Detecting Abnormalities in Flanges using Embedded Devices and Tiny Machine Learning
Embedded devices with micro-machine learning predict flange anomalies, enhancing pipeline safety by automating inspections and reducing human reliance, addressing the inefficiencies of conventional flange inspection methods.
Patent Information
- Application Number
- SA1020243089
- Authority / Receiving Office
- SA · SA
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-06-03
- Publication Date
- 2026-10-06
- Estimated Expiration
- 2044-06-03
AI Technical Summary
Conventional flange inspections in pipeline systems are infrequent, labor-intensive, and reliant on human inspectors, leading to potential flange defects that can cause leaks and safety risks, especially in hard-to-reach locations.
Utilizing embedded devices and micro-machine learning to detect flange anomalies through sensors and AI algorithms, predicting conditions like loose bolts, misalignment, and flow blockages, enabling real-time, accurate assessments without human intervention.
Enables continuous, real-time detection of flange anomalies, improving safety and reducing maintenance costs by predicting and addressing issues before they become hazardous, ensuring the integrity of pipeline systems.
Smart Images

Figure 65138250-D15A-4FEC-A902-7F7789150636 
Figure 86024BDE-F379-4CB7-885F-00F6A779F08F 
Figure D76BAABE-8424-4C68-BBB5-AE1245DB6784
Abstract
Description
Flanges enable connections between pipes in pipeline systems that transport various materials, such as oil, gas, water, and the like. Additionally, flanges can seal pressurized vessels that store different materials. In general, the term "flange" refers to a protruding ridge, lip, rim, or collar, either external or internal, that increases strength. In some cases, a flange allows for the transfer of contact force to another object or provides stability and control over the movement of machinery or its components. Multiple flanges are used to create a flange joint, where two flange collars are joined together using a set of bolts paired with corresponding nuts. A seal is created between the flange collars, providing a strong connection between two pipes. Ensuring the integrity of flanges and flange joints ensures a safe pipeline or pressurized vessel. Flange defects can cause a flange leak, leading to a risk to life and property, field shutdowns, and loss of resources. The 3D models described here enable the detection of lip anomalies. Specifically, current technologies utilize embedded devices and micro-machine learning to detect anomalies. Examples of anomalies include screw loss, misalignment, and flow blockage. One or more sensors are used in conjunction with trained machine learning models to predict lip anomalies. Specifically, data from one or more sensors is processed and fed into an artificial intelligence (AI) algorithm, such as a neural network, to predict anomalies in one or more lips. In some models, the AI algorithm is trained on annotated data for both normal and anomalous lips. One of the advantages of current technologies is the generation of continuous, real-time data used to predict anomalies. Traditionally, on-site flanges are infrequently inspected because conventional inspections require a trained inspector to travel to the pipeline and visually examine the flanges. Furthermore, once on-site, the inspector may find flanges in hard-to-see locations or hidden by the pipeline design. Predictions using current technologies allow for more accurate information about flanges, including their accessibility. Moreover, predictions are made without relying on inexperienced inspectors. Inexperienced inspectors are often unable to correctly assess the condition of flanges during conventional inspections. Current technologies predict anomalies without inspector input, eliminating the need for extensive, costly, and time-consuming inspector training to evaluate flanges. Based on the prediction, flanges are repaired or replaced to ensure a safe and reliable system. Figure 1 is a diagram of a 100-pair flanges. In some configurations, 100-pair flanges are incorporated into a pipeline system. For example, a pipeline system might be a gathering, conveying, or distribution system with multiple flanges forming connections or couplings across the pipeline pipes. Flanges are often under high pressure and can transport hazardous or flammable materials. Accordingly, a faulty flange can create a hazardous environment. Current technologies enable the detection of flanges anomalies using embedded devices and micro-machine learning. These detections inform repair or maintenance operations within the pipeline system. In the examples, depending on the detected anomaly, the flange is repaired or maintained. For instance, loose bolts are tightened, or missing or shorted bolts are replaced. Additionally, flanges are realigned or adjusted in response to detected anomalies. In the example in Figure 1, the coupled flanges include 100 first coupled flange pairs, 110 second coupled flange pairs, 130 third coupled flange pairs, and 150 third coupled flange pairs. The first coupled flange pair, 110, includes bolts 112, 114, 116, and 118. The bolts extend through the bolt holes in hoops 120A and 120B of the 110 coupled flange pair. Nuts 122, 124, 126, and 128 receive the threaded ends of bolts 112, 114, 116, and 118. Tightening nuts 122, 124, 126, and 128 secures hoops 120A and 120B of the 110 coupled flange pair, creating a seal between pipe 101 and pipe 102. A flange gasket (not shown) is placed between hoops 120A and 120B to create a seal. In the example in Figure 1, the second coupled flange pair 130 includes several bolts 132, 136, and 138. The bolts extend through the bolt holes in the hoops 140a and 140b of the coupled flange pair 130. A number of nuts 142, 146, and 148 receive the threaded end of bolts 132, 136, and 138. Nuts 142, 146, and 148 are threaded onto bolts 132, 136, and 138. The paired flanges exhibit 130 anomalies. In these examples, an anomaly refers to a flange condition that deviates from the normal or usual condition that creates a proper seal at the flange. Additionally, an anomaly may refer to a flange condition that falls outside the limits associated with other standards or regulations that describe proper flange conditions. These standards or regulations can include those issued by organizations such as the American Society of Mechanical Engineers (ASME) and the American National Standards Institute (ANSI). Examples of anomalies include short / missing bolts, misalignment, a loose / not properly fitted flange, an incorrect gasket type, a missing gasket, a misplaced gasket, changes in process fluid properties (flow or pressure), and similar issues. The short screw anomaly is illustrated using screw 132 and nut 142. In the examples, the machine learning model predicts the flange condition as having a short screw. The short screw anomaly is indicated by screw 132 being short and failing to fully pass through the screw holes of collar 140A and collar 140B. At screw holes 134 and 144, the missing screw defect is illustrated. In the examples, the machine learning model predicts the flange condition as a missing screw condition. The missing screw anomaly is indicated by the screw holes being empty, reducing the integrity of the coupled flange collar 140A and collar 140B. Collars 140A and 140B of the coupled flange pair 130 exhibit an alignment anomaly. Specifically, collars 140A and 140B exhibit parallel misalignment, indicated by the offset 131 between centerline 135 of collar 140A and centerline 133 of collar 140B. In the examples, the machine learning model predicts the lip condition as a parallel misalignment. The third pair of 150 flanges includes several bolts, 152, 154, 156, and 158. These bolts extend through collars 160a and 160b of the third pair of 150 flanges. Nuts 162, 164, 166, and 168 receive the threaded ends of bolts 152, 154, 156, and 158. Nuts 162, 164, 166, and 168 are threaded onto the corresponding bolts 152, 154, 156, and 158. The pair of 150 flanges exhibits anomalies. For example, collars 160a and 160b of the pair of 150 flanges show a misalignment anomaly. Specifically, collars 160A and 160B exhibit angular misalignment, indicated by an angle of 151 between the centerline 155 of collar 160A and the centerline 153 of collar 160B. In the examples, the machine learning model predicts the flange condition as angular misalignment. Additionally, collars 160A and 160B exhibit a loose / loose-fit anomaly, indicated by a large gap of 170 between collars 160A and 160B. In the examples, the machine learning model predicts the flange condition as loose or loose-fit.For clarity, certain anomalies are shown on a single pair of conjoined lips. However, a pair of lips can exhibit one or more anomalies, either individually or in combination. A single lip or a pair of lips can display any number of anomalies. Furthermore, for ease of description, the anomalies in Figure 1 are shown as they would be perceived by the human eye. However, in some stereoscopic models, the anomalies associated with conjoined lips are so small and subtle that they are not visible to the human eye. For ease of description, the flanges described here include lip collars and a number of screws. However, flanges can be of many different types, such as a Weld Neck Raised Face (WNRF), Socket Weld (SW), Slip-On Flange, Flat Faced (FF), Lap Joint, Ring Joint, Threaded Flange, Reducing Flange, Blind Flange, and so on. Types of flanges include, for example, flat face, raised face, ring joint face, groove, tongue and groove, and male and female faces. Additionally, in the examples, flanges can have a number of finishes, such as serrated or smooth. In the examples, the machine learning model predicts the condition of the flange as either structural damage or damage to the finish of the flange. Figure 2 is a diagram of workflow 200 that enables the detection of anomalies in flanges using embedded devices and micro-machine learning. Workflow 200 uses inputs 210 to build a trained machine learning model 228 through a model-building process 220. Inputs 210 include raw sensor data 212, synthetic / simulation data 214, and other parameters 216. Raw sensor data 212, synthetic / simulation data 214, or other parameters 216 are used to form a training dataset used to train the machine learning model in the model-building process 220. In the examples, raw sensor data 212 is captured using one or more sensors mounted on the flange. For example, the sensors are mounted on the flange body, such as a flange collar. In other examples, the sensors are mounted near the flange. For example, the sensors are mounted on the upstream and downstream inlet and outlet portions of a pipe coupled using a flange.In one example, identical sensors (e.g., sensors of the same type) or sensors with a known relationship to the direction of entry and exit of a lip or lip pair are fitted, and the differences between the identical sensors are used to train a machine learning model or make predictions about the state of the lip. Sensors include, for example, vibration sensors, acoustic sensors, accelerometers, high-speed cameras, gyroscopes, ultrasonic sensors, temperature sensors, and the like. One or more sensors are used to capture data related to the lips. The data captured by the sensors is used to learn signatures or patterns in the sensor data associated with anomalies. In these examples, high-speed cameras positioned near the lips capture images of the lips used for motion zoom. Motion zoom amplifies subtle movements in the video sequence, making it possible to see distortions that might otherwise go unseen. In some microstructures, the initial sensor data (212) is acquired using several pipeline parameters, including intact and normal flanges under normal operating conditions and flanges with at least one abnormal condition. In some microstructures, the initial sensor data is labeled according to the condition of the flange at the time of data acquisition. The labels are a binary classification (e.g., normal, abnormal) of the flange condition. In some examples, the labels are a probability distribution for the presence of one or more conditions. In some models, the initial sensor 212 data is captured exclusively from the normal operating conditions of a healthy, natural flange. Consequently, the machine learning model is trained using sensor data that corresponds to the correct operating conditions of the flange. The machine learning model trained using this sensor data predicts whether the flange is normal, and any deviation from this normal state is considered an anomaly. Input 210 also includes synthetic / simulated data 214. Synthetic / simulated data is used to augment the machine learning model's training dataset. Training a machine learning model with a larger and more extensive dataset increases the accuracy of the detections performed by the trained machine learning model. Obtaining sensor data associated with abnormal lips can be more challenging than obtaining sensor data associated with normal lips. Synthetic / simulated data can represent edge cases or critical situations that do not occur frequently in the real world, resulting in a lack of sensor data representing these edge cases. Edge or critical situations include, for example, bursts, leaks, or other defects in the lip. Combining sensor data results in a more robust machine learning model that can detect both normal and abnormal lips.To generate synthetic / simulated data, an auto-encoder or other unsupervised learning technique of a neural network learns efficient data representations (e.g., encoding) by training a neural network to ignore signal noise. Input 210 also includes the other parameters 216. These other parameters allow for further generalization through the machine learning model. Other parameters include, for example, pipe diameter, pipe material of construction (MOC), service fluid, process parameters (flow, pressure, temperature, etc.), ambient conditions (e.g., ambient temperature, humidity, etc.), flange orientation (e.g., vertical or horizontal), and other deterioration conditions (e.g., material loss, coating damage, etc.). In the examples, the other parameters include details about the flange connection, such as flange size and rating (e.g., size in inches with pressure rating in pounds; gasket type; flange type; process fluid of interest along with properties such as pressure / flow / temperature, etc.). In some models, a training dataset is annotated that includes raw sensor data 212 and / or synthetic / simulated data based on other parameters 216, including parameters related to the pipeline and flange.These other parameters can improve the model's accuracy. In the examples, the other parameters are added directly and automatically to the training dataset using the same sensor on a corresponding embedded device. In these examples, the same sensor is a temperature sensor, pressure sensor, humidity sensor, or flow meter sensor. In these examples, the same sensor data is measured or obtained using a central processing unit (CPU). To train a machine learning model (M228), the model building process involves preprocessing (222) and training (224). The model building process obtains a training dataset that includes one or more inputs (210) and outputs a trained M28 model. Preprocessing (222) extracts features from the initial sensor data (212). In the examples, the initial sensor data (212) is time-sequential data, such as flange vibrations captured by a vibration sensor on or near the flange. Preprocessing the time-series data refines the data by checking for missing values, noisy data, and other inconsistencies, thus facilitating the training of the M28 model and producing a more accurate M28 model. In the examples, preprocessing (222) involves applying Fourier transforms, power transforms, normalization, difference transforms, and standardization to the training dataset.In Training 224, a pre-processed training dataset is used to train a machine learning algorithm such as a neural network, convolutional neural network, or recurrent neural network. The machine learning algorithm is trained to predict lip anomalies based on other parameters and the pre-processed sensor data. In the examples, signal detection using attention transformers is used to detect abnormal features. An attention transformer is a neural network that uses self-attention mechanisms to process the training data. The attention transformer determines the relevance of the training data in the dataset for predicting the output, thus focusing on the most relevant parts of the training dataset when producing the output. In some instances, the machine learning algorithm relies on a transformer architecture that enables long-term signal dependency capture, and the transformer architecture focuses on the input portion that maximizes correct output. The process of building model 220 results in a trained machine learning model 228 that predicts the anomaly at the lip. In some examples, the model building process involves testing, validating, or verifying the machine learning model to generate the trained machine learning model 228. In some examples, invisible input data, including unseen initial sensor data or other parameters associated with a particular lip, are fed into the trained machine learning model, and the trained machine learning model outputs a binary classification (e.g., normal, abnormal) associated with the lip. In some examples, the trained machine learning model produces a probability distribution for one or more anomalies present at one or more lips. In the examples, the trained machine learning model 228 is the input for optimization 229. Optimization 229 transforms the trained machine learning model 228 for execution on an embedded device, resulting in a miniature trained machine learning model 232. As shown in Figure 2, the miniature trained machine learning model is used for inference 230. In inference 230, the model is mounted on a low-power embedded device. In the examples, the embedded device is permanently positioned on or near a flange or flanged connector. Miniaturizing and scaling the machine learning model 228 using miniature machine learning techniques allows the miniature machine learning model 230 to be deployed on microcontrollers and similar low-power devices. In some instances, the trained machine learning model 228 is executed on a remote computer, and the raw sensor data is captured and transmitted to the remote computer, which is capable of more advanced processing. In some instances, the trained machine learning model 228 is used for inference 230 on the remote computer. As illustrated in Figure 2, the 232-trained micromachine learning model (e.g., optimization 229) has been optimized for deployment on resource-constrained devices, such as microcontrollers or low-power embedded systems. The 232-trained micromachine learning model is lightweight, with minimal memory and computing power, compared to unconstrained machine learning models (e.g., 228-trained model). In these examples, the 232-trained micromachine learning model is generated from the 228-trained model through optimization 229, which involves quantization and pruning to reduce the model's parameter size. In some examples, the 232-trained micromachine learning model is implemented using hardware accelerators, including dedicated inference chips, to improve inference speed while reducing power consumption. In the quantization process, weights are converted from float values to integer values, thus reducing the size of the machine learning model. Refining separates or removes weights that do not contribute significantly to the model's output. In inference 230, the trained micro-machine learning model 232 is deployed or updated (for example, 234) on a mesh. Deployment refers to the process of implementing a fully functional machine learning model for production where it can perform detections based on unseen data. In the examples, deployment refers to updating the existing machine learning model on the mesh. Updates occur, for example, when additional training data becomes available or in response to data skew (for example, the statistical properties of the target variable, which the model is trying to predict, have changed over time). Additional training data leads to better accuracy for the trained models. In some micro-scales, the optimal model is implemented on a predefined number of embedded devices (for example, 236). By having multiple sensors or devices permanently mounted on the flanges, additional data improves the accuracy of the trained model. The deployment / update of the machine learning model is described in more detail in Figures 3a–3b. In Inference 230, a 232-trained micromachine learning model is executed on a device embedded in the mesh network. Each embedded device in the mesh network captures sensor data from sensors deployed at or near a specific flange. The embedded devices are paired communically, enabling communication between the embedded devices on the mesh network and external networks and devices. Sensors associated with the flange transmit and relay the findings. In some instances, the mesh network enables a decentralized model, where each 232-trained micromachine learning model is executed on a mesh device and detects anomalies based on unseen sensor data or other parameters.In some embodiments, the mesh network enables a centralized model, and invisible sensor data is sent to a remote computer (e.g., a device on a network other than the mesh network) using the relevant embedded devices of the mesh network, and a fully trained machine learning model (e.g., the machine learning model output 228) detects anomalies based on invisible sensor data or other parameters in the remote computer. In these examples, detection based on invisible sensor data is used to identify any abnormal lip operation so that inspections or other corrective actions can be performed. The use of trained machine learning models improves lip anomaly detection by predicting abnormalities before a lip malfunction occurs. When sensor data predicts an abnormal lip condition, further lip inspections are performed. These additional inspections can be manual human inspections of the lip or further review of the captured sensor data. In these examples, the captured sensor data is transformed into a prediction of normal or abnormal lip conditions. When the sensor data predicts an abnormal lip condition, fluid flow at the lip or other lip use is stopped in a controlled manner to enable corrective actions and / or further inspection. Ultimately, lip operation is improved through predictions that provide insight into the condition of the lip in question. The trained machine learning model 228 and the trained micro-machine learning model 232 allow generalization and adaptation to new, previously unseen data derived from the same training dataset distribution. In the examples, even when trained exclusively on the initial sensor data, the trained model 228 and the trained micro-machine learning model 232 are generalized to adapt to the new, previously unseen data. In the examples, the trained micro-machine learning models are implemented across the relevant embedded devices on a nested network, and the weights of the trained micro-machine learning models are the same for each embedded device. The input parameters for other parameters vary for each embedded device. In some instances, a specialized machine learning model for a specific flange is built by training a machine learning model using data obtained from that specific flange. In this case, this specialized machine learning model receives as input the unseen data captured by the sensors associated with the specific flange. Figure 3A illustrates the architecture of system 300A, which enables lip anomaly detection using embedded hardware and micro-machine learning. In the example shown in Figure 3A, computer system 310 communicates with a master node 320. Master node 320 is in full-duplex communication with a mesh network 330. In the example in Figure 3a, the master node includes a master database 322, an Extract, Transform, and Load (ETL) template 324, and a machine learning algorithm 326. The ETL template 324 is a data integration process that aggregates data from multiple data sources into a single, consistent data store, such as the master database 322. The master database 322 stores detailed information about each node in the network, such as location, lip size, tag number, and machine learning model version. It also stores data collected from sensors at each lip for training and updating the model, as well as historical lip detection in the network. The main node 320 includes machine learning algorithm 326. In some models, machine learning algorithm 326 is the same as or similar to the trained machine learning model 228 from Figure 2. In some models, machine learning algorithm 326 operates in a centralized model, where it obtains data from embedded devices (e.g., sensors 350, 352, 354, 356, 358, 360, 362, 364, and 366) and predicts the lip anomaly. In some models, data from embedded devices is used to retrain machine learning algorithm 326 to improve its performance or accuracy. In some instances, the 326 machine learning algorithm operates in a decentralized model, where each embedded device has its own trained micro-machine learning model. In this decentralized model, there is no relationship between the 326 machine learning algorithm and the micro-machine learning models trained in each embedded device. However, the master node 320 is used to control and disseminate information through the mesh network in the decentralized model. The trained micro-machine learning models in each embedded device are updated and tested, and the master node 320 disseminates the updated models through the mesh network 330. The nested network includes 330 embedded devices (340a, 340b, 340c, ..., and 340n). As shown in Figure 3a, a trained micromachine learning model is present in each embedded device, and each embedded device is associated with a flange. The trained micromachine learning model in each embedded device may be the 232-trained micromachine learning model described in Figure 2. In the examples, an embedded device is permanently mounted to a flange. In the examples, the embedded device is permanently mounted to the flange during flange fabrication. In the examples, the embedded device is permanently mounted to the flange after flange fabrication. In the examples, an embedded device is removablely mounted to a flange during flange fabrication. In the examples, an embedded device is removablely mounted to a flange after flange fabrication. In the examples, the sensors are configured in a pairwise configuration, as illustrated by sensors 350 / 352 and 354 / 356. In a dual-sensor configuration, the first sensor is placed in the direction of the upstream flow relative to the flange, and the second sensor is placed in the direction of the downstream flow relative to the flange.By using at least two sensors in the incoming and outgoing flow directions, differential sensor data associated with the flange can be obtained. For example, when a dual-sensor configuration includes two accelerometers or other vibration sensors, differential vibration can be obtained from the flow within the pipes before and after the flange. In some models, differential sensor data is used to train and predict flange anomalies. For clarity, a fixed number of embedded devices and flanges are shown. However, the 330 mesh network can include any number of embedded devices connected to at least one flange. As shown in Figure 3a, embedded device 340a processes data associated with flange link 342a, embedded device 340b processes data associated with flange link 342b, embedded device 340c processes data associated with flange link 342c, and embedded device 340n processes data associated with flange link 342n. For clarity, Figure 3a shows embedded devices connected to a flange link that includes a pair of physically coupled flanges. However, embedded devices can be connected to a single flange. As shown in Figure 3a, each embedded device is communically paired with one or more sensors. The 340A integrated device is paired with sensors 350 and 352, the 340B integrated device is paired with sensors 354 and 356, the 340C integrated device is paired with sensors 358, 360, 362 and 364, and the 340N integrated device is paired with sensor 366.In some models, the machine learning model trained at each relevant flange is tuned based on the sensor data available at that flange. For example, the machine learning model trained at a particular flange is trained using data captured by sensors associated with that flange to achieve higher prediction accuracy. In other models, the machine learning model is tuned based on data similar to the flange environment, where the sensor data to be fed into the machine learning model is captured, rather than training using sensor data from different sources. The 340A, 340B, 340C, and 340N embedded devices capture sensor data associated with their respective lips and transmit that sensor data using the 330 mesh network. In some configurations, the 330 mesh network is an IPv6-based network. In some examples, the 330 mesh network is an OpenThread network (e.g., 802.15.4 Thread protocol) that routes data from one or more embedded devices that comprise the network. In some examples, the 330 mesh network is a Matter-based network according to the Connectivity Standards Alliance. In some examples, the 330 mesh network is a Bluetooth Low Energy (LE) network according to the Bluetooth Special Interest Group. In some examples, the 330 mesh network is a Bluetooth mesh network according to the Bluetooth Special Interest Group. In the examples, the 330 mesh network is a ZigBee network according to the Communications Standards Alliance. In the examples, network 330 is an ANT network according to the ANT+ Alliance.Figure 3b illustrates a 300b mesh network topology. In the examples, each embedded device (e.g., embedded devices 340a, 340b, 340c, ..., 340n) corresponds to a mesh network node. The nodes are represented by numbered circles in the 300b mesh network topology. The edges represent the communication paths between the nodes and are shown by lines and arrows connecting the nodes. The mesh network configuration allows connection to an external network at a single node of the mesh network. In the example in Figure 3b, node 1 is connected to an external network using a Wi-Fi connection. While node 1 is connected to the external network (e.g., the internet) via Wi-Fi and is considered the root node of the network, the remaining nodes are connected to the external network using communication paths that include the nearest node in the predefined mesh network topology. In some encapsulations, each node of the mesh network is capable of connecting to the Wi-Fi network (for example, it includes devices for connecting to the external network), however, only one node is connected to the external network.In some implementations, the root node enables communication between the IPv4 network (for example, the Internet) and the 330 mesh network. The root node is placed at the edge of the 330 mesh network, and in some implementations, the root node routes data between the 330 mesh network and an external network, such as the Internet. Each embedded node / device is a separate device with its own unique, trained micromachine learning model that executes independently of other embedded nodes / devices. Information is transmitted along communication paths (e.g., edges) to the root node connected to the external network. For example, when the trained micromachine learning model predicts a normal or anomalous edge, the information is transmitted across the embedded network to the external network. For example, in Figure 3b, an anomalous edge is predicted at node 9 and propagated down the sequence: node 9 → node 6 → node 3 → node 1 → Wi-Fi / Internet. As illustrated in Figure 3b, the connection between the two nodes is bidirectional, as shown by the bidirectional edges. This enables communication to and from each node according to a predefined communication protocol. For example, updates or other information are relayed to any node starting with the root node, which is connected to an external network. In some microarchitectures, the root node can receive updates for one or more nodes over the air. In some microarchitectures, communication to and from the nodes involves pre-programmed commands to shut down the relevant nodes, put them into sleep mode, or request information from neighboring nodes. In the examples, relaying and sharing information using a neighboring node is a command that instructs the neighboring node to relay and disseminate information down to the root node. For each command, there are specific predefined conditions that must be met before execution. For example, a node should not be shut down if it is the only means of access to a particular cluster or other nodes on the network. Referring again to Figure 3a, a trained micromachine learning model (e.g., the trained micromachine learning model 232 in Figure 2) is deployed or updated on one or more embedded devices in a mesh network. In some instances, the trained micromachine learning models are updated by updating and improving the machine learning algorithm 326 on the master node. The updates are then sent from the master node 320 via the mesh network 330 to the target set of embedded devices. Such a training and updating scheme is considered centralized training and deployment of a machine learning model. In some configurations, the root node of the mesh network enables two-way communication with the primary node 320. The root node of the mesh network 330 is paired communically with the primary node 320. The root node transmits sensor data to the primary node 320, and the primary node 320 transmits it to the computer system 310. In these examples, the computer system 310 might be a tablet, mobile phone, laptop, or other portable electronic device. The computer system 310 is connected to an IPv4 network and acts as a user interface for interacting with the devices embedded in the mesh network 330. For example, the computer system 310 can send configuration updates to these embedded devices. In some microstructures, the micro-machine learning models process the relevant sensor data locally at the corresponding embedded device and transmit it to the master node 320 via a root node in the mesh network. In other microstructures, the embedded devices transmit the data to the master node 320 for processing. Therefore, predictions can be made locally within the mesh network or at a remote location, such as the master node 320 or computer system 310. Figure 4 is a process flow diagram for detecting lip anomalies using embedded devices and micro-machine learning. In some instances, machine learning models are trained as shown in Figures 3a-3b and 4. Current techniques offer a systematic and automated procedure for assessing lip integrity. At frame 402, sensor data associated with at least one flange is obtained, where sensor data is captured by at least one sensor that is communically coupled to an integrated jihad on a mesh network. In framework 404, a trained micromachine learning model is implemented in the embedded device. Sensor data is fed into the trained micromachine learning model, and the model predicts the state of at least one lip. In the examples, the trained micromachine learning model was built from the trained micromachine learning model using a training dataset that included raw sensor data and synthetic sensor data. The state of at least one lip may be, for example, normal or abnormal. Additionally, the lip state may be a probability distribution of one or more abnormal states. At frame 406, the state is sent to a master node across the mesh network, where further actions are performed in response to the state of at least one lip. In the examples, these additional actions include checking at least one lip. Additionally, in the examples, the master node includes a retrained machine learning algorithm using sensor data from devices embedded in the mesh network. This retrained machine learning algorithm is then used to update the trained micro-machine learning model. Figure 5 is a schematic diagram of a 500 illustrative controller (or control system) for classifying lip integrity using artificial intelligence based on current detection. For example, the 500 controller could be operational according to the 200 workflow shown in Figure 2, the 400 process shown in Figure 4, or any combination thereof. Additionally, the 500 controller could be operational within the 300a system architecture shown in Figure 3a, the 300b nested network topology shown in Figure 3b, or any combination thereof. The 500 controller is intended to incorporate various forms of digital computing hardware, such as printed circuit boards (PCBs), processors, digital circuits, or other components of a supply chain alert management system. Furthermore, the system could include portable storage media, such as Universal Serial Bus (USB) flash drives. For example, USB flash drives might store operating systems and other applications.Portable USB drives can include input / output components, such as a wireless transmitter or a USB connector that can be inserted into the USB port of another computer. The 500 controller includes a 510 processor, 520 memory, a 530 storage device, and a 540 input / output interface, all of which are interconnected with 560 input / output devices (e.g., displays, keyboards, gauges, sensors, valves, and pumps). The 510, 520, 530, and 540 components are connected using the 550 system bus. The 510 processor is capable of processing instructions for execution within the 500 controller. The processor can be designed using any of several architectures. For example, the 510 processor might be a CISC (Complex Instruction Set Computer), a RISC (Reduced Instruction Set Computer), or a MISC (Minimal Instruction Set Computer). In one application, the 510 processor is a single-threaded processor. In another application, the 510 processor is a multi-threaded processor. The 510 processor is capable of processing instructions stored in memory (520) or on storage (530) to display the user interface's graphical information on the input / output interface (540). Memory 520 stores information within controller 500. In one application, Memory 520 is computer-readable medium. In another application, Memory 520 is volatile memory. In yet another application, Memory 520 is nonvolatile memory. The 530 storage device is capable of providing ample storage for the 500 controller. In one application, the 530 storage device is a computer-readable medium. In many different applications, the 530 storage device can be a floppy disk drive, a hard disk drive, an optical disc drive, or a tape drive. The 540 input / output interface provides input / output operations for the 500 controller. In one application, the 560 input / output devices include a keyboard and / or pointing device. In another application, the 560 input / output devices include a display unit for displaying graphical user interfaces. There can be any number of 500 controllers connected to or outside a computer system containing a 500 controller, with each 500 controller connected via a network. Furthermore, the terms "client," "user," and other appropriate terms may be used interchangeably, as appropriate, without exceeding the scope of current disclosure. Moreover, current disclosure assumes that multiple users may use a single 500 controller, and that a single user may use multiple 500 controllers. The applications of the subject and the functional processes described in these specifications may be implemented in digital electronic circuits, in tangible computer programs or firmware, in computer hardware, including the structures described in these specifications and their structural equivalents, or in combinations of one or more of these. Software applications of the described subject may be implemented as one or more computer programs. Each computer program may contain one or more computer program instruction units encoded on a tangible, non-temporary, computer-readable storage medium for execution by a data processing device or for controlling its operation. Alternatively, or in addition, the program instructions may be encoded into / on an artificially generated diffused signal. For example, the signal may be an electrical, optical, or electromagnetic signal generated to encode information for transmission to a suitable receiver for execution by a data processing device.A computer storage medium can be an automatic read / write storage device, an automatic read / write storage substrate, a random access memory or serial memory device, or a combination of computer storage media. The terms "data processing device," "computer," and "electronic computer" (or their equivalents as understood by a specialist) refer to devices that process data. For example, a data processing device can include all types of hardware, instruments, and machines for processing data, including, for instance, a programmable processor, a computer, multiple processors, or multiple computers. The device can also include special-purpose logic circuits, including, for example, a central processing unit (CPU), a field programmable gate array (FPGA), or an application-specific integrated circuit (ASIC). In some applications, a data processing device or an ASIC (or a combination of a data processing device or an ASIC) may be hardware-based or software-based (or a combination of both).The device may optionally include code that creates an execution environment for computer programs, for example, code that constitutes processor firmware, a protocol package, a database management system, an operating system, or a combination of these. The current disclosure assumes the use of data processing devices with or without traditional operating systems, such as Linux, UNIX, Windows, macOS, Android, or iOS. A computer program, which may also be referred to as a program, software, software application, module, software unit, script, or code, can be written in any form of programming language. Programming languages can include, for example, compiled languages, interpreted languages, declarative languages, or procedural languages. Programs can be deployed in any form, including standalone programs, modules, components, subroutines, or units for use in a computing environment. A computer program can, but is not necessarily, correspond to a file in a file system. A program may be stored in a part of a file containing other programs or data—for example, one or more scripts stored in a markup language document—or in a single file dedicated to the program itself, or in multiple formatted files that store a module, subroutines, or pieces of code. A computer program may be deployed for execution on a single computer or on multiple computers located, for example, in one location or distributed across multiple locations connected by a network.While parts of programs illustrated in various figures can be viewed as individual units implementing different features and functions through different objects, methods, or processes, programs can instead comprise a number of sub-modules, third-party services, components, and libraries. Conversely, the features and functions of different components can be combined into single components as needed. Thresholds used for computational decision-making can be defined statically, dynamically, or both statically and dynamically. The methods, operations, or logic flows described in this specification can be implemented by one or more programmable computers that execute one or more computer programs to perform functions by working on input data and producing output. The methods, operations, or logic flows can also be implemented by a device, or the device itself can be a special-purpose logic circuit, for example, a central processing unit (CPU), FPGA, or ASIC. Computers suitable for executing computer programs can rely on one or more general-purpose and specialized microprocessors and other types of central processing units (CPUs). The components of a computer are the CPU, which executes instructions, and one or more memory devices for storing instructions and data. Generally, the CPU can receive instructions and data from (and write data to) memory. A computer can also include, or be effectively coupled with, one or more large-capacity storage devices for data storage. In some applications, a computer can receive data from, and transfer data to, large storage media, including, for example, magnetic disks, magnetic optical discs, or optical discs. Furthermore, a computer can be embedded in another device, for example, a mobile phone, a personal digital assistant (PDA), a portable audio or video player, a game console, a GPS receiver, or a portable storage device such as a USB flash drive. Computer-readable media (temporary or non-temporary, as appropriate) suitable for storing computer program instructions and data can include all forms of permanent / non-permanent and volatile / non-volatile memory, media, and memory devices. Computer-readable media can include, for example, semiconductor memory devices such as random access memory (RAM), read-only memory (ROM), phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and flash memory devices. Computer-readable media can also include, for example, magnetic devices such as tape, cartridges, cassettes, and internal / removable disks.Computer-readable media can also include magnetic optical discs and optical memory devices and technologies, including, for example, DVDs, CD-ROMs, DVD+ / -R, DVD-RAM, DVD-ROM, HD-DVD, and Blu-ray discs. Memory can store various objects or data, including cache memory, classes, frameworks, applications, modules, backup data, functions, web pages, web page templates, data structures, database tables, repositories, and dynamic information. Types of objects and data stored in memory can include parameters, variables, algorithms, instructions, rules, constraints, and references. Additionally, memory can contain logs, policies, security or access data, and report files. The processor and memory can be complemented or combined in special-purpose logic circuits. The applications described in this report can be implemented on a computer equipped with a display device to provide user interaction, including displaying information to (and receiving input from) the user. Display device types can include, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), a light-emitting diode (LED), and a plasma monitor. Display devices may include a keyboard and pointing devices, such as a mouse, trackball, or touchpad. User input to the computer can also be provided using a touchscreen, such as a pressure-sensitive tablet or a multi-touch screen using capacitive or electrical sensing. Other types of devices can be used to provide user interaction, including receiving user feedback, such as sensory feedback (visual, auditory, or tactile).User input can be received in the form of voice, speech, or touch input. Additionally, the computer can interact with the user by sending documents to and from the user's device. For example, the computer can send web pages to a web browser on the user's client device in response to requests from that browser. The term "graphical user interface" or "GUI" can be used in the singular or plural to describe one or more graphical user interfaces and each of the displays of a particular GUI. Therefore, a GUI can represent any graphical user interface, including, but not limited to, a web browser, a touchscreen, or a command-line interface (CLI) that efficiently processes information and presents information results to the user. Generally, a GUI can include a set of user interface elements, some or all of which are associated with a web browser, such as interactive fields, drop-down menus, and buttons. These and other user interface elements may be related to or represent the functions of a web browser. The applications described in this specification can be implemented in a computer system that includes a back-end component, for example, a data server, or a middleware component, for example, an application server. Furthermore, the computing system can include a front-end component, for example, a client computer containing one or both a graphical user interface and a web browser through which the user can interact with the computer. The system components can be connected by any form or medium of wired or wireless digital data communication (or a combination of data communication) in a communications network. Examples of communications networks include a local area network (LAN), a wireless access network (RAN), a metropolitan area network (MAN), a wide area network (WAN), WiMAX (Worldwide Interoperability Microwave Access), a wireless LAN (WLAN) (for example, using 802.11 a / b / g / n or 802.20 or a combination of protocols), all or part of the Internet, or any other communication system or systems in one or more locations (or a combination of communication networks).A network can communicate, for example, with Internet Protocol (IP) packets, frame relays, asynchronous transfer mode (ATM) cells, audio, video, data, or a combination of communication types between network addresses. A computing system can include clients and servers. Clients and servers can generally be geographically separated and typically interact through a network. The client-server relationship can be established through computer programs running on the relevant computers and interacting with them. A pooled file system can be any type of file system accessible from multiple servers for reading and updating. Lock tracking and consistency may not be necessary since the locking of the pooled file system can be implemented at the application layer. Furthermore, Unicode data files can differ from non-Unicode data files. While these specifications contain many specific implementation details, they should not be interpreted as limitations on what can be claimed, but rather as descriptions of features that may be specific to certain applications. Some of the features described in these specifications can also be implemented in separate applications, combined, within a single application. Conversely, many of the features described in a single implementation can also be implemented in multiple applications, separately, or in any suitable subset. Furthermore, although the previously described features may be described as operating in certain combinations and even initially claimed as such, in some cases, one or more of the features may be removed from the claimed combination, and the claimed combination may be directed to a subcombination or a subset of a subcombination. Specific applications of the topic are described. Other applications, modifications, and variations of the described applications fall within the scope of the following safeguards, as will become clear to those experienced in the field. While the operations in the diagrams or safeguards are depicted in a particular order, this should not be understood to mean that these operations must be performed in the exact order shown or sequentially, or that all of the operations shown must be executed (some may be considered optional) to achieve the desired results. In certain circumstances, multitasking or parallel processing (or a combination of multitasking and parallel processing) may be beneficial and should be implemented as appropriate. Furthermore, the separation or integration of different system units and components in the applications described above should not be understood as requiring such separation or integration in all applications. It should be noted that the software components and systems described can generally be integrated together in a single software product or as a package in multiple software products. Therefore, the illustrative applications described above do not define or limit the current disclosure. Other changes, substitutions, and modifications can also be made without deviating from the core and scope of the current disclosure. Furthermore, any purported application is considered to be applicable to at least a computer-executed method; a computer-readable, non-temporary medium that stores computer-readable instructions for performing the computer-executed method; and a computer system that includes computer memory mutually coupled to a hardware processor configured to perform the computer-executed method or the instructions stored on the computer-readable, non-temporary medium. Specific implementations of the subject have been described. Consequently, other implementation methods can be used, all of which fall within the scope of the following safeguards. For example, the actions described in the safeguards can be performed in a different order and still achieve the desired results. For instance, some of the operations shown in the accompanying figures do not necessarily require the specific order or sequence shown to achieve the desired results.
Claims
1. A computer-implemented method for detecting flange abnormalities. The method includes: acquiring, using one or more hardware processors, sensor data associated with at least one flange, where the sensor data is captured by at least one sensor connected to an embedded device on a mesh network; executing, using one or more hardware processors, a trained machine learning model in the embedded device, where the sensor data is fed into the trained machine learning model, and the trained machine learning model predicts the state of at least one flange; and sending, using one or more hardware processors, the state to a master node via the mesh network, where further actions are performed in response to the state of at least one flange. 2.The computer-implemented method, as per safeguard element 1, where a trained machine learning model is built from the initial trained machine learning model using a training dataset that includes raw sensor data and synthesized sensor data.
3. The computer-implemented method, as per safeguard element 1, where additional procedures include checking at least one lip.
4. The computer-implemented method, as per safeguard element 1, where the master node includes a retrained machine learning algorithm using sensor data from embedded devices on the mesh network, and the retrained machine learning algorithm is used to update the trained machine learning model.
5. The computer-implemented method, as per safeguard element 1, where at least one lip is either normal or abnormal. 6.The method is implemented by computer according to protection element 1, where the lip condition is a probability distribution of the presence of one or more abnormalities. 7.An apparatus containing a non-transitory, computer-readable storage medium stores instructions which, when executed by at least one processor, cause at least that processor to perform operations including: acquiring sensor data associated with at least one lip, where the sensor data is captured by at least one sensor connected to an embedded device on a mesh network; executing, by one or more of the apparatus processors, a trained machine learning model in the embedded device, where the sensor data is fed into the trained machine learning model and the trained machine learning model predicts the state of at least one lip; and sending the state to a master node via the mesh network, where further actions are performed in response to the state of at least one lip. 8.
9. The apparatus, according to safeguard element 8, has a trained machine learning model built from the initial trained machine learning model using a training dataset that includes raw sensor data and synthesized sensor data.
10. The apparatus, according to safeguard element 7, includes further actions that involve inspecting at least one lip.
11. The apparatus, according to safeguard element 7, has a master node that includes a retrained machine learning algorithm using sensor data from embedded devices on the mesh network. This retrained machine learning algorithm is then used to update the trained machine learning model.
12. The apparatus, according to safeguard element 7, has a master node that controls the embedded device using commands distributed from the master node to the root node and to a node containing the embedded device.The apparatus, according to safeguard element 7, is designed so that at least one flange condition is either normal or abnormal.
13. The apparatus, according to safeguard element 7, is designed so that the flange condition is a probability distribution for the presence of one or more abnormalities. 14.A system comprises one or more memory modules; one or more hardware processors connected to one or more memory modules. One or more of the hardware processors are configured to execute instructions stored on one or more memory modules to perform operations that include: acquiring sensor data associated with at least one lip, where sensor data is captured by at least one sensor connected to an embedded device on a mesh network; executing a trained machine learning model in the embedded device, where the sensor data is fed into the trained machine learning model, and the model predicts the state of at least one lip; and transmitting the state to a master node via the mesh network, where further actions are performed in response to the state of at least one lip. 15.The system, according to security element 14, has a trained machine learning model built from the initial trained machine learning model using a training dataset that includes raw sensor data and synthesized sensor data.
16. The system, according to security element 14, includes additional procedures that involve inspecting at least one flange.
17. The system, according to security element 14, has a master node that includes a retrained machine learning algorithm using sensor data from embedded devices on the mesh network. This retrained machine learning algorithm is then used to update the trained machine learning model.
18. The system, according to security element 14, has a master node that controls the embedded device using commands distributed from the master node to the root node and to a node containing the embedded device. 19.The system is in accordance with protection element 14, where at least one flange is either normal or abnormal.
