System and method for condition-based monitoring

Abstract

Systems and methods for condition-based monitoring are disclosed, including obtaining first visual data indicative of an operation of an industrial automation equipment, extracting a plurality of features from the first visual data, and determining a subset of the plurality of features of the first visual data, wherein the subset of the plurality of features is indicative of a change in a condition of the operation of the industrial automation equipment. The method further includes obtaining second visual data indicative of the operation of the industrial automation equipment, detecting, using the second visual data, the change in the condition of the operation of the industrial automation equipment based on the subset of the plurality of features, and generating an output to control the operation of the industrial automation equipment based on the detected change in the condition.

Description

Technical Field

[0001] This disclosure generally relates to condition monitoring systems, and more specifically, to condition monitoring systems used to predict changes in processes and functions of industrial operations. Background Technology

[0002] Typically, control systems are integrated into industrial operations to monitor automated processes and functions to determine whether industrial operations are proceeding as expected. Control systems can be used to correct and / or flag deviations from standard operating procedures. Such control can prevent large-scale deviations from standard operating procedures. Previously available control systems monitor changes in data collected from associated equipment, sensor feedback, user observations, or a combination thereof. While monitoring changes in the collected data can be used to control large-scale deviations from standard operating procedures, such responses based on changes in collected data are passive. Passive responses to changes in industrial operations can limit operational efficiency due to downtime, equipment impacts, etc., resulting from detecting deviations from standard operating procedures. Therefore, new technologies are needed to predict changes in the processes and functions of industrial operations.

[0003] This section is intended to introduce the reader to various aspects of the art that may relate to the aspects of the present technology described below and / or claimed. This discussion is intended to help provide the reader with background information to facilitate a better understanding of the various aspects of this disclosure. Therefore, it should be understood that these statements are to be read in this context and not as an admission of prior art. Summary of the Invention

[0004] The following provides an overview of some of the embodiments disclosed herein. It should be understood that these aspects are presented merely to provide the reader with a brief overview of these certain embodiments, and these aspects are not intended to limit the scope of this disclosure. In fact, this disclosure may cover many aspects that may not be set forth below.

[0005] In one embodiment, this disclosure relates to a method comprising: acquiring first visual data indicating the operation of industrial automated equipment; extracting a plurality of features from the first visual data; and determining a subset of the plurality of features of the first visual data, wherein the subset of features indicates a change in the operating condition of the industrial automated equipment. The method further comprises: acquiring second visual data indicating the operation of the industrial automated equipment; using the second visual data, detecting changes in the operating condition of the industrial automated equipment based on the subset of features; and generating an output based on the detected changes in the operating condition to control the operation of the industrial automated equipment.

[0006] In one embodiment, this disclosure relates to a system comprising: one or more camera devices configured to capture visual data associated with the operation of industrial automation equipment; a processing circuitry; and a memory accessible by the processing circuitry. The memory stores instructions that, when executed by the processing circuitry, cause the processing circuitry to perform operations including: acquiring visual data from one or more camera devices; and acquiring process data associated with the operation of the industrial automation equipment. The operations further include: detecting changes in the operating condition of the industrial automation equipment based on a subset of features of the visual data via a clustering engine; extracting one or more tags from the process data associated with the detected changes in the operating condition of the industrial automation equipment; and generating an output based on the one or more tags associated with the detected changes in the operating condition to control the operation of the industrial automation equipment.

[0007] In one embodiment, this disclosure relates to a non-transitory computer-readable medium comprising computer-executable instructions configured, when executed, to cause a processor to perform operations including: receiving first visual data indicating the operation of one or more industrial automated equipment; extracting a plurality of features from the first visual data; and determining a subset of the plurality of features of the first visual data, wherein the subset of features indicates a change in the operating condition of one or more industrial automated equipment. The operations further include: receiving second visual data indicating the operation of one or more industrial automated equipment; using the second visual data, detecting a change in the operating condition of one or more industrial automated equipment based on a subset of the plurality of features; and generating an output based on the detected change in the operating condition to control the operation of one or more industrial automated equipment. Attached Figure Description

[0008] These and other features, aspects, and advantages of this disclosure will be better understood when the following detailed description is read with reference to the accompanying drawings, in which similar reference numerals denote similar parts throughout the drawings:

[0009] Figure 1 An exemplary industrial automation system adopted by a food manufacturer according to various aspects of the present technology is shown;

[0010] Figure 2 A graphical representation of an exemplary control and condition monitoring system that can be employed in any suitable industrial automation system according to various aspects of the present technology is shown;

[0011] Figure 3This is a flowchart illustrating the process of extracting features from visual data to generate a baseline indicating the operation of industrial automation equipment, based on various aspects of this technology.

[0012] Figure 4 This is a flowchart illustrating the process of extracting features from visual data to control the operation of industrial automation equipment, based on various aspects of this technology.

[0013] Figure 5 It is a flowchart illustrating the process of associating one or more markers within the prediction process data of this technology with the formation of one or more new clusters; and

[0014] Figure 6 Based on various aspects of this technology Figure 2 An illustrative implementation of the graphical user interface (GUI) generated within the condition monitoring system. Detailed Implementation

[0015] One or more specific embodiments of this disclosure will now be described. To provide a concise description of these embodiments, not all features of an actual implementation may be described in the specification. It should be understood that, as in any engineering or design project, many implementation-specific decisions must be made to achieve the developer's specific goals, such as compliance with system-related and business-related constraints, which may vary depending on the implementation. Furthermore, it should be understood that while such development efforts may be complex and time-consuming, they remain routine tasks of design, production, and manufacturing for those skilled in the art who benefit from this disclosure.

[0016] When describing the elements of various embodiments of this disclosure, the articles “a,” “an,” “the,” and “described” are intended to mean the presence of one or more elements. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that other elements may be present in addition to those listed.

[0017] Furthermore, as used herein, the terms “real-time,” “real-time (of),” or “near real-time” are used interchangeably and are intended to describe operations performed without any human-perceptible interruptions between operations (e.g., computational operations). For example, as used herein, data associated with the system described herein may be collected, transmitted, and / or used to control computations in “near real-time,” such that during system operation (e.g., while the system is running), data read, data transmission, and / or data processing steps occur once per second, once every 0.1 seconds, once every 0.01 seconds, or even more frequently.

[0018] As mentioned above, control systems can be integrated into industrial operations to monitor automated processes and functions to determine whether industrial operations are proceeding as expected, and can be used to correct deviations from standard operating procedures. Monitoring changes in collected data can be used to control large-scale deviations from standard operating procedures; however, such responses based on changes in collected data can be passive. Such passive responses may limit the operational efficiency of industrial operations due to downtime, equipment impact, etc., resulting from detecting deviations from standard operating procedures. Therefore, new technologies are needed to predict changes in the processes and functions of industrial operations.

[0019] Therefore, this paper discloses a condition monitoring system that provides predictive monitoring capabilities for the processes and functions of an enterprise's industrial operations. The condition monitoring system utilizes visual data (e.g., video data, image data, video, video streams) collected during industrial operations to predict changes in the operation (e.g., changes in operating status) of one or more components of an industrial automation system. Predictions of operational changes can be based on extracting one or more features from the visual data using the condition monitoring system's feature extraction engine. The feature extraction engine can extract features associated with specific operations of the industrial automation system. In this way, the feature extraction engine can be used to reduce the dimensionality of video data while still being able to predict operational changes of components in the industrial automation system. For example, after analyzing video data, the condition monitoring system can extract multiple features; however, a subset of these features can be used to predict operational changes. This subset of features can be compared with one or more thresholds, process data, historical data, etc., to predict changes in specific operations of the industrial automation system. The identification and monitoring of the subset of features can provide finer-grained and proactive control over the industrial operation of the industrial automation system.

[0020] In some implementations, the condition monitoring system may use artificial intelligence (AI) techniques, including vision (e.g., visual data), as sensors to determine a subset of features. The subset of features identified by AI techniques (e.g., neural networks, machine learning) includes features that are not readily apparent to the operating user. In some implementations, the AI ​​techniques use a feature extraction engine to identify the subset of features. The feature extraction engine may include an unsupervised model configured to process unlabeled video data, unlabeled process data, unlabeled historical data, etc. As a non-limiting example in an industrial context, the feature extraction engine of the condition monitoring system may identify a first feature associated with off-axis motion of a first component of a piece of equipment, and a second feature associated with the operating speed of a second component of the equipment.

[0021] The condition monitoring system can monitor combined changes in a first and a second characteristic to predict changes in the operation of the equipment. These combined changes in characteristics provide subtle identification of operational changes that cannot be provided using changes in operational data (e.g., vibration data, current data, drive data, etc.). The condition monitoring system can proactively provide notifications to the control system and / or the operator. The condition monitoring system can directly control the control system to control the operation of the equipment, thereby improving operational efficiency by preventing operational changes from impacting the industrial automation system. Furthermore, by identifying specific characteristics, the condition monitoring system reduces video data processing costs by reducing data dimensionality, thus improving the efficiency of computing resources. Additionally, this embodiment includes a graphical user interface (GUI) designed to provide insights into characteristic changes used to predict operational changes in industrial automation equipment in a concise and organized format.

[0022] In some implementations, the condition monitoring system can provide features extracted by a feature extraction engine to a clustering engine. The clustering engine can generate a baseline indicating the nominal operation of industrial automation equipment. The clustering engine can receive features extracted from video data of the industrial automation equipment and perform stream clustering on data extracted from one or more incoming frames of the video data. The clustering engine can assign incoming frames to one or more clusters. Clusters can be used to determine whether the incoming features of the video data fall within a threshold distance (e.g., threshold deviation) of the baseline. In some cases, these clusters can be formed in near real-time, making it possible to identify deviations relative to the baseline. In some implementations, clusters formed based on incoming frames of video data may deviate from the baseline generated by the clustering engine. Therefore, the operation of the associated industrial automation equipment can be flagged as deviating from nominal operation.

[0023] In some implementations, the condition monitoring system can be used to identify one or more markers that cause the generation of clusters deviating from a baseline, which indicates the nominal operation of industrial automation equipment. These markers can be identified by analyzing process data (e.g., input data from one or more sensors associated with the industrial automation equipment) corresponding to the formation of new clusters (which are associated with deviations from the baseline). In some implementations, the condition monitoring system can be used to identify markers using an automatic correlation between process data and the occurrence of new clusters, obtained using a transformer architecture (e.g., an AI-based encoder-decoder model). For example, a transformer architecture can be used to isolate attention points in process data associated with the occurrence of new clusters from video data. In some cases, attention points can be labeled as one or more markers associated with a subset of process data that causes the operation of the industrial automation equipment to deviate from the baseline. In this way, the condition monitoring system can predict changes in the operation of the industrial automation equipment based on analyzing video data and identifying the formation of one or more new clusters associated with predetermined markers. Such predictions enable proactive control of the industrial operation of the industrial automation system.

[0024] As an introduction, Figure 1 An example industrial automation system 10 used by a food manufacturer is shown. The embodiments described herein can be implemented using various devices shown in the industrial automation system 10 described below. However, it should be noted that although... Figure 1 The example industrial automation system 10 is designed for food manufacturers, but the embodiments described herein can be used in any suitable industry (e.g., transportation, mining, hydrocarbon production, manufacturing, etc.). A brief description of the example industrial automation system 10 used by food manufacturers is provided below to facilitate a more comprehensive understanding of how the embodiments described herein can be applied to industrial equipment to significantly improve the operation of the corresponding industrial automation system. Therefore, the embodiments described herein should not be limited to applications in food manufacturing. Figure 1 The example depicted in the text.

[0025] Now refer to Figure 1 An example industrial automation system 10 for a food manufacturer may include silos 12 and tanks 14. Silos 12 and tanks 14 may store different types of raw materials, such as grains, salt, yeast, sweeteners, flavorings, colorings, vitamins, minerals, and preservatives. In some embodiments, sensors 16 may be positioned inside or around silos 12 and tanks 14, or at other suitable locations within or around the industrial automation system 10, to measure certain properties such as temperature, mass, volume, pressure, humidity, etc.

[0026] Raw materials can be supplied to mixer 18, which mixes the raw materials together according to a specified ratio. Mixer 18 and other machines in industrial automation system 10 can employ certain industrial automation equipment 20 to control the operation of mixer 18 and other machines. Industrial automation equipment 20 may include controllers, input / output (I / O) modules, motor control centers, motors, human-machine interfaces (HMIs), operator interfaces, contactors, starters, sensors 16, actuators, conveyors, drives, relays, protection devices, switching devices, compressors, sensors, actuators, firewalls, network switches (e.g., Ethernet switches, modularly managed, fixedly managed, service routers, industrial, unmanaged, etc.), etc.

[0027] Mixer 18 can provide a mixed compound to depositor 22, which can deposit a certain amount of the mixed compound onto conveyor 24. Depositor 22 can deposit the mixed compound onto conveyor 24 according to the shape and amount that can be specified to a control system of depositor 22. Conveyor 24 can be any suitable conveyor system of various types of machines that transport articles to industrial automation system 10. For example, conveyor 24 can transport the deposited material from depositor 22 to oven 26, which can bake the deposited material. The baked material can be transported to cooling aisle 28 to cool the baked material, so that the cooled material can be transported via conveyor 24 to pallet loader 30. Pallet loader 30 can include a machine that receives a certain amount of the cooled material for packaging. As an example, pallet loader 30 can receive 25 ounces of cooled material, which can correspond to the amount of grain provided in a grain bin.

[0028] Pallet packer 32 can receive a certain amount of cooled material collected from pallet loader 30 into a bag, which can be sealed. Pallet packer 32 can receive a certain amount of cooled material collected into a bag and seal the bag using appropriate machinery. Conveyor 24 can transport the bagged material to case packer 34, which can pack the bagged material into boxes. The boxes can be transported to palletizer 36, which can stack a certain number of boxes on a pallet that can be lifted by forklifts or the like. The stacked boxes can then be transported to shrink wrapper 38, which can shrink wrap the stacked boxes while they are on the pallet to keep them together. The shrink-wrapped boxes can then be transported to a warehouse, etc., by forklift or other suitable transport vehicle.

[0029] To perform the operation of each device in the equipment of the example industrial automation system 10, the industrial automation equipment 20 may provide power to the machine used to perform certain tasks, provide surge protection to the machine, prevent injury to the human operator in the industrial automation system 10, monitor the operation of the corresponding equipment, and transmit data about the corresponding equipment to the supervisory control system 40, etc. In some embodiments, a local control system 42 may be used to control each industrial automation equipment 20 or a group of industrial automation equipment 20. The local control system 42 may include receiving data from a condition monitoring system regarding the operation of the corresponding industrial automation equipment 20, other industrial automation equipment 20, user input, and other suitable inputs to control the operation of the corresponding industrial automation equipment 20.

[0030] One or more sensors 16 can be any number of devices suitable for providing process data (e.g., information about process status), historical data, etc. Sensors 16 can be used to operate the industrial automation equipment of the industrial automation system 10 in real time and / or for a predetermined period of time. In practice, process data can be utilized within process loops monitored and controlled by a condition monitoring system and / or an HMI. Such process loops can be activated based on process data (e.g., input from sensors 16) or direct operator input received via the HMI. Sensors 16 and one or more actuators communicate with the HMI 52 and the industrial automation equipment. Furthermore, sensors 16 and actuators can receive power from one or more additional attachment modules.

[0031] Figure 2 A schematic representation of an exemplary local control system 42 according to the embodiments presented herein is shown, which includes a condition monitoring system 50 that can be employed in any suitable industrial automation system 10. Figure 2 In the diagram, the local control system 42 is shown to include a condition monitoring system 50, a human-machine interface (HMI) 52, one or more databases 54, and a network 56. The local control system 42 is adapted to interface with devices capable of monitoring and controlling various types of industrial automation equipment 58. As an example, the industrial automation equipment 58 may include a mixer 18, a sedimentator 22, a conveyor 24, an oven 26, etc. Figure 1 Other machines or any other suitable equipment described herein.

[0032] In some implementations, one or more attributes of the industrial automation equipment 58 can be monitored and controlled by specific equipment that adjusts control variables used to operate the industrial automation equipment 58. For example, Figure 1The sensor 16 shown can monitor various attributes of the industrial automation equipment 58 and provide the data to the local control system 42, which can adjust the operation of the industrial automation equipment 58 accordingly. In some cases, the industrial automation equipment 58 can be associated with devices used by other equipment. For example, scanners, meters, valves, flow meters, etc., can be installed on the industrial automation equipment 58. Here, the industrial automation equipment 58 can receive data from the associated devices and use the data to perform their respective operations more efficiently. For example, the controller of a motor drive can receive data about the temperature of the connected motor and adjust the operation of the motor drive based on this data.

[0033] As described above, industrial automation equipment 58 can take many forms and includes devices for a wide variety of purposes. For example, industrial automation equipment 58 may include machinery for performing various operations in compression stations, oil refineries, batch operations for food manufacturing, mechanized assembly lines, etc. Therefore, industrial automation equipment 58 may include various operating components, such as electric motors, valves, actuators, temperature elements, pressure sensors, or a large number of machines or devices for manufacturing, processing, material handling, and other applications.

[0034] Additionally, industrial automation equipment 58 may include various types of equipment that can be used to perform various operations (which may be part of an industrial application). For example, industrial automation equipment 58 may include electrical equipment, hydraulic equipment, compressed air equipment, steam equipment, machine tools, protective equipment, refrigeration equipment, power lines, hydraulic lines, steam lines, etc. Some example types of equipment may include mixers, machine conveyors, tanks, skids, dedicated original equipment manufacturer (OEM) machines, etc. Besides the equipment mentioned above, industrial automation equipment 58 may also include... Figure 1 The system comprises certain industrial automation equipment 20, which may include controllers, input / output (I / O) modules, motor control centers, motors, human-machine interfaces (HMIs), operator interfaces, contactors, starters, sensors 16, actuators, drivers, relays, protection devices, switching devices, compressors, firewalls, network switches (e.g., Ethernet switches, modularly managed, fixedly managed, service routers, industrial, unmanaged, etc.), etc.

[0035] In some embodiments, industrial automation equipment 58 may include a communication component that enables the industrial automation equipment 58 to transmit data between the industrial automation equipment 58 and HMI 52 and / or condition monitoring system 50. The communication component may include a network interface that enables the industrial automation equipment 58 to communicate via network 56 through various protocols such as Ethernet / IP®, Control Network®, Device Network®, or any other industrial communication network protocol. Alternatively, the communication component may enable the industrial automation equipment 58 to communicate via various wired or wireless communication protocols such as Wi-Fi, mobile telecommunications technologies (e.g., 2G, 3G, 4G, 5G, LTE), Bluetooth®, near-field communication technologies, etc.

[0036] The condition monitoring system 50 may include one or more camera devices 62, one or more sensors 64, a controller 66, input / output (I / O) ports 68, a processor 70, a memory 72, a communication circuit system 74, a display 76, an image processing circuit system 78, one or more machine learning engines 80, etc. One or more camera devices 62 may include infrared cameras, CCD cameras, DSLR cameras, SLR cameras, mirrorless cameras, or one or more digital cameras. One or more camera devices 62 may include complementary metal-oxide-semiconductor (CMOS) cameras, charge-coupled devices (CCDs), electron-multiplying charge-coupled devices (EMCCDs), InGaAs detectors, one or more photodiodes, pyroelectric sensors, one or more photodetectors, and / or other suitable detectors. The camera devices 62 can capture visual data in one or more regions of the electromagnetic spectrum. For example, camera device 62 may include capturing visual data in the ultraviolet, visible, and / or infrared (IR) regions (e.g., short-wave, mid-wave, and / or long-wave IR). Camera device 62 may collect visual data ranging from 157 nm to 220 nm, 200 nm to 400 nm, 400 nm to 600 nm, 480 nm to 880 nm, 880 nm to 1800 nm, 2000 nm to 3000 nm, 3000 nm to 5000 nm, and / or 8000 nm to 14000 nm. Camera device 62 may collect visual data. Visual data may include video data (e.g., digital representations of moving images and / or audio), image data (e.g., digital representations of one or more still images), or combinations thereof. Visual data may be analyzed, enabling the use of vision as a sensor. Visual data may be analyzed to extract features that are not apparent to the operating user. In some embodiments, AI technology identifies a subset of features in the visual data to determine operating parameters of industrial automation equipment 58. In some implementations, a first camera may be positioned at a first angle, and a second camera may be positioned at a second angle to capture video data from the industrial automation equipment 58 from various angles. The first camera may be of a first type (e.g., an IR camera), and the second camera may be of a second type (e.g., an RBG camera). Thus, camera 62 can capture visual data in various spectral regions, which is provided to the condition monitoring system 50 for further analysis.

[0037] One or more sensors 64 can be any number of devices suitable for providing information about the operating status of industrial automation equipment 58. Sensors 64 may include one or more temperature sensors, pressure sensors, motion sensors, etc. Controller 66 may be configured to receive data from camera device 62 and / or sensors 64 and control one or more additional components of HMI 52, industrial automation equipment 58, and / or condition monitoring system 50.

[0038] I / O port 68 can be an interface capable of being coupled to HMI 52, industrial automation equipment 58, and other peripheral components such as input devices (e.g., keyboard, mouse), sensors, input / output (I / O) modules, etc. Communication circuitry 74 can facilitate wired or wireless communication between various components of condition monitoring system 50 and with external devices (e.g., mobile devices, HMI, etc.). Processor 70 can be any suitable type of computer processor or microprocessor capable of executing computer-executable code. Furthermore, processor 70 can include multiple microprocessors, one or more "general-purpose" microprocessors, one or more application-specific microprocessors, and / or one or more application-specific integrated circuits (ASICs) or some combination thereof. For example, processor 70 can include one or more Reduced Instruction Set Computing (RISC) or Complex Instruction Set Computing (CISC) processors.

[0039] The memory 72 of the condition monitoring system 50 can also be used to store data executed by the processor 70, various other software applications, etc. The memory 72 can represent a non-transitory computer-readable medium (e.g., any suitable form of memory or storage device) that can store processor-executable code used by the processor 70 to perform the various techniques described herein. The processor 70 can be configured to execute instructions. When executed by the processor 70, these instructions can enable the condition monitoring system 50 to control the camera device 62 and / or sensor 64 to provide data for feature extraction. The condition monitoring system 50 may include a display 76 configured to display text or images transmitted from the processor 70 to the display. In addition to and / or as alternatives to the display 76, other devices for user interaction, such as lights (e.g., LEDs), speakers, etc., may be included.

[0040] The condition monitoring system 50 may include video data prepared by an image processing circuitry system 78 (e.g., an image processing circuitry system within processor 70) or uncompensated video data. The image processing circuitry system 78 may work in conjunction with engine 80 to analyze data used to monitor the operation of industrial automation equipment 58. In some embodiments, the image processing circuitry system 78 may preprocess visual data generated by camera device 62. Preprocessing may include cropping, rotating, applying one or more masks, etc., of the visual data. Engine 80 may include a feature extraction engine 82, a clustering engine 84, and / or one or more additional engines. Engine 80 may use AI technologies, including neural networks, machine learning models, etc. Engine 80 may use various AI-based algorithms to analyze video data obtained from camera device 62, sensors 16, 64, etc. Engine 80 may utilize one or more predictive models to analyze video data, process data, etc. Various types of predictive models may be used to analyze data from various sources and generate predictive outputs. Engine 80 may include various types of artificial neural networks (ANNs), such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), etc. The condition monitoring system 50 can also communicate with the database 54, which can store information associated with the sensor 16, relevant process data, etc.

[0041] In some implementations, unsupervised machine learning techniques can be used to train the feature extraction engine 82, i.e., training the prediction model using training data that includes input data (e.g., an unlabeled dataset). In some cases, partially unsupervised machine learning techniques can be used to train the feature extraction engine 82, i.e., training the prediction model using training data that includes input data with a first part (e.g., an unlabeled dataset) without the desired predicted output and a second part (e.g., a labeled dataset) with the desired predicted output. The feature extraction engine 82 can use unsupervised models to process unlabeled visual data, unlabeled process data, unlabeled historical data, etc. The feature extraction engine 82 can use principal component analysis (PCA), kernel PCA, linear discriminant analysis (LDA), generalized discriminant analysis (GRA), T-distributed random neighbor embedding (t-SNE), nonnegative matrix factorization (NMF), etc. For example, the feature extraction engine 82 can receive unlabeled visual data from the camera device 62. The feature extraction engine 82 can use one or more filters to filter the unlabeled visual data. Filters can be used to convert unlabeled visual data into grayscale images, generate depth maps for visual data, etc. The feature extraction engine 82 can vectorize visual data, apply one or more transforms, etc.

[0042] Feature extraction engine 82 can extract one or more features from visual data associated with a specific operation of industrial automation equipment 58 to generate visual data with reduced dimensionality. For example, feature extraction engine 82 may include an unsupervised model configured to process unlabeled video data, unlabeled process data, unlabeled historical data, etc. For example, feature extraction engine 82 can identify various features extracted from visual data collected by camera device 62 that are associated with temporal features, one or more spatial features, or combinations thereof.

[0043] Clustering engine 84 can classify features extracted from visual data by feature extraction engine 82 to generate a baseline indicating the nominal operation of industrial automation equipment 58. The features provided to clustering engine 84 can be unlabeled, thus allowing clustering engine 84 to group one or more clusters based on the similarity between data points. These one or more clusters can be associated with the baseline. Therefore, clustering engine 84 can receive additional visual data during the operation of industrial automation equipment 58 and analyze deviations from the baseline based on the formation of one or more new clusters.

[0044] Figure 3 This is a flowchart of process 100 according to various aspects of the present technology, which is used to generate one or more extracted features 102 from visual data 104 (e.g., image data, video data, or other data including a series of images) to generate a baseline 106 indicating the operation of industrial automation equipment 58. Process 100 can be described by referring to the above. Figure 1 and Figure 2 The disclosed condition monitoring system 50's engine 80 (e.g., feature extraction engine 82 and / or clustering engine 84), computing device and / or HMI 52, or any other suitable computing device or controller, is used to execute the process. Furthermore, the blocks of process 100 may be executed in the order disclosed herein or in any suitable order. For example, certain blocks of process 100 may be executed simultaneously. Additionally, in some embodiments, at least one block of process 100 may be omitted.

[0045] At block 108 of process 100, condition monitoring system 50 may receive visual data 104 and extract one or more features from visual data 104 using feature extraction engine 82. Visual data 104 may be generated by camera device 62 of condition monitoring system 50. Visual data 104 may include preprocessed (e.g., cropped, rotated, masked, etc.) video data. Feature extraction engine 82 may perform feature extraction and output one or more features associated with visual data 104. Feature extraction engine 82 may include an unsupervised machine learning model that can be used to extract features from unlabeled visual data (e.g., visual data 104). In some embodiments, the features extracted by feature extraction engine 82 may be generated from visual data from multiple camera devices 62. For example, visual data 104 may include data on the operation of industrial automation equipment 58 from one or more angles, one or more spectral regions, etc. As a specific, non-limiting example, visual data 104 may include streaming video data acquired by camera device 62, which is arranged to capture video data of material deposited on conveyor 24. As another specific, non-limiting example, visual data 104 may include a series of infrared images acquired by camera device 62 and / or sensor 64, which is arranged to detect gas molecules (e.g., water or carbon dioxide in the baking context) formed or emitted from the material being baked within oven 26. Feature extraction engine 82 can extract features from visual data 104 that can be used to determine baseline 106. Feature extraction engine 82 can determine these features in an unsupervised manner (e.g., without input from the operator) and therefore can be able to identify features that the operator might not otherwise monitor or recognize. That is, while a user might view the visual data and monitor the overall color change of the bread being baked, feature extraction engine 82 can instead identify specific subsets of pixels that correspond to areas where the oven applies heat less evenly but also serve as better visual indicators of the baking process. Processing a subset of the data corresponding to this feature (e.g., a subset of pixels) uses fewer computational resources and thus improves the efficiency of monitoring industrial automation processes by enabling faster identification of conditions.

[0046] At box 110, the condition monitoring system 50 can determine a subset of extracted features associated with changes in the operating condition of the industrial automation equipment 58. This subset of extracted features can be identified by the feature extraction engine 82. For example, the feature extraction engine 82 may include an unsupervised machine learning model that can be used to perform dimensionality reduction while preserving features of the visual data associated with the operation of the industrial automation equipment 58. Therefore, the feature extraction engine 82 can process the visual data 104 to remove one or more redundant features, noise, etc. In some embodiments, the feature extraction engine 82 can determine a subset of extracted features associated with one or more key operations of the industrial automation equipment 58. The subset of extracted features (e.g., extracted feature 102) may include one or more temporal features, one or more spatial features, etc. In some embodiments, the feature extraction engine 82 can update the extracted feature 102 identified as associated with changes in operating condition. For example, a change in the operating settings of a particular piece of equipment may affect the extracted feature 102 identified by the feature extraction engine 82. Therefore, the feature extraction engine 82 can dynamically provide updates to the dimensionality reduction and feature extraction of the visual data.

[0047] At block 112 of process 100, condition monitoring system 50 receives extracted features 102 and uses clustering engine 84 to generate a baseline 106 indicating the nominal operation (e.g., normal operation, expected operation) of industrial automation equipment. In some embodiments, the generation of baseline 106 may be based on unlabeled clusters generated by clustering engine 84. For example, clustering engine 84 may receive extracted features 102 from visual data 104 and divide the extracted features 102 into one or more clusters (e.g., groups of data points) based on the distance between one or more centroids of each group. The centroid (e.g., cluster center) may be the mean or median of all data points within the corresponding cluster. Clustering engine 84 may generate clusters iteratively to minimize the distance between data points, thereby generating one or more distinct clusters. Clusters may be unlabeled, meaning that a cluster may represent a specific characteristic within the extracted features 102, but such a characteristic may not have been identified.

[0048] As a non-restrictive example, consider Figure 1The shrink wrapper 38 in the system can be monitored by visual data generated by the condition monitoring system 50. The visual data can be analyzed by a feature extraction engine 82, and one or more extracted features can be identified as related to the operation of the shrink wrapper 38. A clustering engine 84 can receive the extracted features indicating the operation of the shrink wrapper 38 and divide the extracted features into one or more unlabeled clusters. Unlabeled clusters can be defined using a similarity threshold based on the distance of each data point to the centroid of a particular unlabeled cluster. In some cases, the unlabeled clusters can form a baseline indicating the nominal operation of the shrink wrapper 38. The condition monitoring system 50 can output a baseline 106 including the unlabeled clusters. The baseline 106 can be used to establish normal operating parameters for the industrial automation equipment 58.

[0049] In some implementations, clustering engine 84 can form one or more tagged clusters. At block 114 of process 100, condition monitoring system 50 can receive process data 116 and generate one or more context labels for the one or more clusters formed at block 112. Process data 116 may include data generated from industrial automation equipment 58, historical data, user input, etc. Clustering engine 84 can associate one or more operational characteristics and / or parameters with each of the one or more clusters of baseline 106. For example, condition monitoring system 50 can associate a first label with a first cluster of the baseline indicating nominal operation of shrink wrapper 38. The first label may correspond to a first operation of shrink wrapper 38, such as the rate at which film is fed into shrink wrapper 38. Condition monitoring system 50 can associate a second label with a second cluster of the baseline indicating nominal operation of shrink wrapper 38. The second label may correspond to a second operation of shrink wrapper 38, such as the movement of spacer wheels of shrink wrapper 38. In this way, each cluster constituting the baseline can be associated with a specific operation of shrink wrapper 38 and tagged accordingly. The condition monitoring system 50 can output a baseline 106 including tagged clusters. Baseline 106 can be used to establish normal operating parameters for the industrial automation equipment 58 over a period of time. It should be noted that box 114 of process 100 can be omitted.

[0050] Figure 4 This is a flowchart of process 150 according to various aspects of the present technology, which is used to extract features from visual data 152 to control the operation of industrial automation equipment 58. Process 150 can be referred to above. Figure 1 and Figure 2The disclosed condition monitoring system 50 may be executed using its engine 80 (e.g., feature extraction engine 82 and / or clustering engine 84), computing device and / or HMI 52, or any other suitable computing device or controller. Furthermore, the blocks of process 150 may be executed in the order disclosed herein or in any suitable order. For example, certain blocks of process 150 may be executed simultaneously. Additionally, in some embodiments, at least one block of process 150 may be omitted.

[0051] At box 154 of process 150, the condition monitoring system 50 can receive visual data 152 and extract a subset of features associated with changes in the operating condition of the industrial automation equipment 58 from the visual data 152. The visual data 152 can be obtained from... Figure 2 The visual data 152 is generated by the camera device 62 of the condition monitoring system 50. Figure 3 The visual data 152 is a portion of the visual data 104 and / or visual data obtained after establishing baseline 106. The visual data 152 may be preprocessed (e.g., cropped, rotated, masked, etc.) by one or more additional components of the image processing circuitry system 78 and / or the condition monitoring system 50. The visual data 152 can be captured in near real-time, providing a video stream of the operation of the industrial automation equipment 58. In some embodiments, the visual data 152 may include data from multiple cameras observing one or more different sets of equipment. Additionally and / or alternatively, the visual data 152 may include data on the operation of the industrial automation equipment 58 from one or more angles, one or more spectral regions, etc. The feature extraction engine 82 can perform feature extraction and output one or more features associated with the visual data 104. In some embodiments, the extracted features may correspond to references. Figure 3 The extracted features 102 are discussed. That is, the extracted features may include a subset (e.g., dimensionality-reduced) of the features of the visual data 152 associated with the operation of the industrial automation equipment 58. In some embodiments, the extracted features may include different features for each different piece of equipment observed in the visual data 152.

[0052] At box 156 of process 150, condition monitoring system 50 can receive baseline 106. Baseline 106 indicates the nominal operation of industrial automation equipment 58 within visual data 152. See reference... Figure 3Baseline 106 is generated as described in the diagram. In short, baseline 106 can be generated by clustering engine 84 based on the partitioning of data points containing extracted features derived from visual data collected during normal operation of industrial automation equipment 58. It should be noted that baseline 106 can be generated based on visual data different from visual data 152. That is, the visual data 104 used to generate baseline 106 may include additional visual data and / or a subset of visual data 152 corresponding to the nominal operating time period of the industrial automation equipment 58 of interest.

[0053] At box 156, the condition monitoring system 50 can determine whether the features extracted from visual data 152 deviate from baseline 106. In some embodiments, the condition monitoring system 50 can determine deviation from baseline 106 based on the formation of one or more new clusters (e.g., one or more data points) beyond a threshold distance from the clusters constituting baseline 106. Deviation from baseline 106 can indicate a change in the operation of associated industrial automation equipment. The threshold distance can be based on the distance between a specific data point and the centroids of the clusters of the baseline, a distance ratio (based on the ratio between the distance of the specific data point to its nearest centroid and its distance to the second nearest centroid), etc. A specific data point can be identified as an outlier. Multiple data points identified as exceeding the threshold distance from baseline 106 can be identified as new clusters. If no deviation from baseline 106 is detected for the extracted features, the condition monitoring system 50 can return to box 154. However, if the condition monitoring system 50 detects a deviation from the baseline 106, the condition monitoring system 50 can proceed to box 160, because the formation of new clusters can indicate potential anomalies in the operation of the industrial automation equipment 58.

[0054] At box 160 of process 150, condition monitoring system 50 can label one or more new clusters generated based on extracted features as associated with potential operational anomalies. Condition monitoring system 50 can identify one or more new clusters based on data points exceeding a threshold distance from baseline 106 (which can indicate changes in the operation of industrial automation equipment 58). Therefore, condition monitoring system 50 can provide outputs associated with detected changes. The output can be an alarm, notification, or visualization providing information to the user and / or HMI 52 to anticipate changes in production and / or operation. Such changes in production and / or operation can be provided before they have an impact on production and / or operation because changes in the characteristics of industrial automation equipment 58 can be determined before changes are detected in the associated process data. In this way, condition monitoring system 50 can predict changes in operation and / or production based on changes that are imperceptible to human operators and / or conventional monitoring systems.

[0055] In some implementations, it may be beneficial for the condition monitoring system 50 to identify one or more specific operating parameters that cause deviations from the baseline 106 indicating the nominal operation of the industrial automation equipment 58. Therefore, process 150 may proceed to block 162, whereby the condition monitoring system 50 identifies one or more markers associated with potential operational anomalies. These markers may be associated with one or more variables of process data, such as temperature data, vibration data, current data, drive data, etc. (See below for reference.) Figure 5 The discussion focuses on the identification of 50 tags in the status monitoring system.

[0056] At block 164 of process 150, condition monitoring system 50 can control the operation of industrial automation equipment 58 based on identified markers. Controlling the operation of industrial automation equipment 58 may include condition monitoring system 50 outputting control signals that change operating parameters (e.g., the speed of conveyor 24, the temperature provided by oven 26, etc.). Industrial automation equipment 58 can be controlled based on instructions provided from condition monitoring system 50 to HMI 52 and / or directly to the controller of industrial automation equipment 58. In this way, condition monitoring system 50 can provide finer-grained and proactive control of industrial operations, thereby reducing downtime.

[0057] As described above, the condition monitoring system 50 can identify one or more markers associated with potential operational anomalies. These markers make it easier for the condition monitoring system 50 to (e.g., with less computational resources) identify future occurrences of the anomaly or other operations that might lead to it. Therefore, Figure 5 This is a flowchart of process 200 according to various aspects of the present technology, which predicts the formation of one or more new clusters associated with one or more markers within process data. Process 200 can be described by referring to the above. Figure 1 and Figure 2 The disclosed condition monitoring system 50 may be executed using its engine 80 (e.g., feature extraction engine 82 and / or clustering engine 84), computing device and / or HMI 52, or any other suitable computing device or controller. Furthermore, the blocks of process 200 may be executed in the order disclosed herein or in any suitable order. For example, certain blocks of process 200 may be executed simultaneously. Additionally, in some embodiments, at least one block of process 200 may be omitted.

[0058] At box 204 of process 200, condition monitoring system 50 receives visual data 202 associated with one or more potential anomalies. Visual data 152 can be compared with, for example, [reference needed]. Figure 4 The predicted changes during the operation of the described industrial automation equipment 58 are correlated. Figure 4Compared to visual data 152, visual data 202 can be a subset of visual data 104. That is, visual data 202 can include data occurring within a time window preceding one or more potential anomalies (e.g., 10 seconds ago, 30 seconds ago, 1 minute ago, 2 minutes ago, 30 minutes ago, 1 hour ago). At box 204, the condition monitoring system 50 can identify attention points in visual data 202 corresponding to the formation of new clusters. The condition monitoring system 50 can use a transformer architecture to determine the time-varying relationships between extracted features to identify the formation of new clusters and determine attention points. Attention points can be based on geometric relationships between data point locations within a specific cluster and / or between one or more clusters, semantic relationships between data points based on the characteristics of the extracted features, etc.

[0059] At box 206 of process 200, condition monitoring system 50 receives process data 208 associated with the operation of industrial automation equipment 58. Process data 208 can be collected during the corresponding time period of visual data collection 202. Process data 208 can be... Figure 3 This is a portion of process data 116. At box 206, the condition monitoring system 50 can extract one or more markers from process data 208 that are associated with attention points. That is, the markers can identify changes in variables within process data 208 that lead to one or more potential anomalies. The markers can correspond to changes in process data, such as changes in temperature data, vibration data, pressure data, current data, drive data, etc. Marker extraction from process data can be based on the correlation between attention points in visual data 202 and changes in variables in process data 208. In some embodiments, changes in process data identified based on correlation with attention points can be detected before deviations within process data 208 itself are detected. In some embodiments, markers can be identified based on the association between visual data and historical process data. That is, the condition monitoring system 50 can be trained using historical data to pre-identify markers corresponding to the formation of new clusters from the process data.

[0060] In some implementations, the condition monitoring system 50 can iteratively extract tags associated with the formation of one or more additional clusters. In this way, visual data associated with additional potential anomalies can be used to build a list of relationships between tags and the formation of different clusters. Therefore, the condition monitoring system 50 can provide information related to various changes in the operation of the industrial automation equipment 58. At block 210 of process 200, the condition monitoring system 50 can output one or more relationships between tags and the formation of one or more clusters. During real-time monitoring of visual data related to the operation of the industrial automation equipment, the clustering engine 84 can use these relationships to not only tag potential changes in the operation of the industrial automation equipment 58, but also provide information related to the source of such changes. Providing relationships between tags and the formation of new clusters enables the condition monitoring system 50 to identify the causes of operational changes in components of the industrial automation system and reduce operational downtime.

[0061] The condition monitoring system 50 can enable Figure 2 The display 76 generates visualizations that help operators or users monitor the operation of industrial automation equipment 58. Figure 6 Based on various aspects of this technology Figure 2 This is an illustrative implementation of a graphical user interface (GUI) 300 generated within a condition monitoring system 50. The condition monitoring system 50 may provide the GUI 600 via a display 76 and / or one or more additional displays (e.g., generating and / or displaying the GUI 600). The condition monitoring system 50 may display a screen 302 on the GUI 300, which has one or more widgets 304, 306, 308 that can be used to visualize and provide information related to: visual data, baselines, clusters, etc., associated with the output of the feature extraction engine 82 and / or clustering engine 84 of the condition monitoring system 50. In this way, the condition monitoring system 50 may receive user input via one or more widgets 304, 306, 308. The GUI 300 of the condition monitoring system 50 may provide alarms, notifications, status updates, user requests, tagged actions, action recommendations, etc.

[0062] In some implementations, components 304, 306, and 308 of the condition monitoring system 50 include a clustering component 304, a baseline monitoring component 306, a visual data component 308, etc. The clustering component 304 may include one or more clusters 310, 312, and 314 associated with a baseline indicating the nominal operation of the industrial automation equipment 58. Each cluster 310, 312, and 314 may each include a series of data points generated by the clustering engine 84 by analyzing features extracted by the feature extraction engine 82. As shown, clusters 310, 312, and 314 may include boundaries 316 (e.g., cluster edges) corresponding to perimeters defined by one or more thresholds. The thresholds may be based on distances from the centroids of clusters 310, 312, and 314.

[0063] In some embodiments, clustering engine 84 can identify new clusters 318. New clusters 318 can indicate changes in the operation of industrial automation equipment 58. New clusters 318 can consist of one or more data points indicating features extracted by feature extraction engine 82. As shown, attention point 320 can be identified as a data point corresponding to a first data point outside the boundary 316 of clusters 310, 312, 314, which correspond to the nominal operation of the industrial automation equipment. In some embodiments, baseline monitoring widget 306 can be used to monitor the formation of new clusters. Therefore, baseline monitoring widget 306 can include a graph for monitoring deviations over time relative to baseline 322. Baseline 322 can be a characteristic of clusters 310, 312, 314 indicating the nominal operation (e.g., expected operation) of the industrial automation equipment 58. Therefore, baseline 322 can be monitored to determine deviations associated with new clusters and / or data points formed outside clusters 310, 312, 314 (which remain within the boundary 316). Therefore, one or more thresholds 324 may correspond to thresholds determined by clusters 310, 312, 314 that constitute the baseline 322 indicating the nominal operation of the industrial automation equipment 58.

[0064] In some implementations, the baseline monitoring widget 306 can indicate the formation of new clusters 318 when the graphical line 326 exceeds a threshold 324 of the baseline 322. The line 326 exceeding the threshold 324 can correspond to the formation of new clusters 318. An attention point 320 can correspond to a first point 328 on the line 326 where data points exceed the baseline 322. The visual data widget 308 can display data from... Figure 2 The visual data can be video data from one or more camera devices 62. In some embodiments, the visual data widget 308 can display preprocessed data, filtered data, etc. As shown, the visual data can be associated with one or more objects 330 moving on the conveyor belt 332.

[0065] The techniques proposed and claimed herein are referenced and applied to specific examples of material objects and actual properties that clearly improve upon the art and are therefore not abstract, intangible, or purely theoretical. Furthermore, if any claim appended to this specification contains one or more elements designated as “means for [executing] [capability]…” or “steps for [executing] [capability]…”, such elements are intended to be interpreted according to 35 USC 112(f). However, for any claim containing elements designated in any other manner, such elements are not intended to be interpreted according to 35 USC 112(f).

[0066] While only certain features of the embodiments described herein are illustrated and described, many modifications and alterations will occur to those skilled in the art. Therefore, it is to be understood that the appended claims are intended to cover all such modifications and alterations falling within the true spirit of the embodiments described herein.

Claims

1. A method comprising: Acquire first-view data indicating the operation of industrial automation equipment; Extract multiple features from the first visual data; Determine a subset of the plurality of features of the first visual data, wherein the subset of the plurality of features indicates changes in the operating status of the industrial automation equipment; Acquire second visual data indicating the operation of the industrial automation equipment; Using the second visual data, changes in the operating status of the industrial automation equipment are detected based on a subset of the multiple features; and Outputs are generated based on the detected changes in the operating conditions to control the operation of the industrial automation equipment.

2. The method according to claim 1, wherein, The plurality of features includes one or more temporal features, one or more spatial features, or a combination thereof.

3. The method according to claim 1, wherein, The second visual data includes a subset of the first visual data.

4. The method according to claim 1, wherein, Detecting changes in the operating status of the industrial automation equipment includes determining a subset of the plurality of features that deviate from a baseline, wherein the baseline comprises one or more clusters of data points representing the subset of the plurality of features.

5. The method of claim 4, further comprising determining that a subset of the plurality of features deviates from the baseline by a threshold deviation or more.

6. The method of claim 1, further comprising providing one or more notifications based on detecting changes in the operating status of the industrial automation equipment.

7. The method of claim 1, further comprising providing a graphical user interface (GUI), the GUI being configured to display a subset of the plurality of features, wherein, The GUI is configured to display changes in the operating status of the industrial automation equipment based on a subset of the plurality of features.

8. The method of claim 1, further comprising using a feature extraction engine to extract the plurality of features from the first visual data and determine a subset of the plurality of features of the first visual data.

9. The method of claim 1, further comprising using a clustering engine to detect changes in the operating status of the industrial automation equipment based on a subset of the plurality of features.

10. The method of claim 1, comprising: Acquire process data associated with the operation of the industrial automation equipment; Identify the attention points in the second visual data that correspond to the formation of the new cluster; Extract one or more tags from the process data that are associated with the attention point; as well as Output one or more relationships between the one or more tags and the formation of one or more clusters.

11. A system comprising: One or more camera devices, said one or more camera devices being configured to capture visual data associated with the operation of industrial automation equipment; Processing circuit system; as well as A memory accessible by the processing circuitry system, the memory storing instructions that, when executed by the processing circuitry system, cause the processing circuitry system to perform operations including: The visual data is acquired from one or more camera devices; Acquire process data associated with the operation of the industrial automation equipment; Changes in the operating status of the industrial automation equipment are detected by a clustering engine based on a subset of features of the visual data. Extract one or more markers from the process data that are associated with the detected changes in the operating status of the industrial automation equipment; as well as Output is generated based on one or more flags associated with the detected changes in the operating conditions to control the operation of the industrial automation equipment.

12. The system according to claim 11, wherein, When the instruction is executed, it causes the processing circuit system to perform the following operations: Acquire additional visual data indicating the expected operation of the industrial automation equipment; Multiple features are extracted from the additional visual data via a feature extraction engine; as well as A subset of the plurality of features of the additional visual data is determined, wherein the subset of the plurality of features corresponds to a baseline indicating the nominal operation of the industrial automation equipment.

13. The system according to claim 12, wherein, The feature extraction engine is an unsupervised model configured to extract the multiple features from unlabeled visual data.

14. The system according to claim 11, wherein, The clustering engine is an unsupervised model, and the clustering engine is configured to generate one or more clusters corresponding to a baseline indicating the nominal operation of the industrial automation equipment.

15. The system according to claim 11, wherein, When the instruction is executed, it causes the processing circuit system to perform the following operations: Identify the attention points in the visual data that correspond to the formation of the new cluster; as well as Extract one or more tags from the process data that are associated with the attention point.

16. The system of claim 11, comprising one or more sensors associated with the industrial automation equipment, wherein, The output of one or more sensors includes the process data.

17. A non-transitory computer-readable medium comprising computer-executable instructions, which, when executed, are configured to cause a processor to perform operations including: Receive first visual data indicating the operation of one or more industrial automation equipment; Extract multiple features from the first visual data; Determine a subset of the plurality of features of the first visual data, wherein the subset of the plurality of features indicates changes in the operating status of the one or more industrial automation equipment; Receive second visual data indicating the operation of the one or more industrial automation devices; Using the second visual data, changes in the operating status of the one or more industrial automation equipment are detected based on a subset of the plurality of features; and Outputs are generated based on the detected changes in the operating conditions to control the operation of one or more industrial automation devices.

18. The non-transitory computer-readable medium according to claim 17, wherein, The processor performs the following operations: Acquire process data associated with the operation of the industrial automation equipment; Identify the attention points in the second visual data that correspond to the formation of the new cluster; Extract one or more tags from the process data that are associated with the formation of the new cluster; as well as Output one or more relationships between the one or more tags and the formation of one or more clusters.

19. The non-transitory computer-readable medium according to claim 17, wherein, The processor performs operations including detecting changes in the operating status of the industrial automation equipment by determining that a subset of the plurality of features deviates from a baseline, wherein the baseline comprises one or more clusters of data points representing a subset of the plurality of features.

20. The non-transitory computer-readable medium according to claim 19, wherein, The processor performs the following operations: determining that a subset of the plurality of features deviates from the baseline by a threshold deviation.

Images