or structure of the source data collected by the sensor.

[0065] The method of the present invention can be applied to the case where the same type of signal (e.g., vibration signal) is obtained by different types of vibration sensors. FIG. 5A to FIG. 5C The method of the present invention can also be applied to the case where different types of signals (for example, vibration signals and magnetic signals) are acquired by different types of sensors (for example, vibration sensors and magnetic flux sensors). The result is shown in Figure 5D as shown in .

[0066] In both cases, the original pre-existing classifier can be modified based on the mapping between the sensor frequency response distributions in order to create a modified classifier capable of identifying faults in signals acquired by a sensor of a different type than the type of sensor upon which the pre-existing classifier was trained. In both cases, the modified classifier can be applied to signals acquired by a sensor of a different type than the type of sensor upon which the pre-existing classifier was trained with greater accuracy than would be obtained by applying the pre-existing classifier in its original, unmodified form. Furthermore, in both cases, the modified classifier can be applied to signals acquired by a sensor of a different type than the type of sensor upon which the pre-existing classifier was trained with greater accuracy than would be obtained with a new classifier trained using only signals acquired by sensors of a different type.

[0067] Obtaining a small new data set from a different type of sensor (eg, first plurality of sensors 102) has been described above with respect to sensors 102 acquiring signals and optionally associated operating condition data from machine 104. Figure 1 The system for acquiring a small data set is preferably implemented by those elements included in the dashed box 115.

[0068] exist Figure 1 In the embodiment of the present invention shown in , the first plurality of sensors 102 are preferably all sensors of the same type (ie, sensor type S1). However, as previously mentioned, the first plurality of sensors 102 may alternatively include more than one type of sensor. Figure 2 An alternative embodiment of the system 100 (represented herein as system 100A) is shown including more than one type of sensor in the first plurality of sensors 102. Turning now to Figure 2 , the first plurality of sensors 102 may include sensors S1, S11 to S1n on each of machines 1 to N of the plurality of machines 104. It should be understood that the first plurality of sensors 102 may include any number of sensors S1 to S1N (such as two or more sensors). These sensors may be, for example, vibration sensors, flux sensors, current sensors, temperature sensors, or a combination of sensors for sensing other parameters related to the operation of the machines 104 (such as torque, displacement, input line frequency, etc.). The sensors S1, S11 to S1N may be of different types from each other, but it is preferred for the sensor group incorporated into each of the machines 104 to be of the same type. However, it should be understood that the system 100A may allow for the situation where some missing data from one of the sensors 102 is present, which may be estimated, for example, by using multivariate statistics.

[0069] The system 100A may be generally similar to the system 100 , except that in the system 110A, multiple types of sensors are included in the first plurality of sensors 102 , and the description of the system 100 also generally applies to the system 100A.

[0070] Continue to refer to Figure 1 and Figure 2 , the small new data set (based on which the pre-existing classifier 114 can be modified by mapping learning to provide the modified classifier) ​​can include much less data than the amount of data based on which the pre-existing classifier 114 was initially trained. For example, the small new data set can include less than 400 sets, less than 300 sets, less than 200 sets, less than 100 sets, less than 50 sets, or less than 30 sets of each of the first plurality of sets of signals, and optionally include the first plurality of sets of operating condition data of the first plurality of mechanical machines 104. In addition, the small new data set can be a smaller number of sets of signals corresponding to a faulty operating state of a mechanical machine in the first plurality of mechanical machines 104 (such as less than 100 sets, less than 90 sets, less than 80 sets, less than 70 sets, less than 60 sets, less than 50 sets, less than 30 sets, or less than 20 sets of signals corresponding to a faulty operating state of a mechanical machine in the first plurality of mechanical machines 104). As described above, in some cases, the small new data set may not even include signals corresponding to a faulty operating state of a mechanical machine in the first plurality of mechanical machines 104. This is in contrast to much larger data sets (such as several thousand samples) upon which pre-existing classifiers 114 may have been trained.

[0071] It should be understood that the first plurality of sensors 102 are different from the second plurality of sensors upon which the pre-existing classifier 114 is trained, but the machines 104 preferably share common characteristics with each other and with the machines upon which the pre-existing classifier 114 is trained. The machines 104 may or may not be the same as those upon which the pre-existing classifier 114 is pre-trained.

[0072] The first plurality of sets of signals acquired by the first plurality of sensors 102 and optionally the first plurality of sets of operating condition data collected at the data collection unit 110 of the first plurality of mechanical machines 104 may be supplied to the pre-existing fault classifier 114. The first plurality of sets of signals may be pre-processed by the signal processor 108 before being supplied to the pre-existing fault classifier 114.

[0073] Subsequently, the pre-existing fault classifier 114 is preferably modified based on the plurality of sets of signals and optionally based on the plurality of sets of operating condition data for the first plurality of mechanical machines 104 to produce a modified classifier 116 .

[0074] The modified classifier 116 can be an algorithmic classifier that can be executed by one or more processors, which can be the same or different processors as the processor that executes the pre-existing classifier 114. For example, the pre-existing classifier 114 and the modified classifier 116 can be embodied within a data processing unit. The one or more processors that execute the pre-existing fault classifier 116 can be remote processors, such as located in the cloud, or can be local processors. For example, the one or more processors that execute the modified classifier 116 can be located within multiple sensors 102 in the sensors 102.

[0075] The pre-existing fault classifier 114 is modified by adapting the classifier to a new data set including the plurality of sets of signals acquired by the sensors 102 and optionally the plurality of sets of operating condition data of the first plurality of mechanical machines 104. For example, Figure 1 As shown in , in the case where the first plurality of sensors 102 only includes sensor type S1, the modified classifier 116 may be adapted to sensor type S1. Figure 2 As shown in , where the first plurality of sensors includes a plurality of sensor types S1, S11, etc., the modified classifier 116 may be adaptable to those plurality of sensor types. By adjusting the pre-existing classifier 114 to produce the modified classifier 116, the classifier is calibrated for a data set that is different from the data set on which the original training classifier was based. The adjustment includes a mapping between the original sensor characteristics on which the previous training classifier 114 was based and the sensor characteristics of the first plurality of sensors 102.

[0076] In one embodiment of the invention, the adjustment includes a mapping between the distribution of sensor frequency responses of the original plurality of sensors upon which the classifier 114 was previously trained and the distribution of sensor frequency responses of the first plurality of sensors 102. The mapping may be between sensor frequency response distributions, rather than simply between sensor frequency responses, because both the frequency responses of the sensors upon which the classifier 114 was previously trained and the frequency responses of the first plurality of sensors 102 may be distributed. Such a distribution may result from actual variations between each plurality of sensors, such as variations in machine characteristics, variations in sensor locations, sensor mounting, the exact type of sensor mounting, etc. New classifier scores

[0077] The mapping may be performed in a supervised manner. In this embodiment, a data collection unit 110 may be included in the system 100, and the labeled signals are supplied to a pre-existing fault classifier 114. Figure 3 1 to provide more details on how to modify the pre-existing fault classifier 114 by mapping between sensors in a supervised manner.

[0078] The mapping may optionally be performed in an unsupervised manner. In this case, the data collection unit 110 need not be included in the system 100, and no signal of the associated operating state is supplied to the pre-existing fault classifier 114. The pre-existing fault classifier 114 may be modified by unsupervised learning to map the difference between the sensor frequency response distribution of the sensor on which the classifier 114 was previously trained and the sensor frequency response distribution of the new sensor (e.g., the first plurality of sensors 102).

[0079] The mapping may optionally be performed in a semi-supervised manner, wherein the labeled signals are fed to a pre-existing fault classifier 114, and the mapping between sensors is performed in an unsupervised manner to produce a modified classifier 116. Figures 3 to 4B to provide more details about the unsupervised mapping.

[0080] Having been adapted for use with data from the first plurality of sensors 102 through the mapping, the modified classifier 116 is now ready for use in classifying signals acquired by one or more types of sensors of the same type as the first plurality of sensors 102 .

[0081] An example of the use of the modified classifier 116 for identifying faults in a signal acquired by a sensor of the same type as the first plurality of sensors 102 is further shown in Figure 1 and Figure 2 In. Figure 1 and Figure 2 As shown in FIG. , at least one sensor 122 (e.g., Figure 1 Sensor S1 and Figure 2 Sensors S1 to S1N in the system 100 may acquire at least one set of signals emitted from at least one given mechanical machine (depicted herein as mechanical machine X, indicated by reference numeral 124). The mechanical machine 124 may share at least one characteristic with the first plurality of mechanical machines 104 and the plurality of mechanical machines based on the signals upon which the pre-existing fault classifier 114 is trained. The mechanical machine 124 may or may not be a component of the first plurality of mechanical machines 104. It should be understood that although the mechanical machine 124 is shown herein as being implemented as a single machine, this is for simplicity only, and the system 100 may include any number of mechanical machines 124 (such as one, two, or more mechanical machines 124) that may be the same or different from one another, as long as the mechanical machines 124 share at least one characteristic as described above.

[0082] At least one sensor 122 is one of the first plurality of sensors 102, which means that the sensor 122 is of the same type as the first plurality of sensors 102. Figure 1In the case where the first plurality of sensors 102 includes sensors of a single type (ie, sensor S1), the sensor 122 also has sensor type S1. Figure 2 As shown in , in the case where the first plurality of sensors 102 includes multiple types of sensors (i.e., sensors S1 to S1N), the at least one sensor 122 also includes the same multiple types of sensors (i.e., sensors S1 to S1N). At least one sensor 122 is preferably operable to acquire at least one set of signals emitted from a given mechanical machine 124. The at least one set of signals can be a "signal snapshot" sensed by the at least one sensor 122 over a short period of time. For example, the signal can be sensed over a period of several seconds (such as 1 second to 4 seconds). The at least one set of signals can optionally include multiple "signal snapshots" over time (for example, a four-second "signal snapshot" measured every hour over a period of hours, days, or even months). Optionally, the at least one set of signals can include signals that are continuously monitored over a longer period of time or more than one period of time. For example, the signal can be continuously monitored every millisecond.

[0083] The signals acquired by the at least one sensor 122 may be pre-processed, for example, by analog or digital processing capabilities of the sensor 122 itself or by other hardware and / or software signal processing components 128. Figure 1 and Figure 2 While the signal processor 128 is shown as a separate element from the at least one sensor 122, the signal processing functionality may be incorporated into one or more of the sensors 122. Signal preprocessing may include at least one of digitization, compression, feature extraction, and signal representation in the time domain or frequency domain.

[0084] In one embodiment of the present invention, the signal acquired by the at least one sensor 122 may be uploaded to a remote server (such as a server in the cloud). The signal processing function 128 may be executed at the remote server.

[0085] A set of signals emitted from at least one given mechanical machine (herein exemplified as mechanical machine X) and acquired by at least one sensor (herein exemplified as sensor 122) may be provided to the modified classifier 116. The modified classifier 116 may be configured to automatically identify at least one fault of the mechanical machine X based on the set of signals emitted therefrom. It should be understood that since the modified classifier 116 has been adapted to classify signals acquired by sensors of the type of the first plurality of sensors 102, the modified classifier 116 may be precisely applied to signals acquired by the at least one sensor 122.

[0086] Identifying faults by the modified classifier 116 may include identifying one or more specific faults of the monitored machine 104. Depending on the specific machine 104 being monitored, the identified specific faults may include bearing wear, mechanical looseness, misalignment, imbalance, electrical faults, or other faults of the rotating machine.

[0087] Identifying a fault may optionally include identifying that machine X is in a fault state (i.e., an abnormal state relative to a normal, healthy operating state of machine X), but not identifying a specific fault. In this case, fault identification identifies the machine as not operating in a healthy manner, but does not identify what the specific cause of the unhealthy operation is.

[0088] The systems 100 and 100A may additionally include an output device 130. The output device 130 may be operable to receive the fault identification output by the modified classifier 116 and provide a human-perceptible output, the human-perceptible output including at least an identification of a fault of at least one given mechanical machine 124. The human-perceptible output may include at least one of a visual, tactile, or auditory output. Preferably, at least one of a repair or maintenance operation is performed based on the human-perceptible output.

[0089] For example, where the modified classifier 116 is executable by a remote processor, the fault identification output by the modified classifier 116 is transmitted to the output machine 130 .

[0090] The output device 130 may also be operable to provide a prediction of failure of at least one given mechanical machine 124 due to a fault identified by the modified classifier 116 without performing a recommended maintenance operation on the at least one given mechanical machine 124, wherein the at least one given mechanical machine 124 will actually fail if the recommended maintenance operation is not performed.

[0091] In some cases, maintenance 132 may be performed on a given machine 124 in response to human-perceivable output provided by an output machine 130. The output device 130 may optionally be operably coupled to a controller of the machine 124 and may adjust the operation of the machine 124 in response to a fault identification. For example, the machine 124 may be shut down, may be operated at reduced power, or otherwise adjusted. Such adjustments may be automatic or may be directed by a human expert in response to human-perceivable output provided by the output machine 130.

[0092] Now refer to Figures 3 to 4B Provides more details of the pre-existing classifier 114 itself and how the pre-existing classifier 114 may be adjusted or calibrated in order to produce the modified classifier 116, Figures 3 to 4B They are Figure 1 and Figure 2A simplified diagram of the modification of a classifier implemented by any system of the type shown in FIG. 1 and a diagram showing ... Figure 1 or Figure 2 1 and 100A are simplified corresponding block diagrams of components of two possible systems of classifiers used in the system. In one embodiment of the present invention, the system 100 and 100A may include a non-transitory computer-readable storage medium having computer-executable instructions stored thereon for executing, by a processor, a modification method of the classifier (as described below with respect to Figure 4A and Figure 4B Such a processor may be a remote processor or a local processor.

[0093] like Figure 3 As shown in , the pre-existing fault classifier 114 may be an artificial neural network (ANN) classifier 300 for supervised fault and anomaly detection. For example, the ANN 300 may be configured to identify specific faults or performance anomalies in a machine having a rotating component. Without loss of generality, the fault classifier 114 having a multi-layer perceptron architecture is shown in Figure 3 However, it should be understood that the system and method of the present invention can be applied to any type of fault classifier, regardless of its structure. More specifically, to illustrate the principles of the present invention, Figure 3 The number of layers and neurons shown in are only highly simplified examples. In addition, the pre-existing fault classifier 114 can be an unsupervised fault classifier as described above (such as an autoencoder, a deep belief network, a clustering-based classifier, a K-means or hidden Markov model or any other type of model capable of performing unsupervised learning).

[0094] Figure 3 The upper panel 302 in FIG. 4 shows an example of the classifier 114 in its original, unmodified form after training of the classifier 114 and before any modification thereof for mapping between sensor types in accordance with the present invention.

[0095] The pre-existing classifier 114 may have a multi-layer architecture. The data layer 308 may be a layer that includes input data in the form of multiple sets of sensor signals. The input layer 310 may be an initial layer that inputs the sensor signal set into the network 300. At each of the multiple subsequent hidden layers 312, the signal set is fused with respective weights and an activation function applied to the combined layer output before forwarding the signal set to the next layer of the hidden layer 312. The process repeats itself for each layer of the hidden layer 312 until the output layer 314 is reached. The output layer 314 generates a fault score. The score represents the identification of the fault. Identifying the fault may include identifying a current specific fault or anomaly or predicting a specific fault that is about to occur in the future. The fault may include any type of machine anomaly.

[0096] In the network layer Figure 3 The general mathematical expression of the ANN architecture shown in panel 302 is:

[0097]

[0098] Where s is the neuron input, which for the first layer 310 is the sensor signal, σ is the activation function, w is the weight, b is the bias term, and i and j are the indices operating on the layer input data points and neurons, respectively.

[0099] Training the network 300 to generate a pre-existing classifier 114 can be performed by referring to Figure 4A and Figure 4B To understand better.

[0100] First go to Figure 4A , the system 400 for training the pre-existing classifier 114 may include a second plurality of sensors 402. The sensors 402 are preferably coupled to respective second plurality of mechanical machines 404. The second plurality of sensors 402 may all be of the same second type (represented herein as sensor type S2). Optionally, the sensors 402 may include multiple types of sensors coupled to each respective second plurality of mechanical machines 404 in the second plurality of mechanical machines 404. The case where the second plurality of sensors 402 includes multiple types of sensors is shown in Figure 4B System 400A is generally similar to system 400 except that system 400A includes multiple types of sensors 402 therein. Figure 4B , should understand, Figure 4BThe second plurality of sensors 402 shown in FIG. 4 may include any number of sensors S2 to S2N (such as two or more sensors). These sensors may be, for example, vibration sensors, flux sensors, current sensors, temperature sensors, or a combination of sensors for sensing other parameters associated with the operation of the machine 404 (such as torque, displacement, input line frequency, etc.). The sensors S2 to S2N may be of different types from each other, but preferably are of the same type for each set of sensors in the plurality of sets of sensors associated with each of the machines 404.

[0101] The sensor 402 (eg, sensor S2 or sensors S2 to S2N) may be different from Figure 1 The first plurality of sensors 102 or different Figure 2 The first plurality of sensors 102 are of sensor types S1 - S1N. Figure 1 , Figure 2 , Figure 4A and Figure 4B Any combination of the embodiments shown in is possible. That is, the first plurality of sensors 102 may include only one type of sensor S1, and the second plurality of sensors 402 may include only one type of sensor S2 (different from S1). Figure 1 and Figure 4A The first plurality of sensors 102 may include only one type of sensor S1, the second plurality of sensors 402 may include multiple types of sensors S2 to S2N, and at least some of the multiple types of sensors S2 to S2N are different from S1 ( Figure 1 and Figure 4B Embodiment); the first plurality of sensors 102 may include multiple types of sensors S1 to S1N, and the second plurality of sensors 402 may include only one type of sensor S2 that is different from at least some of S1 to S1N ( Figure 2 and Figure 4A The first plurality of sensors 102 may include a plurality of types of sensors S1 to S1N, the second plurality of sensors 402 may include a plurality of types of sensors S2 to S2N, and at least some of the plurality of types of sensors S2 to S2N are different from S1 to S1N ( Figure 2 and Figure 4B of the embodiment of the invention).

[0102] In some embodiments, the first plurality of sensors 102 and the second plurality of sensors 402 may include some sensors in common, for example, the first plurality of sensors 102 may include vibrators and magnetic sensors, and the second plurality of sensors 402 may include only magnetic sensors.

[0103] Here, as an example, the second plurality of mechanical machines 404 is shown to include mechanical machine 1, mechanical machine 2 to mechanical machine N, where N can be any number of mechanical machines (such as two or more mechanical machines). Typically, the sensors 402 are coupled to the mechanical machines 404 in a one-to-one corresponding arrangement, where one of the sensors 402 is coupled to a corresponding one of the machines 404 in the machines 104. Other arrangements are also possible, where a single sensor can be arranged to sense signals from more than one mechanical machine 404. The sensor 402 can be in physical contact with the machine 404 (such as mounted directly or indirectly on the machine 404). For example, where the sensor 402 is an optical sensor, the sensor 402 can optionally be physically separated from the machine 404 (such as located at a given distance from the machine 404).

[0104] In one embodiment of the present invention, the sensors of the second plurality of sensors 402 may have substantially the same frequency response as each other. The frequency response of the second plurality of sensors 402 may be referred to as a frequency response distribution of the sensors 402. The first plurality of sensors 102 may also have substantially the same frequency response as each other. The frequency response of the first plurality of sensors 102 may be referred to as a frequency response distribution of the sensors 102. Since the sensors 402 are of a different type than the type of the sensors 102, the frequency response distribution of the first plurality of sensors 102 is different from the frequency response distribution of the second plurality of sensors 402. For example, the first plurality of sensors 102 may be single-axis vibration sensors and the second plurality of sensors 402 may be multi-axis vibration sensors, or vice versa, or the first plurality of sensors may be single-axis vibration sensors and the second plurality of sensors may be flux sensors. The sensors 102 and 402 or the sensors 402 and 102 may measure, by way of example only, vibration and flux signals; vibration and current signals; vibration and temperature signals; current and flux signals; and vibration and machine internal pressure signals, respectively.

[0105] The sensor 402 is preferably operable to sense signals emitted from the mechanical machine 404. The mechanical machines 1 to N, which are members of the second plurality of mechanical machines 404, are preferably characterized by one or more shared characteristics between each other and one or more shared characteristics with the mechanical machine 104. The mechanical machines 1 to N may or may not be the same machines, as long as the mechanical machines 1 to N have at least one shared characteristic in common. For example, the shared characteristic may refer to a type, model, manufacturer, physical characteristics or dimensions, operating characteristics or parameters, or other shared characteristics (the other shared characteristics indicate that the observed behavior of one mechanical machine in the plurality of mechanical machines may be typical of another mechanical machine in the plurality of mechanical machines).

[0106] The second plurality of sensors 402 are preferably operable to acquire a second plurality of sets of signals emitted from the second plurality of mechanical machines 404. The plurality of sets of signals may be "signal snapshots" sensed by an appropriate one of the sensors 402 over a short period of time. For example, the signals may be sensed over a period of a few seconds (such as 1 to 4 seconds). Each set of signals may optionally include a plurality of "signal snapshots" over time (e.g., a four-second "signal snapshot" measured every hour over a period of hours, days, or even months). Alternatively, each set of signals may include signals monitored continuously over a longer period of time or over more than one period of time. For example, the signals may be monitored continuously every millisecond.

[0107] The signal acquired by the sensor 402 may be pre-processed, for example, by analog or digital processing capabilities of the sensor 402 itself or by other hardware and / or software signal processing components 408. For example, the signal processing may be at least one of digitizing, compressing, extracting features from the signal, and representing the signal in the time domain or frequency domain.

[0108] The systems 400 and 400A may optionally include a data collection unit 410. The data collection unit 410 is preferably operable to receive a second plurality of sets of operating condition data regarding mechanical machines in the second plurality of mechanical machines, each set of operating condition data indicating an operating state of a mechanical machine in the second plurality of mechanical machines 404, each operating state being associated with at least one of the plurality of sets of signals acquired by the second plurality of sensors 402.

[0109] The operating status data collected at data collection unit 410 is preferably in the form of machine status diagnostics supplied by a human expert such as an engineer. The plurality of sets of signals acquired by first plurality of sensors 402 are optionally provided to data collection unit 410 directly or indirectly by sensors 402 and / or by signal processor 408.

[0110] These human experts can analyze the plurality of sets of signals acquired by the second plurality of sensors 402 and label each of the plurality of sets of signals as representing a particular operating state of the corresponding mechanical machine 404 from which the signals were emitted. The human expert can interact with a user interface (e.g., with a user interface of the data collection unit 410, or with a user interface of another device capable of communicating between the human expert and the data collection unit) to input operating state data. The signals and labels applied to the data collection unit 410 can be accumulated and stored in a database in the data collection unit 410. In one embodiment of the present invention, the data collection unit 410 can be located in a remote server such as a server in the cloud.

[0111] Identifying a fault by a human expert may include identifying one or more specific faults of the monitored machine 404. Depending on the specific machine 404 being monitored, the identified specific faults may include bearing wear, mechanical looseness, misalignment, imbalance, electrical faults, or other faults of the rotating machine. Identifying a fault may optionally include identifying that the machine 404 is in a fault state (i.e., an abnormal state relative to a normal, healthy operating state of the identified machine 404), but not identifying a specific fault. In this case, the human expert fault identification identifies the machine as not operating in a healthy manner, but does not identify what the specific cause of the unhealthy operation is.

[0112] It should be understood that in this example, the system 400 / 400A preferably includes a data collector 410, and the signals are preferably labeled by human experts in order to train the fault classifier 300 in a supervised manner. However, in other embodiments of the present invention, the data collector 410 may be omitted and the signals need not be labeled by human experts. In this case, the pre-existing fault classifier 114 may be initially trained by using unsupervised learning.

[0113] Return now Figure 3 The network 300 can be trained by supplying the network 300 with multiple sets of signals acquired by the second plurality of sensors 402 as input data at layer 308 and supplying the network 300 with related operating state data labeled by a human expert as the required corresponding output at the output layer 314. The network 300 can be trained using back propagation and gradient descent algorithms with respect to predefined data labels. Training parameters such as loss function, learning rate, optimizer type, etc. are preferably selected relative to the output score. For example, for binary fault detection, a cross entropy loss function, an Adam optimizer, and an L2 regularization term can be selected. The training process is first applied to a data set that is consistent across all sensors 402 in the entire data set, which means that for each vector in the data point, there is preferably no mixing between the sensors 402.

[0114] Once training is complete, the parameters of the network 300 established based on the training are preferably fixed. The parameters include activation functions and weights. The network 300 now constitutes a pre-existing classifier (such as the pre-existing classifier 114) based on the set of sensor signals acquired by the second plurality of sensors 402. The pre-existing classifier 114 in the form of the trained network 300 shown in panel 302 is now configured to accurately classify new input data that has the same or similar structure and is acquired from the same or similar source as the source of the data based on which the pre-existing classifier 114 was trained. In this case, the pre-existing classifier 114 in the form of the trained network 300 is configured to accurately classify new input signals that are acquired from the same or similar source as the data based on which the pre-existing classifier 114 was trained. Figure 4A and Figure 4B ) is emitted by a machine having a shared characteristic and is transmitted by a second plurality of sensors 402 ( Figure 4A and Figure 4B ). However, as described in detail above, the pre-existing classifier 114 in the form of the trained network 300 may not accurately classify new input signals sensed by sensors of a different type than the second plurality of sensors 402 (such as the first plurality of sensors 102), even though these new input signals are emitted from machines that share characteristics with the machines 404 (such as the first plurality of machines 104). In this case, the classifier accuracy will be significantly reduced due to the difference in frequency responses of the different types of sensors.

[0115] In order to enable the pre-existing classifier 114 to accurately classify a group of signals from sensors (e.g., the first plurality of sensors 102) that are of a different type than the type of sensors (e.g., the second plurality of sensors 402) upon which the classifier was previously trained, the pre-existing classifier 114 may be modified. In accordance with a preferred embodiment of the present invention, the modification of the pre-existing classifier 114 in the form of the network 300 is shown in FIG. Figure 3 In the lower panel 440.

[0116] As in Figure 3 The network 300 may be modified to produce a modified network 442. Preferably, the network 300 is modified by adding at least one additional layer 450 to the network 300. The at least one additional layer 450 may be referred to as a "mapping layer" and is configured to learn frequency response differences between the original sensor type (i.e., the second plurality of sensors 402) based on which the classifier 114 in the form of the network 300 is trained and the new data is collected from the new sensor type (i.e., the first plurality of sensors 102).

[0117] The configuration of the mapping layer 450 may be achieved by fixing the structure and parameters of all layers of the network 300 and retraining the classifier using a new data set acquired from the first plurality of sensors 102, as described above with respect to Figure 1 and Figure 2 115. In this way, since the parameters of all other layers of network 442 have been configured and fixed and cannot be changed, network 442 is forced to optimize the weight values ​​of mapping layer 450 relative to the frequency response difference of the two sensors. As a result, mapping layer 450 is forced to learn the weight values ​​of the mapping layer 450 in the presence of sensor 402 ( Figure 4A and Figure 4B) and the raw sensor signal group provided by sensor 102 ( Figure 1 and Figure 2 ) provides a mapping between new sensor signal groups.

[0118] Alternatively, the structure and parameters of all layers of the network 300 need not be fixed, but can be adjusted during retraining of the classifier using new data sets acquired from the first plurality of sensors 102. In this case, the parameters of the network 300 are used as a starting point for the adjusted parameters of the modified classifier 442.

[0119] The configuration of the mapping layer 450 is preferably selected with respect to the nature of the frequency response differences between the different types of sensors (eg, the first plurality of sensors 102 and the second plurality of sensors 402 ).

[0120] In the case of mapping between different sensors of the same type (eg, both the first plurality of sensors 102 and the second plurality of sensors 402 are vibration sensors, but of different types with different frequency responses), the frequency response difference is typically a linear function of the signal frequency.

[0121] In the case of mapping between different sensors of different types (e.g., the first plurality of sensors 102 and the second plurality of sensors 402 are vibration sensors and magnetic sensors with different frequency responses, respectively), the frequency response difference is usually a nonlinear function, so a nonlinear mapping is required. Accordingly, the activation function can (but does not have to) take a nonlinear form.

[0122] In some cases, the mapping of the mapping layer 450 can be assisted by providing new sensor data in a more complex form (e.g., a more complex form of signals acquired by the first plurality of sensors 102) to the mapping layer 450. For example, inverse data, logarithmic data, or other forms of data can be provided. The processing of modifying the input signal can be optimized relative to the classifier accuracy, thereby modifying the new sensor data based on the classifier accuracy performance feedback, and then retraining the mapping layer 450 using the modified data.

[0123] The mapping layer 450 is preferably incorporated into the network 442 as the first layer after the data layer 308 and is configured to receive input data in the form of a new data set. The mapping layer 450 is upstream of the original input layer 310 and precedes the original input layer 310 with respect to the input data. The location of the mapping layer 450 in the network 442 is important because the location of the mapping layer 450 as the initial layer in the network allows the mapping layer 450 to learn the mapping between sensor types and adjust the input new data to be in an appropriate form for continuing to other hidden layers 312 downstream in the network 442.

[0124] In an optional embodiment of the present invention, a pre-existing fault classifier may be initially trained in an unsupervised manner by providing a large amount of sensor data to the pre-existing fault classifier so that the classifier learns how to identify anomalies in the sensor data related to the machine operating conditions. For example, an autoencoder NN may be used to learn a low-dimensional representation of the sensor data and cluster-based classification, which is then used to identify outliers (anomalies). According to a preferred embodiment of the present invention, such a pre-existing unsupervised fault classifier 114 may then be modified in an unsupervised manner to produce a modified fault classifier 116, which is adjusted to identify anomalies in sensor data acquired by a different type of sensor than the type of sensor on which the pre-existing fault classifier 114 was initially trained. For example, the modified fault classifier 116 may be produced by adding an initial mapping layer to the autoencoder NN, fixing the parameters of all layers of the network except the mapping layer, and training the mapping layer to learn the mapping between the original sensor type and the new sensor type. During the training of the mapping layer, another mapping layer (such as an inverse mapping layer) may be additionally added at the output of the NN after the decoder for the training process. Cluster-based classification can then be used to classify the output in order to identify discrete values ​​(anomalies).

[0125] The modification of the classifier performance as a result of the mapping learning of the present invention is shown in the following exemplary figure:

[0126] exist Figure 5A , an example of the performance of the modified classifier of the present invention is shown in the form of a precision-recall curve. In this example, the fault classifier is initially trained and validated based on a data set including more than 100,000 labeled signal groups, 7,000 of which represent machine faults acquired by a single-axis vibration sensor from a rotating machine including a bearing. After training, the fault classifier is tested based on a data set including 20,000 signal groups, 1,400 of which correspond to machine faults. The fault classifier trained in this way can be referred to as a pre-existing fault classifier.

[0127] Then, according to a preferred embodiment of the present invention, a pre-existing classifier is modified to be able to classify faults based on signals obtained from a rotating machine including a bearing by a triaxial vibration sensor. A new data set of signals obtained by a triaxial vibration sensor and used to modify the pre-existing fault classifier includes a training and validation data set of 256 signal groups, of which only 20 groups of signals are related to faulty machine conditions (i.e., bearing wear). As described above, the pre-existing fault classifier is modified based on this very small signal group.

[0128] The modified fault classifier is then applied to a test set of 7149 signal groups, of which 486 are related to faulty machine states of signals obtained from a triaxial vibration sensor for a rotating machine including bearings for fault detection, in order to test its performance (line 502). For comparison purposes, a completely new classifier is trained "from scratch" with the same 256 signal groups and applied to the test set of the same 7149 signal groups (line 504). For completeness of comparison, the original pre-existing classifier in unmodified form is also applied to the 7149 example data sets (line 506). Figure 5A It can be clearly seen that despite the extremely small dataset provided, the modified classifier performs best in fault identification. Although other example scores can be used, Figure 5A The scores listed in are mean average precision scores.

[0129] exist Figure 5B , an example of the performance of the modified classifier of the present invention is shown in the form of a precision-recall curve. In this example, the fault classifier is initially trained and validated based on a data set including more than 100,000 labeled signal groups, 7,000 of which represent machine faults acquired by a single-axis vibration sensor from a rotating machine including a bearing. After training, the fault classifier is tested based on a data set including 20,000 signal groups, 1,400 of which correspond to machine faults. The fault classifier trained in this way can be referred to as a pre-existing fault classifier.

[0130] The pre-existing classifier is then modified according to a preferred embodiment of the present invention so as to be able to classify faults based on signals obtained from a rotating machine including a bearing by a triaxial vibration sensor. The new data set of signals obtained by the triaxial vibration sensor and used to modify the pre-existing fault classifier includes 924 signal groups, of which only 70 groups of signals are related to faulty machine conditions (i.e., bearing wear). As described above, the pre-existing fault classifier is modified based on this very small signal group. The modified fault classifier is then applied to a set of 6841 signal groups, which include 436 groups of signals corresponding to faulty machine conditions of signals obtained from a rotating machine including a bearing for fault detection by a triaxial vibration sensor, in order to test its performance (line 510). For comparison purposes, a completely new classifier is trained "from scratch" with the same 924 groups of signals, and the completely new classifier is also applied to the 6841 example data sets (line 512). For completeness of comparison, the original pre-existing classifier in an unmodified form is also applied to the 6841 example data sets (line 514). As considered Figure 5BIt can be clearly seen that despite the extremely small dataset provided, the modified classifier performs best in fault identification. Although other example scores can be used, Figure 5B The scores listed in are mean average precision scores.

[0131] exist Figure 5C , an example of the performance of the modified classifier of the present invention is shown in the form of a precision-recall curve. In this example, the fault classifier is initially trained and validated based on a data set including more than 100,000 labeled signal groups, 7,000 of which represent machine faults acquired by a single-axis vibration sensor from a rotating machine including a bearing. After training, the fault classifier is tested based on a data set including 20,000 signal groups, 1,400 of which correspond to machine faults. The fault classifier trained in this way can be referred to as a pre-existing fault classifier.

[0132] The pre-existing classifier is then modified according to a preferred embodiment of the present invention so as to be able to classify faults based on signals obtained from a rotating machine including a bearing by a triaxial vibration sensor. The new data set of signals obtained by the triaxial vibration sensor and used to modify the pre-existing fault classifier includes 5924 groups of signals, of which only 350 groups of signals are related to faulty machine conditions (i.e., bearing wear). As described above, the pre-existing fault classifier is modified based on this very small group of signals. The modified fault classifier is then applied to a set of 1481 example signal groups, the 1481 example signal groups including 56 groups of signals corresponding to faulty machine conditions of signals obtained from a rotating machine including a bearing for fault detection by a triaxial vibration sensor, in order to test its performance (line 520). For comparison purposes, a completely new classifier is trained "from scratch" with the same 5924 groups of signals, and the completely new classifier is also applied to the 1481 example data sets (line 522). For completeness of comparison, the original pre-existing classifier in unmodified form is also applied to the 1481 example data sets (line 524). As considered Figure 5C It can be clearly seen that the modified classifier performs best in fault identification. Although other example scores can be used, Figure 5C The scores listed in are mean average precision scores.

[0133] exist Figure 5D, an example of the performance of the modified classifier of the present invention is shown in the form of a precision-recall curve. In this example, the fault classifier is trained and validated based on a data set including more than 100,000 labeled signal groups, 7,000 of which represent machine faults acquired by a single-axis vibration sensor from a rotating machine including a bearing. After training, the fault classifier is tested based on a data set including 20,000 signal groups, 1,400 of which correspond to machine faults. The fault classifier trained in this way can be referred to as a pre-existing fault classifier.

[0134] The pre-existing classifier is then modified according to a preferred embodiment of the present invention so as to be able to classify faults based on signals obtained from a rotating machine including a bearing by a magnetic sensor. The new data set of signals obtained by the magnetic sensor and used to modify the pre-existing fault classifier includes 3411 groups of signals, of which only 222 groups of signals are related to faulty machine conditions (i.e., bearing wear). As described above, the pre-existing fault classifier is modified based on this very small group of signals. The pre-existing fault classifier is then applied to a test signal group of 754 signals, which only includes 50 example signal sets corresponding to machine faults obtained from a motor including a bearing for fault detection by a magnetic sensor, in order to test its performance (line 530). For comparison purposes, a completely new classifier is trained with the same 3411 groups of magnetic signals, and the completely new classifier is also applied to the 754 example magnetic data sets (line 532). For completeness of comparison, the original pre-existing classifier in an unmodified form is also applied to the same 754 example magnetic data sets (line 534). As considered Figure 5D It can be clearly seen that the modified classifier performs best in fault identification. Although other example scores can be used, Figure 5D The scores listed in are mean average precision scores.

[0135] exist FIG. 5A to FIG. 5D In the above example, the data set on which the classifier 114 was initially trained consisted of data sensed from more than 40,000 independent rotating machines (primarily including motors, pumps, fans, gearboxes, coolers, and compressors). The bearing records were recorded over a three-year period, some of which were measured more than once. Each bearing was measured along three axes by a uniaxial piezoelectric vibration sensor. In the case where the new data set includes data from a triaxial vibration sensor, this is a MEMS triaxial vibration sensor.

[0136] As is evident from the above data, the failure rate in the data set is approximately 7%. The classifier was designed to detect single bearing failures such that even in the case where an otherwise non-faulty machine bearing produces a signal that exhibits characteristics of bearing wear, the classifier is able to correctly classify the non-faulty bearing as such due to its proximity to the faulty bearing and due to acoustic wave propagation between bearings. Labeling of the data was performed by more than 10 human experts.

[0137] Now refer to Figure 6 , Figure 6 is a simplified flow chart showing the steps included in a mechanical machine fault identification method based on transfer learning according to a preferred embodiment of the present invention.

[0138] Figure 6 6 shows a method 600 for machine fault identification. As shown in a first step 602, a first set of signals emitted from a plurality of mechanical machines may be acquired by a first plurality of sensors. The sensors in the first plurality of sensors are preferably of the same first type as each other. Optionally, the sensors in the first plurality of sensors may be of multiple types. The sensors in the first plurality of sensors may have substantially the same sensor frequency response, which may be referred to as a first sensor frequency response distribution.

[0139] As shown in a second step 604, a first set of machine operating condition data corresponding to the first set of signals acquired in step 602 may optionally be acquired. The machine operating condition data may include an identification of an operating state associated with each of the first set of signals. The operating state may be a fault or non-fault state.

[0140] As shown in the third step 606, the plurality of sets of signals and optionally corresponding operating condition data may be supplied to a pre-existing fault classifier that is pre-trained to identify faults in the machine based on signals emitted from the machine and acquired by a second plurality of sensors. The second plurality of sensors upon which the pre-existing fault classifier is trained may be of types that are the same as each other but different from the types of the first plurality of sensors. Alternatively, the second plurality of sensors upon which the pre-existing fault classifier is trained may optionally be of a plurality of types, at least some of which are different from the types of the first plurality of sensors. The sensors in the second plurality of sensors may have substantially the same sensor frequency response, which may be referred to as a second sensor frequency response distribution. The second sensor frequency response distribution may be different from the first sensor frequency response distribution of the first plurality of sensors.

[0141] The pre-existing fault classifier is modified based on the new data supplied to the pre-existing fault classifier, preferably using a transfer learning method, as shown in a fourth step 608. The modification may include adding at least one mapping layer to the pre-existing fault classifier, the mapping layer being aware of the frequency response difference between the first sensor frequency response distribution and the second sensor frequency response distribution.

[0142] As shown in a fifth step 610, an additional set of signals may then be acquired from at least one machine using the same sensor type as the first plurality of sensors.

[0143] As shown at a sixth step 612, the modified classifier generated at step 608 may be applied to additional signal groups to identify faults in the machine that generated the additional signal groups.

[0144] As shown in a seventh step 614, machine maintenance or repairs may be performed based on the identified fault.

[0145] It should be understood that all of the various machines described as generating signals at steps 602, 606, and 610 may be the same machine or different machines having shared characteristics as described above.

[0146] Those skilled in the art will appreciate that the present invention is not limited to what has been specifically shown and described above. The scope of the present invention includes combinations and sub-combinations of the various features described above and their modifications, all of which are not in the prior art.

Claims

1. A method for identifying a fault of at least one mechanical machine, the method comprising: causing a first plurality of sensors coupled to corresponding first plurality of mechanical machines to acquire a first plurality of sets of signals emitted from the first plurality of mechanical machines, the first plurality of mechanical machines sharing at least one characteristic; supplying at least the first plurality of sets of signals of the first plurality of mechanical machines to a pre-existing fault classifier, the pre-existing fault classifier being pre-trained to automatically identify faults of a second plurality of mechanical machines based on signals emitted from the second plurality of mechanical machines and pre-acquired by a second plurality of sensors, the second plurality of sensors being of a different type than the first plurality of sensors, the second plurality of mechanical machines sharing the at least one characteristic; modifying the pre-existing fault classifier by using transfer learning based on at least the first plurality of sets of signals of the first plurality of mechanical machines to provide a modified fault classifier, wherein the pre-existing fault classifier comprises a neural network, the neural network comprising a data layer and an input layer for receiving data from the data layer, and modifying the pre-existing fault classifier comprises adding at least one mapping layer to the neural network, the at least one mapping layer being added between the data layer and the input layer, such that the at least one mapping layer is configured to receive the data from the data layer in the modified fault classifier; applying the modified fault classifier to at least one additional set of signals acquired by at least one sensor of the first plurality of sensors and emanating from at least one given mechanical machine sharing the at least one characteristic, the modified fault classifier being configured to automatically identify at least one fault of the at least one given mechanical machine based on the at least one additional set of signals; as well as A human-perceptible output is provided by an output device, the human-perceptible output comprising at least an identification of the fault of the at least one given mechanical machine, and at least one of a repair or maintenance operation is performed based on the human-perceptible output.

2. The method according to claim 1, further comprising: After causing the first plurality of sensors to acquire the first plurality of sets of signals and before supplying the first plurality of sets of signals to the pre-existing fault classifier: obtaining a first plurality of sets of operating condition data for a mechanical machine of the first plurality of mechanical machines, each set of operating condition data indicating an operating state of a mechanical machine of the first plurality of mechanical machines, each operating state being associated with at least one of the plurality of sets of signals; said supplying at least said first plurality of sets of signals to said pre-existing fault classifier further comprises supplying said operating condition data of said first plurality of mechanical machines to said pre-existing fault classifier; Modifying the pre-existing fault classifier by using transfer learning based at least on the first plurality of sets of signals further includes modifying the pre-existing fault classifier by using transfer learning based additionally on the first plurality of sets of operating condition data of the first plurality of mechanical machines.

3. The method according to claim 1 or 2, characterized in that: Identifying the fault includes identifying a specific fault of the at least one given mechanical machine and predicting a failure of the at least one given mechanical machine due to the specific fault without performing a recommended maintenance operation on the at least one given mechanical machine, wherein the at least one given mechanical machine will indeed fail if the recommended maintenance operation is not performed.

4. The method according to claim 2, characterized in that: Aside from adding the at least one mapping layer, the neural network is not otherwise modified by the modifying.

5. The method according to claim 4, characterized in that The first plurality of sensors have a first frequency response distribution, and the second plurality of sensors have a second frequency response distribution, and the mapping layer is configured to map between the first frequency response distribution and the second frequency response distribution.

6. The method according to claim 1, characterized in that The first plurality of sensors are operative to sense signals of the same type as signals sensed by the second plurality of sensors.

7. The method according to claim 6, characterized in that The same type of signal includes one of a vibration signal, a magnetic flux signal, an electric current, a temperature, and a machine internal pressure signal.

8. The method according to claim 1, characterized in that: The first plurality of sensors and the second plurality of sensors are operative to sense signals of different types relative to each other.

9. The method according to claim 8, characterized in that The mutually different types of signals include at least one of a vibration and a magnetic flux signal; a vibration and a current signal; a vibration and a temperature signal; a current and a magnetic flux signal; and a vibration and a machine internal pressure signal.

10. The method according to claim 2, characterized in that At least some of the operating states of a mechanical machine in the first plurality of mechanical machines as indicated by the first plurality of sets of operating condition data are faulty operating states.

11. The method according to claim 10, characterized in that The first plurality of sets of signals and the first plurality of sets of operating condition data for the first plurality of mechanical machines include less than 30 sets of the faulty operating conditions.

12. A system for identifying a fault in at least one mechanical machine, the system comprising: a first plurality of sensors coupled to a corresponding first plurality of mechanical machines and operable to acquire a first plurality of sets of signals emanating from the first plurality of mechanical machines, the first plurality of mechanical machines sharing at least one characteristic; The data processing unit is operated as follows: receiving the first plurality of sets of signals from the first plurality of mechanical machines, the data processing unit comprising a pre-existing fault classifier, the pre-existing fault classifier being pre-trained to automatically classify operating conditions of a second plurality of mechanical machines based on signals emitted from the second plurality of mechanical machines and pre-acquired by a second plurality of sensors, the second plurality of sensors being of a type different from that of the first plurality of sensors, the second plurality of mechanical machines sharing the at least one characteristic; modifying the pre-existing fault classifier by using transfer learning based on at least the first plurality of sets of signals of the first plurality of mechanical machines to provide a modified fault classifier, wherein the pre-existing fault classifier comprises a neural network, the neural network comprising a data layer and an input layer for receiving data from the data layer, and modifying the pre-existing fault classifier comprises adding at least one mapping layer to the neural network, the at least one mapping layer being added between the data layer and the input layer, such that the at least one mapping layer is configured to receive the data from the data layer in the modified fault classifier, and applying the modified fault classifier to at least one additional set of signals acquired by at least one sensor of the first plurality of sensors and emanating from at least one given mechanical machine sharing the at least one characteristic, the modified fault classifier being configured to automatically identify at least one fault of the at least one given mechanical machine based on the at least one additional set of signals, and an output device in communication with the data processing unit and operable to provide a human-perceptible output including at least an identification of the fault of the at least one given mechanical machine, at least one of a repair or maintenance operation being performed based on the human-perceptible output.

13. The system of claim 12, further comprising a data collection unit operable to obtain a first plurality of sets of operating condition data for mechanical machines of the first plurality of mechanical machines, each set of operating condition data indicating an operating state of a mechanical machine of the first plurality of mechanical machines, each operating state being associated with at least one of the plurality of sets of signals; The data processing unit is operative to receive the operating condition data of the first plurality of mechanical machines and to modify the pre-existing fault classifiers based additionally on the first plurality of sets of operating condition data of the first plurality of mechanical machines by using transfer learning.

14. The system according to claim 12 or 13, characterized in that Identifying the fault includes identifying a specific fault of the at least one given mechanical machine and predicting a failure of the at least one given mechanical machine due to the specific fault without performing a recommended maintenance operation on the at least one given mechanical machine, wherein the at least one given mechanical machine will indeed fail if the recommended maintenance operation is not performed.

15. The system according to claim 13, characterized in that Aside from adding the at least one mapping layer, the neural network is not otherwise modified by the modifying.

16. The system according to claim 12, characterized in that The first plurality of sensors have a first frequency response distribution, and the second plurality of sensors have a second frequency response distribution, and the mapping layer is configured to map between the first frequency response distribution and the second frequency response distribution.

17. The system according to claim 12, characterized in that The first plurality of sensors are operative to sense signals of the same type as signals sensed by the second plurality of sensors.

18. The system according to claim 17, characterized in that The same type of signal includes one of a vibration signal, a magnetic flux signal, an electric current, a temperature, and a machine internal pressure signal.

19. The system according to claim 12, characterized in that The first plurality of sensors and the second plurality of sensors are operative to sense signals of different types relative to each other.

20. The system according to claim 19, characterized in that The mutually different types of signals include at least one of a vibration and a magnetic flux signal; a vibration and a current signal; a vibration and a temperature signal; a current and a magnetic flux signal; and a vibration and a machine internal pressure signal.

21. The system according to claim 13, characterized in that At least some of the operating states of a mechanical machine in the first plurality of mechanical machines as indicated by the first plurality of sets of operating condition data are faulty operating states.

22. The system according to claim 21, characterized in that The first plurality of sets of signals and the first plurality of sets of operating condition data for the first plurality of mechanical machines include less than 30 sets of the faulty operating conditions.

Images