Optimizing the accuracy of machine learning algorithms for monitoring the operation of industrial machines

Abstract

Method for updating a machine learning algorithm of a computer system that actively monitors at least one industrial machine behavior model of at least one industrial machine, operating by providing additional training for the machine learning algorithm, comprising: - Monitoring of at least one industrial machine behavior model; - Identifying at least one first ambiguous segment of the at least one industrial machine behavior model that has a first set of features, and identifying a corrective solution recommendation associated with the first ambiguous segment; - Determining at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features, wherein the first ambiguous segment and the second ambiguous segment represent unclear behavior of at least one component of at least one industrial machine, and wherein the first ambiguous segment and the second ambiguous segment are each based on at least one obtained sensor reading of at least one sensor that is associated with the at least one industrial machine; - Determine whether the similarity between the first set of features and the second set of features exceeds a predetermined threshold; and - Updating the computer system's machine learning algorithm so that the corrective solution recommendation is assigned to the second ambiguous segment when the similarity exceeds the specified threshold.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority over the preliminary US application No. 62 / 717 855 filed on August 12, 2018, the contents of which are hereby incorporated by reference. TECHNICAL AREA

[0002] The present disclosure relates generally to maintenance systems for machines and in particular to the monitoring of machine operations for the improvement of machine processes. BACKGROUND

[0003] Communication, data processing, cloud computing, artificial intelligence, and other computer-based technologies have advanced considerably in recent years, ushering in new technology and production sectors. Furthermore, many industrial technologies that have been in use since the 1970s or earlier are still in operation today. Existing solutions related to these industrial technologies have often undergone only minor improvements, resulting in only slight increases in production and yield.

[0004] In modern manufacturing practices, manufacturers often have to adhere to strict production schedules and deliver flawless or near-flawless product quality. Consequently, these manufacturers risk significant losses if an unexpected machine failure occurs. A machine failure is an event that occurs when a machine deviates from its correct operating condition. Faults, which are typically deviations from the correct state of the machine, are not necessarily failures themselves, but they can lead to and indicate potential future failures. In addition to failures, faults can also cause unusual machine behavior that can negatively impact performance.

[0005] The average downtime due to machine failure for typical manufacturers (i.e., the average time production is partially or completely shut down due to a machine failure) is 17 days per year, meaning 17 days of lost production and, consequently, lost revenue. For a typical 450-megawatt power plant turbine, for example, a single day of downtime can cost a manufacturer over $3 million in lost revenue. Such downtime can also incur additional costs for repairs, safety precautions, and similar expenses.

[0006] Power plants spend billions of US dollars annually to ensure reliability. In particular, billions are allocated to backup systems and redundancies to minimize production downtime. Additionally, monitoring systems can be used to quickly detect outages and thus expedite the return to production when a failure occurs. However, existing monitoring systems typically detect outages only after or immediately before the outage begins.

[0007] Furthermore, existing solutions for monitoring machine failures typically rely on a set of predefined rules for each machine. These rule sets do not consider all the data that can be collected about the machine and are used only to check certain key parameters, while ignoring the rest. Moreover, these rule sets must be pre-defined by engineers or other human analysts. As a result, existing solutions may only utilize a portion of the collected data, leading to a waste of computing resources associated with the transmission, storage, and processing of unused data. Moreover, the failure to consider all relevant data can result in missed or otherwise inaccurate failure detection or prediction.

[0008] Furthermore, existing solutions often rely on periodic testing at predetermined intervals. Therefore, even with existing solutions capable of predicting failures, machine maintenance requests are typically returned even when the machine is not immediately at risk of failure. Such premature replacement and maintenance lead to material waste and expenses for replacing perfectly functional parts. Moreover, these existing solutions often result in repairs being initiated only after a fault has occurred. Consequently, failures cannot be prevented, leading to downtime and lost revenue.

[0009] Furthermore, existing monitoring and maintenance solutions often require specialized testing equipment. Consequently, these solutions typically require specialized operators who are highly trained in the operation of the respective monitoring and maintenance system. The need for specialized operators can be inconvenient and costly, and introduces potential sources of human error. Moreover, given the sheer volume of data that can be collected for a given machine and the minute variations within that data, a human analyst is unable to adequately predict impending failures.

[0010] It would therefore be advantageous to offer a solution that overcomes the challenges mentioned above. SUMMARY

[0011] What follows is a summary of several exemplary embodiments of the disclosure. This summary is intended to provide the reader with a basic understanding of these embodiments and does not fully define the scope of the disclosure. It is not a comprehensive overview of all embodiments under consideration and is not intended to identify important or critical elements of all embodiments, nor to delimit the scope of any one or all aspects. Its sole purpose is to present some concepts of one or more embodiments in simplified form as a prelude to the more detailed description that will be presented later. For the sake of simplicity, the term "certain embodiments" may be used here to refer to a single embodiment or several embodiments of the disclosure.

[0012] Certain embodiments disclosed herein comprise a method for optimizing machine learning algorithms for monitoring the operation of industrial machines, comprising: monitoring at least one industrial machine behavior model of at least one industrial machine; identifying at least one first ambiguous segment of the at least one industrial machine behavior model, which has a first set of features, and identifying a corrective solution recommendation associated with the first ambiguous segment; identifying at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features; and determining whether a similarity between the first set of features and the second set of features exceeds a predetermined threshold.and updating a machine learning algorithm of the at least one industrial machine behavior model to assign the corrective solution recommendation to the second ambiguous segment when it is determined that the similarity has exceeded the specified threshold.

[0013] Certain embodiments disclosed herein also include a non-transitory, computer-readable medium on which instructions are stored to cause a processing circuit to execute a method, wherein the method comprises: monitoring at least one industrial machine behavior model of at least one industrial machine; identifying at least one first ambiguous segment of the at least one industrial machine behavior model, which has a first set of features, and identifying a corrective solution recommendation associated with the first ambiguous segment; identifying at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features; and determining whether a similarity between the first set of features and the second set of features exceeds a predetermined threshold.and updating a machine learning algorithm of the at least one industrial machine behavior model to assign the corrective solution recommendation to the second ambiguous segment when it is determined that the similarity has exceeded the specified threshold.

[0014] Certain embodiments disclosed herein also include a system for optimizing machine learning algorithms for monitoring the operation of an industrial machine, comprising: a processing circuit; and a memory, wherein the memory contains instructions which, when executed by the processing circuit, configure the system to: monitor at least one industrial machine behavior model of at least one industrial machine; identify at least one first ambiguous segment of the at least one industrial machine behavior model having a first set of features, and identify a corrective solution recommendation associated with the first ambiguous segment; identify at least one second ambiguous segment of the at least one industrial machine behavior model having a second set of features;to determine whether a similarity between the first set of features and the second set of features exceeds a predetermined threshold; and to update a machine learning algorithm of the at least one industrial machine behavior model to assign the corrective solution recommendation to the second ambiguous segment when it is determined that the similarity has exceeded the predetermined threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The subject matter disclosed herein is specifically highlighted and clearly claimed in the claims at the end of the description. The aforementioned and other objectives, features, and advantages of the disclosed embodiments will be evident from the following detailed description in conjunction with the accompanying drawings. Fig. Figure 1 is a network diagram used to describe the various disclosed embodiments. Fig. Figure 2 is a schematic representation of the management server system according to one embodiment. Fig. Figure 3 is a flowchart illustrating a procedure for improving the accuracy level of a machine learning algorithm suitable for monitoring machine operation according to one embodiment. Fig. Figure 4 is a flowchart that illustrates a reinforcement learning-based method for automatically providing corrective solution recommendations for machine operation according to one embodiment. Fig. Figure 5 is a flowchart illustrating a reinforcement learning-based procedure for updating a machine learning algorithm set up to monitor machine operation according to one embodiment. Fig. Figure 6 is an example simulation showing the representation of an ambiguous segment in a machine behavior model according to one embodiment. DETAILED DESCRIPTION

[0016] It is important to note that the embodiments disclosed herein are only examples of the many advantageous uses of the innovative teachings contained herein. In general, the statements made in the description of the present application do not necessarily limit any of the various claimed embodiments. Furthermore, some statements may apply to some inventive features but not to others. In general, unless otherwise stated, individual elements may be plural and vice versa without affecting the generality of the invention. In the drawings, identical numerals refer to identical parts in multiple views.

[0017] The disclosed reinforcement learning-based method is used to identify ambiguous segments in a machine behavior model intended for optimizing a machine learning algorithm to monitor the operation of an industrial machine. The machine behavior model is based on sensory inputs received from one or more sensors of the machine. In response to the identification of such an ambiguous segment, a query is generated and sent to a client device. A response, i.e., an input, is then received with respect to the query. This response is used to update a machine learning algorithm configured to monitor machine operation and, in particular, to predict an impending machine failure in a timely manner. In a further embodiment, a first ambiguous segment is identified and compared with a second ambiguous segment.If the two segments are classified as similar above a predefined threshold, a corrective recommendation is determined for the first segment that is suitable for the second ambiguous segment.

[0018] Fig. Figure 1 shows an example network diagram 100, which is used to describe the various disclosed embodiments. The example network diagram 100 includes a machine monitoring system (MMS) 130, a management server 140, a database 150, and a client device 160, which are connected via a network 110. The example network diagram 100 further includes a plurality of sensors 120-1 to 120-n (hereinafter referred to individually as a sensor 120 and collectively as sensors 120, where n is an integer equal to or greater than 1) which are connected to the machine monitoring system 130. The network 110 can be, but is not limited to, a wireless network, a cellular or wired network, a local area network (LAN), a wide area network (WAN), a metropolitan area network (MAN), the Internet, the World Wide Web (WWW), similar networks, and any combination thereof.

[0019] The client device 160 can be a personal computer, a laptop, a tablet computer, a smartphone, a wearable computing device, a protocol, a data source (e.g., a database), or any other device capable of receiving notifications and / or displaying maintenance and failure time predictions, results of supervised analysis, unsupervised analysis of machine operating data, and similar information.

[0020] The sensors 120 are located near (e.g., in physical proximity to) a machine 170. The machine 170 can be any machine whose performance can be represented by sensory data, such as an industrial machine used in industrial environments, but not limited to a turbine, motor, welding machine, 3D printer, injection molding machine, a combination thereof, a part thereof, and the like. Each sensor 120 is configured to detect sensory inputs such as, but not limited to, sound signals, ultrasonic signals, light, motion tracking indicators, temperature, energy consumption indicators, and the like, based on the operation of the machine 170. The sensors 120 can include, among other things, sound detection sensors, motion tracking sensors, energy consumption meters, temperature meters, and the like.Each of the sensors 120 can, but does not have to, be communicatively or otherwise connected to the machine 170 (such a connection is in . Fig. 1 not shown for the sake of simplicity and without limiting the disclosed embodiments).

[0021] The sensors 120 are connected to the machine monitoring system 130. The machine monitoring system 130 can be configured to store and preprocess the raw sensor inputs received from the sensors 120. Alternatively or additionally, the machine monitoring system 130 can be configured to periodically retrieve the collected sensor inputs, which are stored, for example, in the database 150. Preprocessing can include, among other things, data cleansing, normalization, rescaling, upscaling, reformatting, noise filtering, a combination thereof, and similar processes.

[0022] The management server 140 typically includes at least one processing circuit (not shown) and a memory (not shown), the memory containing instructions which, when executed by the processing circuit, configure the management server 140 as described below. According to one embodiment of the disclosure, the instructions stored in the memory are those which configure the system 100 to perform the procedure described below. The memory may also contain data collected by the sensors 120; however, such data may also be stored in a data storage device, such as a database 150, wherein, in certain embodiments, the memory of the management server 140 stores data and / or instructions therein or retrieves them from it.

[0023] In one embodiment, the management server 140 is configured to monitor at least one first machine behavior model of a machine (e.g., machine 170). The machine behavior model can be represented, for example, by a graph that aggregates a multitude of sensory inputs associated with a multitude of machine components and / or processes executed by a machine (e.g., machine 170). In another embodiment, the machine behavior model can be represented by metamodels, each metamodel being associated with a component of the machine. The metamodels are based on the indicative sensory inputs relating to their respective components and can be used to identify anomalies in the operation of each component of the machine.In another embodiment, the first machine behavior model can be divided into a plurality of segments. The segments can be defined by time frame, start and end point of at least one abnormal operating behavior of at least one component of the machine, as represented by the graph, etc.

[0024] In one embodiment, the management server 140 is configured to detect at least one first ambiguous segment in the at least one first machine behavior model. An ambiguous segment may contain features that, for example, have not been detected, determined, or analyzed in previous segments of the same machine or in similar machines. The ambiguous segment may represent abnormal behavior of at least one component of the machine. For example, the ambiguous segment may include the exceeding of a new threshold that has never been exceeded before, new behavioral patterns that have never occurred before, and similar phenomena.

[0025] In one embodiment, the management server 140 is configured to generate at least one notification based on the detection of the at least one first ambiguous segment. The at least one notification includes at least one query that can be generated in response to the detection of at least a part of the ambiguous segment. The query can contain at least one question, the answer to which can make it possible to determine a root cause for the formation of the unusual features or parameters of the first ambiguous segment. The root cause can be undesirable circumstances, such as an accumulation of gases within a specific component of a machine (e.g., machine 170). In another embodiment, the query can contain at least one question, the answer to which can narrow down the options for the formation of the unusual features or parameters of the first ambiguous segment.The management server 140 can be configured to send the notification to at least one client device (e.g., client device 160).

[0026] In one embodiment, the management server 140 is configured to monitor at least a part of a first machine behavior model with respect to at least one machine (e.g., machine 170). In another embodiment, the monitoring enables the generation of a variety of analyses related to the operation of the at least one machine or a component of the machine, such as anomalies, trends, energy consumption parameters, expected maintenance requirements, and the like. The behavior model consists of sensor inputs received from a variety of sensors (e.g., sensors 120) of a machine (e.g., machine 170).

[0027] The behavioral model can indicate at least one normal machine behavior, one abnormal machine behavior, one trend indicating an impending machine failure, one ambiguous machine behavior, and the like. Ambiguous behavior can be represented by parameters, values, sequences, and the like associated with at least one component of a machine (e.g., machine 170) that the management server 140 cannot classify or whose meaning or influence it cannot determine. The first machine behavioral model can contain a variety of segments. Each segment can be distinguished from other segments, for example, by time intervals, changes in the graph of the first machine behavioral model indicating increasing or decreasing values ​​above or below a certain threshold, and the like.

[0028] In one embodiment, the management server 140 is configured to determine at least one first ambiguous segment in the at least one first machine behavior model. The ambiguous segment can be represented by parameters, values, sequences, and the like, which are associated with at least one component of a machine (e.g., machine 170) that the management server 140 cannot classify, nor whose meaning or influence on the machine can be determined. The ambiguous segment represents unclear behavior of at least one component of machine 170. For example, an ambiguous segment of the first machine behavior model can contain a parameter value that is considered relatively high compared to average values ​​of that parameter. An ambiguous segment can, for example,This indicates downtime, a fault related to one or more machine components, and similar events that cannot be determined beyond a certain level of certainty. The level of certainty may refer to the presence of an ambiguous event or to a timeframe within which the ambiguous event occurred. For example, Management Server 140 may be configured to detect that downtime has occurred, but the correct timeframe of the downtime may not be unambiguous for Management Server 140. Determining at least one initial ambiguous segment can be achieved using at least one machine learning model.

[0029] In one embodiment, the management server 140 is configured to generate at least one notification containing at least one query based on the detection of the first ambiguous segment. In another embodiment, the management server 140 sends the notification to at least one client device (e.g., client device 160). In yet another embodiment, the notification can be sent to a log, a database, or the like. The notification can be an electronic message sent via email, text messaging (SMS), or the like. The query can contain textual and / or verbal elements. For example, the query can contain open-ended or closed-ended questions such as "Has a failure occurred?", "What are the symptoms?", and "What is the solution?". The query can be generated with respect to the ambiguous segment values.For example, after receiving a sequence of relatively low values ​​from the first machine behavior model, management server 140 can generate a query relating to the aforementioned sequence. Following the same example, the query might be: "Has any downtime occurred?"

[0030] In one embodiment, the management server 140 is configured to receive at least one input from a client device (e.g., client device 160) in response to a query. The input can be, for example, user feedback and can be entered by a user operating a client device (e.g., client device 160). In another embodiment, the input can be received from a log, a database, or the like. In yet another embodiment, the input can include a corrective solution recommendation, an answer to a closed or open question, a description of the root cause, confirmation of the machine learning algorithm's estimate regarding the ambiguous segment (the estimate can relate to the detection and / or prediction of one or more machine failures), and so on. The input can include, for example, a word, a sentence, a number, a part thereof, a combination thereof, etc.The input can be, for example, without restriction, "Yes," "No," "Increase in pressure gauge," "Open pressure valves," and the like. As an example, a query such as "Has a shutdown occurred?" is sent to a client device and displayed on a display unit (not shown) of the client device. The user's response to the query, e.g., "Yes" or "No," is then received at the management server 140. It should be noted that there can be multiple queries and / or a sequence of related inputs.

[0031] In one embodiment, the Management Server 140 is configured to update a machine learning algorithm, such as a deep learning model, based on the received input. This model is suitable for detecting abnormal behavior in a variety of machine behavior models, identifying patterns and / or trends that may indicate impending machine failures, and similar tasks. In another embodiment, the received input is used to adjust a deep learning reward function, resulting in a continuous improvement in the accuracy of the machine learning based on the received input.

[0032] It should be noted that if no input is received from client device 160, the management server 140 is configured to generate one or more corrective solution recommendations regarding the specific ambiguous segment. This is achieved, for example, by comparing the features of the ambiguous segment with one or more previous segments of one or more machine behavior models that were previously analyzed and determined. Following the same example, the comparison allows for the identification of a high degree of similarity between the features of the ambiguous segment and the previous segments, so that one or more corrective solution recommendations previously associated with the previous segments can also be associated with the ambiguous segment.

[0033] Fig. Figure 2 shows an example block diagram of the management server 140, configured according to one embodiment. The management server 140 comprises a processing circuit 210 coupled to a memory 220, a data storage 230, a network interface 240, and a machine learning (ML) unit 250. In one embodiment, the components of the machine failure predictor 140 are connected via a bus 260.

[0034] The processing circuit 210 can be implemented as one or more hardware logic components and circuits. Examples of hardware logic components that can be used include FPGAs (Field Programmable Gate Arrays), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), graphics processing units (GPUs), tensor processing units (TPUs), general-purpose microprocessors, microcontrollers, digital signal processors (DSPs), and the like, or other hardware logic components capable of performing calculations or other manipulations of information.

[0035] The memory 220 can be volatile (e.g., RAM), non-volatile (e.g., ROM or flash memory), or a combination thereof. In one embodiment, computer-readable instructions for implementing one or more embodiments disclosed herein can be stored in the data memory 230.

[0036] In one embodiment, the memory 220 is configured to store software. Software is to be understood in the broadest sense and means any type of instruction, whether it is called software, firmware, middleware, microcode, hardware description language, or otherwise. The instructions may also include code (e.g., in source code format, binary code format, executable code format, or any other suitable code format). When executed by the one or more processors, the instructions cause the processing circuit 210 to perform the various procedures described herein.

[0037] The data storage medium 230 can be a magnetic storage medium, an optical storage medium and the like, and can be implemented, for example, as flash memory or other storage technology, CD-ROM, Digital Versatile Discs (DVDs) or any other medium that can be used to store the desired information.

[0038] Network interface 240 enables the management server 140 to communicate with the machine monitoring system 130, e.g., via network 110, to receive raw and / or pre-processed sensor inputs. Additionally, network interface 240 allows the management server 140 to communicate with the client device 160 to send and receive inputs, etc.

[0039] The machine learning unit 250 is configured to perform machine learning based on sensor inputs received via the network interface 240, as further described herein. In one embodiment, the machine learning unit 250 is further configured to determine ambiguous segments in a machine behavior model of a machine, as described above. In another embodiment, the machine learning unit 250 is further configured to apply a deep learning model used to estimate a reward function, i.e., an input received from a client device. In another embodiment, the machine learning unit 250 is further configured to determine predictions for failures of the machine 170 based on one or more machine learning models.In another embodiment, the machine learning unit 250 is also configured to provide at least one recommendation, such as a corrective solution recommendation, to avoid or mitigate the identified predicted failures. For example, the recommendation may indicate that an exhaust pipe on machine 170 should be replaced with a new exhaust pipe in the near future to prevent failures.

[0040] It should be understood that the embodiments described here do not refer to those in Fig. 2. The specific architecture shown is limited, and other architectures can be used without deviating from the scope of the disclosed embodiments.

[0041] Fig. Figure 3 is an exemplary flowchart 300 illustrating a method for improving the accuracy level of a machine learning algorithm configured to monitor machine operation according to one embodiment. In one embodiment, the method can be performed by a management server, e.g., the management server 140. Fig. 1.

[0042] In S310, at least one initial machine behavior model of a first machine is monitored, for example, by a management server. This monitoring enables the generation of a wide range of analyses related to the operation of the at least one machine or a component of the machine. These analyses can include anomalies, trends, energy consumption parameters, expected maintenance requirements, and similar data.

[0043] In S320, at least one first ambiguous segment is determined in the first machine behavior model. The ambiguous segment represents unclear behavior of at least one component of the machine, which is represented by parameters, values, sequences, and the like, which, for example, cannot be classified by the machine or whose meaning or influence on the machine cannot be determined.

[0044] In S330, at least one notification containing at least one query is generated based on the determination of the first ambiguous segment. The notification can be customized to be sent to a specific client device (e.g., client device 160).

[0045] With S340, the notification is sent to a client device (e.g., client device 160). The notification can be in the form of an electronic message sent via email, SMS, MMS, internet-based messaging services, etc.

[0046] With S350, at least one input is received from a client device (e.g., client device 160) in response to the query. This input could be, for example, direct user feedback entered by a user with the client device.

[0047] In S360, a machine learning algorithm is updated based on at least one input. The machine learning algorithm can be, for example, a deep learning model configured to detect abnormal behavior in a variety of machine behavior models, identify patterns and / or trends that may indicate impending machine failures, and similar tasks associated with one or more machines.

[0048] Fig. Figure 4 is an exemplary flowchart 400, which represents a reinforcement learning-based method for automatically providing corrective solution recommendations for machine operation according to one embodiment.

[0049] In S410, an initial industrial machine behavior model associated with an initial industrial machine (e.g., machine 170) is monitored to identify and analyze an initial ambiguous segment. An industrial machine behavior model can be represented, for example, by a graph summarizing a variety of sensor inputs associated with a variety of components of the initial industrial machine and / or processes performed by the initial industrial machine. The initial ambiguous segment may contain features that have not been analyzed in previous segments of the same industrial machine behavior model or in similar types of industrial machine behavior models with similar features. For example, the initial ambiguous segment might include the exceeding of a predetermined threshold that was not previously exceeded, a new sequence of parameters that has never occurred before, and similar occurrences.The analysis of the first ambiguous segment may include the extraction of one or more features associated with the first ambiguous segment, such as parameters received from sensory inputs using the machine sensors of at least one component of the industrial machine at the time of the ambiguous segment.

[0050] In S420, an initial set of features is determined that relate to the first ambiguous segment. This initial set of features consists of parameters of at least one component of the first industrial machine at a specific point in time, for example, when an ambiguous segment is detected indicating unusual behavior of at least one component of the first industrial machine. Examples of such behavior include exceeding a predetermined threshold for one of the following parameters: an operating temperature, a rotational speed of a component of the industrial machine, various parameters that measure the productivity of the industrial machine, and the like.

[0051] In S430, a second ambiguous segment of a second industrial machine behavior model, which may be associated with the first industrial machine or with a second industrial machine, is monitored to identify and analyze this second ambiguous segment. A machine behavior model can be represented, for example, by a graph that aggregates a variety of sensory inputs associated with a variety of components of the first industrial machine and / or processes performed by the first industrial machine. The second ambiguous segment may contain features that were not analyzed in previous segments of the same industrial machine behavior model or in similar types of industrial machine behavior models with similar features. The analysis of the second ambiguous segment may involve extracting one or more features associated with the second ambiguous segment, such as...Parameters received from sensory inputs using sensors from at least one component of an industrial machine at the time of the second ambiguous segment.

[0052] In S440, a second set of features is determined that relate to the second ambiguous segment. This second set of features consists of parameters of at least one component of an industrial machine at a specific point in time, for example, when an ambiguous segment is detected indicating unusual behavior of at least one component of the industrial machine. Examples of such behavior include exceeding a predetermined threshold for one of the following parameters: an operating temperature, a rotational speed of a component of the industrial machine, various parameters that measure the productivity of the industrial machine, and the like.

[0053] In S450, it is determined whether the first set of features is similar to the second set of features above a predefined threshold. If so, execution continues with S460; otherwise, execution continues with S430. The threshold is used to distinguish similar ambiguous segments from dissimilar ambiguous segments. Similarity between two feature sets of a first and a second ambiguous segment can include, for example, similar sensor input values, similar starting points of the ambiguous segment, time frames, and the like. In one embodiment, the similarity determination can be performed using one or more machine learning methods, a deep learning method, and / or a statistical approach.In one embodiment, the determination can be carried out using a similarity function, which is a function that provides a quantitative value representing the similarity between the two groups of features.

[0054] In S460, at least one recommendation, such as a corrective solution recommendation previously determined with respect to the first ambiguous segment, is linked to the second ambiguous segment. The corrective solution recommendation can be retrieved, for example, from a database. In one embodiment, the previously determined recommendation can be received as input from a client device (e.g., client device 160) after a notification containing a query regarding the first ambiguous segment has been sent to the client device and user feedback regarding the first ambiguous segment has been received. In another embodiment, after receiving the recommendation, it is stored, for example, in a database and can be linked to the first set of features of the ambiguous segment to which the recommendation refers.In another embodiment, the machine learning algorithm of the at least one industrial machine behavior model is updated to link the corrective solution recommendation with the second ambiguous segment.

[0055] In the optional step S470, a notification regarding the corrective solution recommendation is sent to a client device. The recommendation, determined to be suitable for the second ambiguous segment based on similar characteristics, can be offered to a user to make changes in machine operation, thus preventing, for example, machine failure. In a further embodiment, S470 can also include adapting the recommendation based, for example, on the machine type, machine characteristics, the second set of characteristics of the at least one second segment, and the like.

[0056] Fig. Figure 5 is an exemplary flowchart 500 illustrating a reinforcement learning-based method for updating a machine learning algorithm configured to monitor machine operation according to an embodiment.

[0057] In S510, an initial ambiguous segment of a first machine behavior model is determined, indicating a presumed downtime. The presumed downtime can be determined based on ambiguous parameters from sensory inputs received by one or more sensors of the machine. Ambiguous parameters can be represented by unusual parameters whose meaning, i.e., their influence on machine operation, has not been determined.

[0058] In S520, an initial query is sent to a client device (e.g., client device 160), the response of which makes it possible to determine whether a downtime has occurred. In one embodiment, S520 further includes generating the initial query with respect to, for example, the first ambiguous segment features.

[0059] S530 determines whether downtime has occurred based on a response received from the client device. If so, execution continues with S540; otherwise, execution continues with S535. In one embodiment, S530 further includes analyzing an initial input, such as a user response, using one or more machine learning techniques to determine whether downtime has occurred.

[0060] If S535 detects that no downtime has occurred, the machine learning algorithm, which is configured to monitor machine operation and, in particular, to predict machine failures, is updated. This update can be achieved using the first input received from the client device in relation to the initial query.

[0061] In S540, a second query is sent to a client device (e.g., client device 160), the response of which allows verification of the downtime timeframe. In one embodiment, S540 further includes generating the second query upon receiving a positive user response to the first query.

[0062] S550 determines whether a downtime timeframe specified by management server 140 is correct, and if so, execution continues with S555; otherwise, execution continues with S560. This determination can be based on a second input, such as a response received from a client device (e.g., client device 160) to the second query. In one embodiment, S550 further includes analyzing the response, such as user feedback, using one or more machine learning techniques to determine whether the downtime timeframe initially determined by management server 140 is correct.

[0063] If the timeframe for the downtime is determined to be correct, the machine learning algorithm in S555, which is configured to monitor machine operation and specifically predict machine failures, is updated. This update can be performed using a second input received from the client device in response to the second query.

[0064] In S560, a third query, the response to which enables the determination of a correct failure timeframe, is sent to a client device (e.g., client device 160). In one embodiment, S560 further includes generating the third query in response to a negative user response to the second query.

[0065] At S570, after receiving a third input from the client device regarding the third query, the management server 140 updates the machine learning algorithm based on the received correct failure timeframe specified in the third input.

[0066] Fig. Figure 6 is an example simulation showing the representation of an ambiguous segment in a machine behavior model according to one embodiment. The in Fig.The simulation shown in Figure 6 contains a graph 610, which represents a machine behavior model as received by one or more sensors of the monitored machine. By analyzing graph 610, ambiguous segments such as segment 620 can be identified. An ambiguous segment may contain features that have not been identified, determined, or analyzed in previous segments of the same machine or in similar machines. For example, the ambiguous segment might contain the exceeding of a new threshold that has never been exceeded before, a new parameter sequence that has never occurred before, and similar features.

[0067] The various embodiments disclosed herein can be implemented as hardware, firmware, software, or any combination thereof. Furthermore, the software is preferably implemented as an application program tangibly embodied on a program storage unit or a computer-readable medium consisting of parts or specific devices and / or a combination of devices. The application program can be loaded onto and executed by a machine having any suitable architecture. Preferably, the machine is implemented on a computer platform comprising hardware such as one or more central processing units (“CPUs”), memory, and input / output interfaces. The computer platform may also include an operating system and microinstruction code.The various processes and functions described here can be part of the microinstruction code, part of the application program, or any combination thereof, that can be executed by a CPU, regardless of whether such a computer or processor is explicitly represented. Furthermore, various other peripheral units can be connected to the computer platform, such as an additional data storage unit and a printing unit. Additionally, a non-transient computer-readable medium is any computer-readable medium other than a transitory transmission signal.

[0068] As used herein, the phrase "at least one of" followed by a list of elements means that each of the listed elements can be used individually, or any combination of two or more of the listed elements can be used. For example, if a system is described as "at least one of A, B, and C," the system can contain A alone; B alone; C alone; A and B in combination; B and C in combination; A and C in combination; or A, B, and C in combination.

[0069] All examples and conditional language cited herein serve educational purposes to facilitate the reader's understanding of the principles of the disclosed embodiment and the concepts contributed by the inventor to the advancement of the prior art, and are to be interpreted as not constituting a limitation to these specifically cited examples and conditions. Furthermore, all statements contained herein that describe principles, aspects, and embodiments of the disclosed embodiments, as well as specific examples thereof, are to be understood as encompassing both structural and functional equivalents thereof. Moreover, it is intended that such equivalents include both currently known equivalents and those developed in the future, i.e., all developed elements that perform the same function regardless of their structure.

Claims

[1] Method for updating a machine learning algorithm of a computer system that actively monitors at least one industrial machine behavior model of at least one industrial machine, operating by providing additional training for the machine learning algorithm, comprising: - Monitoring of at least one industrial machine behavior model; - Identifying at least one first ambiguous segment of the at least one industrial machine behavior model that has a first set of features, and identifying a corrective solution recommendation associated with the first ambiguous segment; - Determining at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features, wherein the first ambiguous segment and the second ambiguous segment represent unclear behavior of at least one component of at least one industrial machine, and wherein the first ambiguous segment and the second ambiguous segment are each based on at least one obtained sensor reading of at least one sensor that is associated with the at least one industrial machine; - Determine whether the similarity between the first set of features and the second set of features exceeds a predetermined threshold; and - Updating the computer system's machine learning algorithm so that the corrective solution recommendation is assigned to the second ambiguous segment when the similarity exceeds the specified threshold. [2] Method according to claim 1, further comprising: Generating a notification regarding the corrective solution recommendation for the second ambiguous segment; and Sending the notification to a client device. [3] Method according to claim 1, wherein the first ambiguous segment indicates a presumed downtime of the at least one industrial machine. [4] Method according to claim 1, wherein determining that the similarity has exceeded the predetermined threshold is carried out using at least one of the following: a machine learning method, a deep learning model, a statistical approach and a similarity function. [5] Method according to claim 1, further comprising: Sending an initial query to a client device regarding the first ambiguous segment to determine if a detected downtime has occurred; and Determine whether downtime has occurred, based on a response to the initial query. [6] Method according to claim 5, further comprising: Update the machine learning algorithm when it is determined that no downtime has occurred. [7] Method according to claim 5, further comprising: Sending a second query to a client device to determine if a downtime timeframe is correct when downtime is detected; and Determine whether the downtime timeframe is correct, based on a Answer to the second query. [8] Method according to claim 7, further comprising: Update the machine learning algorithm when it is determined that the timeframe for the downtime is correct. [9] Method according to claim 7, further comprising: Sending a third query to a client device to determine an updated downtime timeframe if it is found that this downtime timeframe is incorrect; and Updating the machine learning algorithm with the updated timeframe. [10] Non-transitory computer-readable medium on which instructions are stored to cause a processing circuit to perform a method for updating a machine learning algorithm of a computer system that actively monitors at least one industrial machine behavior model of at least one industrial machine, operating by providing additional training for the machine learning algorithm, wherein the method comprises: - Monitoring at least the industrial machine behavior model of at least one industrial machine; - Identifying at least one first ambiguous segment of the at least one industrial machine behavior model that has a first set of features, and identifying a corrective solution recommendation associated with the first ambiguous segment; - Determining at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features, wherein the first ambiguous segment and the second ambiguous segment represent unclear behavior of at least one component of at least one industrial machine, and wherein the first ambiguous segment and the second ambiguous segment are each based on at least one obtained sensor reading of at least one sensor that is associated with the at least one industrial machine; - Determine whether the similarity between the first set of features and the second set of features exceeds a predetermined threshold; and - Updating the computer system's machine learning algorithm so that the corrective solution recommendation is assigned to the second ambiguous segment when the similarity exceeds the specified threshold. [11] System for updating a machine learning algorithm of a computer system that actively monitors at least one industrial machine behavior model of at least one industrial machine, operating by providing additional training for the machine learning algorithm, comprising: - a processing circuit; and - a memory, wherein the memory contains instructions which, when executed by the processing circuit, set up the system: - to monitor at least one industrial machine behavior model of at least one industrial machine; - to identify at least one first ambiguous segment of the at least one industrial machine behavior model that has a first set of features, and to identify a corrective solution recommendation associated with the first ambiguous segment; - to determine at least one second ambiguous segment of the at least one industrial machine behavior model, which has a second set of features, wherein the first ambiguous segment and the second ambiguous segment represent unclear behavior of at least one component of at least one industrial machine, and wherein the first ambiguous segment and the second ambiguous segment are each based on at least one obtained sensor reading of at least one sensor that is associated with the at least one industrial machine; - to determine whether a similarity between the first set of features and the second set of features exceeds a predetermined threshold; and - to update the computer system's machine learning algorithm to assign the corrective solution recommendation to the second ambiguous segment when the similarity exceeds the specified threshold. [12] System according to claim 11, wherein the system is further configured as follows: to generate a notification regarding the corrective solution recommendation for the second ambiguous segment; and to send the notification to a client device. [13] System according to claim 11, wherein the first ambiguous segment indicates a presumed downtime of the at least one industrial machine. [14] System according to claim 11, wherein determining that the similarity has exceeded the predetermined threshold is carried out using at least one of the following: a machine learning method, a deep learning model, a statistical approach and a similarity function. [15] System according to claim 11, wherein the system is further configured as follows: to send an initial query to a client device regarding the first ambiguous segment to determine whether a detected downtime has occurred; and to determine, based on a response to the initial query, whether downtime has occurred. [16] System according to claim 15, wherein the system is further configured as follows: to update the machine learning algorithm when it is determined that no downtime has occurred. [17] System according to claim 15, wherein the system is further configured as follows: to send a second query to a client device to determine if a downtime timeframe is correct if downtime is detected; and to determine whether the timeframe for the downtime is correct, based on a response to the second query. [18] System according to claim 17, wherein the system is further configured as follows: to update the machine learning algorithm when it is determined that the timeframe for the downtime is correct. [19] System according to claim 17, wherein the system is further configured as follows: to send a third query to a client device to determine an updated downtime timeframe if it is found that this downtime timeframe is incorrect; and to update the machine learning algorithm with the updated timeframe.

Images