JP2023552461A - Systems and methods for corrosion and erosion monitoring of pipes and vessels - Google Patents

Abstract

The present disclosure relates to the field of corrosion and erosion monitoring of pipes and vessels. More specifically, the present disclosure relates to systems and methods for corrosion and erosion monitoring of pipes and vessels, the systems/methods including ultrasonic thickness monitoring using longitudinal waves and one or more waveguides. Combined with the use of ultrasonic area monitoring, whereby representative thickness measurements are complemented by area monitoring features to detect localized corrosion/erosion between representative thickness measurement locations. In another embodiment, a system and method for optimized asset health monitoring that includes an analytical solution is disclosed. [Selection diagram] Figure 1

Description

(Related application)
This application is filed on February 28, 2020 with attorney reference number MX-2020-PAT-0029-US-PRO entitled "SYSTEM AND METHOD FOR CORROSION AND EROSION MONITORING OF PIPES AND VESSELS." U.S. provisional patent for the name Related to Application No. 62/982,751. The disclosures of the above applications are incorporated herein by reference in their entirety.

(Field of invention)
The present disclosure relates to the field of corrosion and erosion monitoring of pipes and vessels. Specifically, the present disclosure relates to corrosion and/or erosion monitoring systems that include mechanical components, hardware, software, analysis, and/or combinations thereof. In one embodiment, the mechanical components and hardware may include one or more ultrasound transducers, base units, gateways, and/or combinations thereof. The system may further include a software platform for remote monitoring. The system may further include analysis tools for front-end services and back-end services for remote monitoring and/or diagnostics in some embodiments. More specifically, in some embodiments, the present disclosure may relate to systems and methods for corrosion and erosion monitoring of pipes and vessels, where the systems/methods include ultrasonic thickness monitoring using longitudinal waves. , combined with ultrasonic area monitoring using one or more waveguides, whereby representative thickness measurements are complemented by area monitoring features to detect localized corrosion/erosion between representative thickness measurement locations. Detect. In another embodiment, a system and method for optimized asset health monitoring that includes an analytical solution is disclosed.

The use of ultrasound transducers to ultrasonically monitor the condition and integrity of structural assets, including pipes and pressure vessels, such as those used in the oil, gas and power generation industries, is well known. Corrosion and erosion monitoring systems and techniques that incorporate/use ultrasonic transducers are currently known, including in-location thickness monitoring and area monitoring (also known as waveguide inspection). However, these two systems and techniques are typically separate from each other. Additionally, internal corrosion of piping systems is sometimes monitored using radiation (RT) thickness testing, in addition to ultrasonic (UT) testing, on selected components at predetermined intervals over the life of the system. Measure the wall thickness of.

Thickness monitoring ultrasonic transducers and systems utilizing them typically measure the thickness of a pipe/vessel wall at the spot where the ultrasonic transducer is provided; in other words, it does not provide any information regarding the thickness of the pipe/vessel wall at the location surrounding the exact spot. Therefore, if corrosion/erosion is occurring at a location other than where an ultrasound transducer is provided, the corrosion/erosion will likely not be detected unless thickness monitoring is accompanied by ultrasound transducer mapping. Naturally, ultrasound transducer mapping increases inspection costs. However, it is advantageous for these ultrasonic transducers and systems to be permanently installed in pipes/vessels.

Conversely, area monitoring ultrasound transducers and systems utilizing them typically measure pipe/vessel wall thickness over a larger area of the pipe/vessel wall; is beyond where a thickness monitoring ultrasonic transducer is provided on the pipe/vessel. Such area monitoring ultrasound transducers and systems utilizing them typically produce a pipe/vessel wall thickness map over the area being measured. In theory, such a generated thickness map would be useful, but currently such waveguide inspections are performed using common hardware whose segments generate 10 to 20 different waveguide modes. The large number of wave modes and complex analysis adversely affect the reliability of the test results. Additionally, waveguide tests are typically not permanently installed on pipes and vessels. In addition, highly localized corrosion must be reliably detected in temporarily installed waveguide systems such as those described in API 574 (API 574, Inspection practices for piping system components, 4th edition, 2016). I can't.

In addition, existing permanently installed corrosion monitoring systems use appropriate data to determine sensor placement in industrial facilities such as refineries and petrochemical plants that use piping systems to transport fluids. Can not do it. The piping system may transport the fluid to one or more tanks and/or chemical processing units. Some piping systems handle specialized fluids at predetermined temperatures and/or pressures. These piping systems may transport highly corrosive fluids at high temperatures and pressures.

Additionally, many industrial facilities face health and safety concerns. They may transport fluids that may be flammable and/or toxic. Therefore, failure of the piping system can result in leaks to the atmosphere and/or exposure to plant personnel. Furthermore, some equipment operates without scheduled outages for several years. Therefore, the reliability of the piping system and its components is important.

In addition to health and safety concerns, unplanned outages due to piping system failures are problematic from a business consequences perspective. The condition of piping systems is monitored to accurately predict their remaining life and determine safe repair or replacement dates, taking into account the potential safety, health, environmental, and business risks associated with piping failures. do.

As a result of the foregoing, certain individuals will appreciate improvements in systems and methods for pipe and vessel corrosion and erosion monitoring.

In the following description of various exemplary embodiments, reference is made to the accompanying drawings, which form a part of this specification, and which illustrate, by way of illustration, various implementations in which aspects of the disclosure may be practiced. The morphology is shown. It should be understood that other embodiments may be utilized and structural and functional modifications may be made without departing from the scope of this disclosure. Note that various connections between elements are discussed in the description below. These connections are typical and may be direct or indirect, wired or wireless, unless otherwise specified, and this specification is not intended to be limiting in this regard. Please note that.

Configuring a system of one or more computers to perform a particular operation or action by installing software, firmware, hardware, or a combination thereof on the system that causes the system to perform the action during operation. Can be done. One or more computer programs may be configured to perform particular operations or actions by including instructions that, when executed by a data processing device, cause the device to perform the actions. One general aspect includes a method for down-selecting among probe assemblies installed in a piping system. The method also includes setting a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter for the model before training the model. The method also groups the first set of probe assemblies based at least on historical pipe wall thickness measurements collected from probe assemblies installed on the piping system over a period of time by the model executed on the processor. including doing. The method also includes assigning a unique group ID to each set of probe assemblies. The method also includes selecting, by the model, an optimization function among a plurality of optimization functions for the model after training the model. The method also includes identifying, by the model, a single probe assembly corresponding to each group ID for pipe wall thickness monitoring of the piping system. The method also includes transmitting a single probe assembly pipe wall thickness measurement from each group ID for inspection by a thickness monitoring controller associated with the piping system. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs stored on one or more computer storage devices, each configured to perform the actions of the method.

Implementations may include one or more of the following features and other features. The method may include one or more steps for ignoring all remaining probe assemblies within each group ID except for a single probe assembly from each group ID during testing. The grouping of the first set of probe assemblies is further based at least on inspection information provided to the system and historical pipe wall thickness measurements collected over a period of time from probe assemblies installed on the piping system. The piping system can include a tank, and a first of the probe assemblies is configured to measure wall thickness of the tank. The method also includes the step of: storing historical pipe wall thickness measurements collected over an extended period of time from probe assemblies installed in the piping system in a computer memory communicatively coupled to the processor; training the model using historical pipe wall thickness measurements stored in the method. The model may include an artificial neural network. Implementations of the described techniques may include hardware, methods or processes, or computer software on a computer-accessible medium.

One common aspect includes a system for detecting global corrosion (eg, lack of localized corrosion) on multiple components that transport material over a distance. The system can also include a plurality of probe assemblies attached to one or more of the components, the probe assemblies detecting corrosion (e.g., global corrosion and/or localized corrosion) to the component. The method may include at least a thickness monitoring ultrasonic transducer and an area monitoring ultrasonic transducer configured to. The system may also include a data store configured to store historical wall thickness measurements collected over a period of time from measurements performed by the probe assembly. The system may also include a model trained with hyperparameters on historical wall thickness measurements in the data store, the hyperparameters may include a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter. The system may also include a monitoring device, the monitoring device may include a processor and a memory storing computer-executable instructions, the computer-executable instructions, when executed by the processor, cause the system to assigning a unique group ID to each set of probe assemblies based on the first set of probe assemblies; and selecting an optimization function from among the plurality of optimization functions based on the model. selecting, and identifying, based on the model and the selected optimization function, a probe assembly corresponding to each group ID for wall thickness monitoring of the component; and a thickness monitoring controller associated with the component. and transmitting wall thickness measurements of the probe assembly from each group ID for inspection. In another embodiment, the system may output a list of unique identifiers corresponding to any group ID instead of sending wall thickness measurements for inspection. The inspector can receive the output of the system and react accordingly, as described in various embodiments disclosed herein. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs stored on one or more computer storage devices, each configured to perform the actions of the method.

Implementations may include one or more of the following features and other features. The system may include two or more probe assemblies of the plurality of probe assemblies, the probe assembly identified from each group ID, the memory of the monitoring device storing computer-executable instructions; when executed by the processor, causes the system to: ignore all remaining probe assemblies within each group ID except for two or more probe assemblies from each group ID during inspection; verifying that the wall thickness measurements of the two or more probe assemblies from the ID are global corrosion and not localized corrosion. The probe assembly wall thickness measurements from the first group ID may include wall thicknesses of pipe components in the probe assembly. The probe assembly wall thickness measurements from the first group ID may include wall thicknesses of tank components in the probe assembly. The method (i) generates a probability plot of all pipe wall thickness measurements associated with a piping system, (ii) groups the plotted pipe wall thickness measurements by nominal thickness, and (iii) Pipe wall thickness of a single probe assembly by identifying nonlinear relationships in probability plots of pipe wall thickness measurements grouped by nominal thickness to confirm structural corrosion (e.g., lack of localized corrosion) It may include verifying that the measurement is global corrosion (eg, lack of localized corrosion). Pipe wall thickness monitoring may include steps for analyzing original wall thickness, wall thickness loss over time, calibration errors, and repeatability errors of measurement locations by the probe assembly. Implementations of the described techniques may include hardware, methods or processes, or computer software on a computer-accessible medium.

Implementations may include one or more of the following features and other features. The method generates a probability plot of all pipe wall thickness measurements associated with a piping system, groups the plotted pipe wall thickness measurements by nominal thickness, and groups the plotted pipe wall thickness measurements by nominal thickness and By identifying a nonlinear relationship in a probability plot of pipe wall thickness measurements grouped by nominal thickness to confirm the absence of overall corrosion (e.g. , absence of localized corrosion). Pipe wall thickness monitoring may include steps for analyzing original wall thickness, wall thickness loss over time, calibration errors, and repeatability errors of measurement positions by the probe assembly. Implementations of the described techniques may include hardware, methods or processes, or computer software on a computer-accessible medium.

The present disclosure is presented by way of example and not as a limitation to the accompanying drawings, in which like reference numbers indicate similar elements.

1 is a diagram of a system for corrosion/erosion monitoring; FIG. 2 is a diagram of the thickness monitoring controller and piezoelectric assembly of the system of FIG. 1; FIG. 3 is a diagram of the thickness monitoring controller of FIG. 2; FIG. 3 is a diagram of a switch assembly forming part of the piezoelectric assembly of FIG. 2; FIG. 3 is a diagram of the piezoelectric assembly of FIG. 2; FIG. 1 is a diagram of a method for corrosion/erosion monitoring; FIG. 1 is a diagram of a method for corrosion/erosion monitoring; FIG. 1 is a diagram of a method for corrosion/erosion monitoring; FIG. FIG. 3 is a diagram for displaying signal modulation. FIG. 3 is a diagram for displaying signal modulation. FIG. 3 is a diagram for displaying signal modulation. FIG. 3 is a diagram for displaying signal modulation. 13A and 13B (collectively referred to as “FIG. 13”) are illustrations of one example piping with an installed MUT sensor in accordance with one or more aspects of the features disclosed herein. . 13A and 13B (collectively referred to as “FIG. 13”) are illustrations of one example piping with an installed MUT sensor in accordance with one or more aspects of the features disclosed herein. . 1 is an example network architecture of an industrial facility in accordance with various aspects of the present disclosure. FIG. 3 is an illustration of probe assembly grouping in an embodiment of the present disclosure. 16A, 16B, and 16C (collectively referred to as "FIG. 16") are graphs showing probability plots of measurements to verify global corrosion as opposed to localized corrosion. 16A, 16B, and 16C (collectively referred to as “FIG. 16”) are graphs illustrating levels of risk to TML in accordance with various aspects disclosed herein. 16A, 16B, and 16C (collectively referred to as “FIG. 16”) illustrate shifts in curves depicting the level of risk for TML after downselection, in accordance with various aspects disclosed herein. do. 1 is a graphical plot showing the cumulative thickness distribution of a tube with naphthenic acid corrosion. 2 is a corrosion sensor analysis graph showing TML measurements by date in an embodiment of the present disclosure. 8A is another corrosion sensor analysis graph showing TML measurements by date, similar to FIG. 8A, but with higher grouping sensitivity settings; FIG. 8A is yet another corrosion sensor analysis graph illustrating TML measurements by date, similar to FIG. 8A, but with higher grouping sensitivity settings; FIG. 4 is a graph in accordance with one or more aspects of the present disclosure. 3 is a graph in accordance with one or more aspects of the present disclosure. 3 is a graph in accordance with one or more aspects of the present disclosure. 4 is a graph in accordance with one or more aspects of the present disclosure. 1 illustrates an example artificial neural network configured to operate in conjunction with the systems, methods, and algorithms disclosed herein. 1 is a flowchart illustrating example steps of a method performed in accordance with some embodiments disclosed herein. 2 is a simplified pipe and instrumentation diagram (PID) corresponding to an exemplary corrosion/erosion monitoring system as shown in FIG. 1, according to some embodiments disclosed herein; FIG.

In the following description of various exemplary embodiments, reference is made to the accompanying drawings, which form a part of this specification, and which illustrate, by way of illustration, various implementations in which aspects of the disclosure may be practiced. The morphology is shown. It should be understood that other embodiments may be utilized and structural and functional modifications may be made without departing from the scope of this disclosure. Note that various connections between elements are discussed in the description below. These connections are typical and may be direct or indirect, wired or wireless, unless otherwise specified, and this specification is not intended to be limiting in this regard. Please note that.

The present disclosure is illustrated in the drawings, although embodiments may permit different forms. In addition, specific descriptions of the disclosure are to be considered illustrative of the principles of the disclosure, and are not intended to limit the disclosure to what is illustrated and described herein. Embodiments will be described in detail herein. Thus, unless stated otherwise, the features disclosed herein may be combined to form additional variations not otherwise shown for brevity. It is further understood that in some embodiments, one or more elements shown by way of example in the drawing(s) may be eliminated and/or replaced with alternative elements within the scope of this disclosure. Dew.

Aspects of the present disclosure relate to monitoring and detecting corrosion and/or erosion of pipes, vessels, and other components in industrial facilities. The monitoring system may include a software platform for remote monitoring and analysis of historical measurements collected by multiple sensors attached to pipes and components. The monitoring system can include analytical tools for monitoring, diagnosing, and/or predicting localized and/or global corrosion. By using the analysis system disclosed herein, the thickness monitoring locations (TML) can be optimized to reduce, i.e. down-select, the number of measurement locations without compromising risk, among other things. obtain. As described in this disclosure, by strategically reducing the number of probe assemblies that need to be sampled during an inspection, through down-selecting, the amount of inspection time/cost is reduced while the industry The facility's risk profile is maintained (or even reduced).

A system 100 for monitoring pipe/vessel corrosion and erosion is shown in FIGS. 1 and 2. System 100 includes a data analysis and visualization platform 110, an optional gateway 120, a thickness monitoring controller 130, a thickness monitoring ultrasound transducer 140 used for standardization purposes, and at least one probe assembly 150. include. Each probe assembly 150 includes a switch assembly 160, at least one thickness monitoring ultrasound transducer 170, and at least one area monitoring ultrasound transducer 180.

Data analysis and visualization platform 110 includes a data analysis unit 112 and a visualization unit 114. 1 data analyzer 112 is typically cloud-based, powered software configured to receive signals, typically wirelessly, from one or both of gateway 120 or thickness monitoring controller 130. These signals are analyzed by the data analysis unit 112 and converted into images for display on the visualization unit 114. Visualizer 114 may include any suitable device, such as a computer monitor, tablet, etc., of the type that assists the individual monitoring platform 110 in understanding information regarding corrosion/erosion identified by system 100. It can be a telephone, etc. The individual may also be able to change the images/information on the visualization portion 114 by providing further input to the software.

Gateway 120 may be provided to receive signals, typically wirelessly, from thickness monitoring controller 130 and transmit such signals, typically wirelessly, to platform 110. For example, instead of having each thickness monitoring controller 130 in a facility with its own data plan, it may be more economical to use gateway 120 to establish a cellular connection. In such a case, thickness monitoring controller 130 uses, for example, the XBee protocol to communicate with gateway 120. In another example, if the location of the thickness monitoring controller 130 does not have a good cellular connection, the gateway 120 can be installed at a higher location to establish a cellular connection, and the thickness monitoring controller 130 can be configured using, for example, an XBee protocol. is used to send data to gateway 120.

As best shown in FIG. 3, the thickness monitoring controller 130 includes a modem 131, a microprocessor 132, a pulser 133, an analog-to-digital converter (ADC) 134, an adjustable gain amplifier 135, a transmit channel 136, and a receive channel. Contains 137. Modem 131 is configured to communicate with one or both of platform 110 and gateway 120. Modem 131 may use any suitable communication option, including, but not limited to, XBee915MHz and LTE-M/NB. Modem 131 is configured to communicate with microprocessor 132. Microprocessor 132 may be any type of microprocessor that provides the desired functionality. One such microprocessor 132 is the LPC4370 manufactured and sold by NXP Semiconductors. Microprocessor 132 is configured to communicate with both pulser 133 and ADC 134. Pulsar 133 is preferably a high voltage pulser capacitor. ADC 134 is preferably 16 bit, 2 msps (million samples per second), but other ADC types can be provided as desired. ADC 134 is configured to communicate with an adjustable gain amplifier 135 (also commonly known as a variable gain amplifier). Adjustable gain amplifier 135 preferably has a decibel range of 26 to 54 dB and a frequency range of 10 kHz to 300 kHz, although other ranges may be provided as appropriate. Pulsar 133 is configured to communicate with transmit channel 136 to transmit signals to transmit channel 136 . Adjustable gain amplifier 135 is configured to communicate with receive channel 137 to receive signals from receive channel 137 . Thickness monitoring controller 130 is preferably configured to accommodate a desired number of amplitude scans (“A-scans”) (or waveform displays). In the illustrated embodiment, controller 130 is configured to accommodate 16 A-scans (one from pachymetry ultrasound transducer 140 and five from each of three different probe assemblies 150). Of course, if the number of probe assemblies 150 changes and/or the number of ultrasound transducers 170/180 is included in each probe assembly 150 (as described in further detail below), the controller 130 may It should be appreciated that it may be configured to accommodate more or less than 16 A-scans, depending on the requirements.

Thickness monitoring ultrasound transducer 140 is configured to receive signals from transmit channel 136 of thickness monitoring controller 130 and further configured to transmit signals to receive channel 137 of thickness monitoring controller 130 . As mentioned above, the thickness monitoring ultrasonic transducer 140 is used for standardization purposes and is therefore integrated into a group of ultrasonic transducers (in this example at least one thickness monitoring ultrasonic transducer 170 and at least one area monitoring ultrasonic transducer 170). The acoustic transducer 180) serves to calibrate the measurement system when utilized. Standardized thickness monitoring ultrasound transducer 140 functions to ensure that system 100 always operates in the same manner and functions properly, as required by industry standards. In the illustrated embodiment, standardized thickness monitoring ultrasound transducer 140 is configured to perform a single A-scan. In practice, the thickness monitoring ultrasound transducer 140 is typically placed on a standardized block or thickness-calibrated metal strip to function as a standardized transducer.

As shown in FIG. 1, system 100 includes three different/separate probe assemblies 150A, 150B, 150C (each also referred to as probe assembly 150). Depending on the system 100, the number of probe assemblies 150 provided in the system 100 may be less than three (e.g., one or two) or more than three (e.g., four), as desired. 1, 5, etc.). Depending on the number of probe assemblies 150 provided in the system 100, minor changes/modifications may need to be made to the system 100, as will be understood by those skilled in the art.

As mentioned above, each probe assembly 150 includes a switch assembly 160. As best shown in FIG. 4, the switch assembly 160 includes a power supply 161, a transmit switch 162, a microcontroller 163, a memory 164, a receive switch 165, an amplifier 166, and an optional resistance temperature detector (RTD) interface 167. include. Power supply 161 communicates with transmission channel 136 of thickness monitoring controller 130 . Transmit switch 162 communicates with transmit channel 136 of thickness monitoring controller 130 . Transmit switch 162 preferably has five "switch" channels 162a, 162b, 162c, 162d, 162e, the purpose and function of each of which is described herein. Microcontroller 163 is in communication with transmit channel 136 of thickness monitoring controller 130, transmit switch 162, memory 164, and receive switch 165. Microcontroller 163 may be any type of microcontroller that provides the desired functionality. One such microcontroller 163 is the PIC18 manufactured and sold by Microchip Technology. Preferably, memory 164 is non-volatile memory. The receive switch 165 preferably has four "switch" channels 165a, 165b, 165c, 165d, the purpose and function of each of which will be described below. Amplifier 166 is in communication with receive channel 137 of thickness monitoring controller 130 and receive switch 165 . Amplifier 166 preferably has an amplification of 26-48 dB and a frequency range of 10 kHz to 300 kHz, although other levels/ranges may also be suitably provided. Amplifier 166 is also preferably a two-stage amplifier, providing 26 dB amplification for the single-stage option and 48 dB amplification for the two-stage option, which may populate or depopulate the components on the amplifier board. may be selectable by populating. An optional RTD interface 167 is provided if at least one thickness monitoring ultrasound transducer 170 incorporates an RTD 171 (described below). The illustrated embodiment allows each switch assembly 160 to collect five A-scans (one from the thickness monitoring ultrasound transducer 170 and one from each of the four area monitoring ultrasound transducers 180) by the controller 130. be ordered to do so.

As mentioned above, each probe assembly 150 includes at least one thickness monitoring ultrasound transducer 170. As shown in FIG. 1, each probe assembly 150 includes one thickness monitoring ultrasound transducer 170. Depending on the system 100 and probe assemblies 150, the number of thickness monitoring ultrasound transducers 170 provided within each probe assembly 150 may be two or more (e.g., two, three, four, etc.) as desired. ). As will be appreciated by those skilled in the art, minor changes/modifications may need to be made to the probe assemblies 150 and/or the system 100 depending on the number of thickness monitoring ultrasound transducers 170 provided in each probe assembly 150. is possible. Each thickness monitoring ultrasonic transducer 170 can optionally have an RTD 171 associated with it to measure the temperature of the pipe/vessel at or near the location where the thickness measurement is being performed. Each thickness monitoring ultrasound transducer 170 communicates with the fifth "switch" channel 162e of the transmit switch 162 and, if the thickness monitoring ultrasound transducer 170 includes an RTD 171, also communicates with the RTD interface 167.

Thickness monitoring ultrasound transducer 170 (and 140) operates by generating high frequency ultrasound (eg, 5 MHz). These ultrasound waves are commonly referred to as longitudinal waves (LW), and thus thickness monitoring ultrasound transducer 170 is sometimes referred to as an LW transducer. In the illustrated embodiment, each thickness monitoring ultrasound transducer 170 is configured to perform a single A-scan. Unlike the thickness monitoring ultrasonic transducer 140, the thickness monitoring ultrasonic transducer 170 is not placed on a standardized block or thickness calibrated metal piece, but rather on the pipe/vessel and Measure the thickness of the pipe/vessel where it will be installed.

As mentioned above, each probe assembly 150 includes at least one area monitoring ultrasound transducer 180. As shown in FIGS. 1, 2, 3, 4, and 5, each probe assembly 150 includes four area monitoring ultrasound transducers 180A, 180B, 180C, and 180D (also referred to as area monitoring ultrasound transducer 180, respectively). (called). Depending on the system 100 and probe assemblies 150, the number of area monitoring ultrasound transducers 180 provided within each probe assembly 150 is optionally less than four (e.g., one, two, or three). Or there may be more than four (eg, five, six, etc.). As will be appreciated by those skilled in the art, minor changes/modifications may need to be made to the probe assemblies 150 and/or the system 100 depending on the number of area monitoring ultrasound transducers 180 provided in each probe assembly 150. could be. A first area monitoring ultrasound transducer 180A communicates with a first "switch" channel 162a of transmit switch 162 and a first "switch" channel 165a of receive switch 165. A second area monitoring ultrasound transducer 180B communicates with a second "switch" channel 162b of transmit switch 162 and a second "switch" channel 165b of receive switch 165. A third area monitoring ultrasound transducer 180C communicates with a third "switch" channel 162c of transmit switch 162 and a third "switch" channel 165c of receive switch 165. A fourth area monitoring ultrasound transducer 180D communicates with a fourth "switch" channel 162d of transmit switch 162 and a fourth "switch" channel 165d of receive switch 165.

In one embodiment, probe assembly 150 may include a thickness transducer 170 and a set of area transducers 180 individually wired to switch/preamp assembly 160. In different embodiments, thickness transducer 170 and area transducer 180 can be combined into a single larger probe wired to switch/preamplifier assembly 160 via a single multi-conductor cable. In another embodiment, there may be a larger set of probes (thickness + 2 areas, area + area, etc.).

Area monitoring ultrasound transducer 180 operates by generating low frequency ultrasound (eg, 50kHz to 500kHz). These ultrasound waves are commonly referred to as guided wave (GW), and therefore area monitoring ultrasound transducer 180 is sometimes referred to as a GW transducer. One such type of waveguiding from GW transducers, namely shear horizontal zero waves (referred to as SH0 in plates or T(0,1) in pipes), is interesting due to their non-dispersive behavior. In the illustrated embodiment, each area monitoring ultrasound transducer 180 is configured to perform a single A-scan.

The GW transducer 180 is preferably in the form of a piezoelectric patch transducer, but may alternatively take other forms, such as, for example, a shear plane piezoelectric element. In a preferred embodiment, as best shown in FIGS. 1 and 5, GW transducers 180A, 180B, 180C, 180D are arranged in a rectangular configuration around LW transducer 170, with GW transducer 180A The GW transducer 180B is arranged above and to the left of the LW transducer 170, the GW transducer 180C is arranged below and to the right of the LW transducer 170, and the GW transducer 180D is arranged below and to the left of the LW transducer 170. placed above and to the right of. When applied to a pipe/vessel, a straight line from GW transducer 180A to GW transducer 180B is parallel to a straight line from GW transducer 180C to GW transducer 180D, and a straight line from GW transducer 180A to GW transducer 180D is parallel to the straight line from GW transducer 180A to GW transducer 180D. It is parallel to the straight line from 180B to GW transducer 180C. Further, when applied to a pipe/vessel, a straight line from GW transducer 180A to GW transducer 180C intersects LW transducer 170, a straight line from GW transducer 180B to GW transducer 180D intersects LW transducer 170, and " An "X-shaped" configuration is provided.

System 100, when associated with a pipe/vessel, may be utilized to measure corrosion/erosion of the pipe/vessel. In one embodiment, one method 200 of measuring pipe/vessel corrosion/erosion is described below and illustrated in FIGS. 6, 7, and 8.

Method 200 includes manually measuring 205 the actual longitudinal velocity and temperature of the pipe/vessel being inspected.

Method 200 includes manually measuring 210 the actual waveguiding velocity and temperature of the pipe/vessel being tested.

Method 200 includes performing 215 a standardized thickness measurement using standardized thickness monitoring ultrasound transducer 140 and RTD 171 (similar to thickness monitoring ultrasound transducer 140, standardized thickness monitoring ultrasound transducer 170 is also optional). It should be understood that RTD 171 can optionally be incorporated).

Method 200 includes performing 220 a measurement using probe assembly 150A. Step 220 includes a substep 220a of performing thickness measurements using thickness monitoring ultrasound transducer 170 and RTD 171. Step 220 includes a substep 220b of performing area thickness monitoring using area monitoring ultrasound transducers 180A, 180B, 180C, 180D at a first frequency. Substep 220b includes substep 220b1 of performing an axial scan, whereby area monitoring ultrasound transducer 180A is excited and data is recorded using area monitoring ultrasound transducer 180B. The measurements made in sub-step 220b1 are repeated the number of times specified in the configuration settings and average A-scan. Substep 220b includes substep 220b2 of performing an axial scan, thereby exciting area monitoring ultrasound transducer 180C and recording data using area monitoring ultrasound transducer 180D. The measurements made in sub-step 220b2 are repeated the number of times specified in the configuration settings and average A-scan. Substep 220b includes substep 220b3 of performing a circumferential scan, thereby exciting area monitoring ultrasound transducer 180A and recording data at area monitoring ultrasound transducer 180D. The measurements made in sub-step 220b3 are repeated the number of times specified in the configuration settings and average A-scan. Substep 220b includes substep 220b4 of performing a circumferential scan, whereby area monitoring ultrasound transducer 180C is excited and data is recorded at area monitoring ultrasound transducer 180C. The measurements made in sub-step 220b4 are repeated the number of times specified in the configuration settings and average A-scan. Thus, channels 162a, 162c (associated with GW transducers 180A, 180C) serve as waveguide transmit channels, and channels 162b, 162d (associated with GW transducers 180B, 180D) serve as waveguide receive channels. functions as The receive path further proceeds via amplifier 166 to receive channel 137 of thickness monitoring controller 130 .

Step 220 includes sub-step 220c repeating sub-step 220b at a second frequency, the second frequency being different from the first frequency.

Step 220 includes sub-step 220d repeating sub-step 220b at a third frequency, the third frequency being different from both the first frequency and the second frequency.

Method 200 includes step 225, which includes repeating step 220 to take measurements using probe assembly 150B.

Method 200 includes step 230, which includes repeating step 220 to take measurements using probe assembly 150C.

Accordingly, the method 200 combines ultrasonic thickness monitoring using longitudinal waves with ultrasonic area monitoring using guided waves, and in a preferred embodiment only one special non-dispersive shear wave mode (SH0 or T(0 , 1)). Rather than attempting to create a thickness map, the method 200 obtains representative thickness measurements, and the thickness map detects localized corrosion/erosion between the representative thickness measurement locations. is complemented by area monitoring features. The system 100 generates two separate and different excitation signals from two different types of ultrasound transducers, e.g. LW transducer 170 and GW transducer 180, e.g. high frequency ultrasound (5 MHz) for thickness monitoring and one for area monitoring. Utilizes new electronics that use a single circuit to deliver low frequency ultrasound (50-500 kHz). Each excitation signal must be generated and processed separately. More specifically, pulser 133 of controller 130 is a digital switch that can only deliver predetermined fixed voltage levels: high voltage, low voltage, and zero voltage. The high and low voltage levels are typically adjustable within the ranges of 5V to 90V and -5V to -90V, although different voltage levels are also acceptable. Microprocessor 132 signals pulser 133 to output one of the fixed voltage levels, eg, a high voltage for a specified period of time, to transmit channel 136. Example of a pulse used to excite LW transducer 170: Processor 130 commands pulser 133 to output 0V, then a high voltage for a period of 100ns, then a low voltage for a period of 100ns, then 0V. . The described sequence produces a bipolar square wave at a 5 MHz frequency suitable for exciting the LW transducer 170. For GW transducer 180, different frequencies and signal amplitudes are required.

As best shown in FIGS. 9, 10, 11, and 12, the waveform required to excite the GW transducer 180 is a Hanning window signal (e.g., half-wave) shown as 330 in FIG. It can have a fairly complex shape, for example a 5-cycle sine wave superimposed on a cycle cosine), which allows for a smoother transition from a no-signal state to a signal state. To generate GW transducer 180, an appropriate waveform combination of the digital output of pulser 133, shown as waveform 300 in FIGS. Impedance is used. The impedance of GW transducer 180 in the frequency range used to generate GW waves (50-500 kHz) typically consists mostly of capacitance. This capacitance of the transmit channel 136 and the series resistance described above form a low pass filter. The pulser 133, under instructions from the microprocessor 132, generates a high frequency (typically in the range of tens of MHz) digital waveform 300, which when passed through the capacitance of the transmit channel 136 and the GW transducer 180 (as shown in FIG. ) A waveform 310 different from the one initially output from the pulser 133 is obtained. When the various high frequency digital waveforms from pulser 133 pass through the series resistance of transmit channel 136 and the capacitance of transducer 180, they pass through a range of analog waveforms, such as a sine wave without a Hanning window, shown as 320 in FIG. It is possible to generate a sine wave with a Hanning window, as shown as 330 in FIG. Of course, waveforms other than those described and illustrated can be generated.

In one embodiment, a chirp signal can be used to simultaneously excite multiple frequencies from a single channel. Appropriate software filtering can decode the individual frequency responses from a single A-scan.

By using system 100 and method 200, the time-of-flight and amplitude of echoes reflected from defects on pipes/vessels can be evaluated. More specifically, by transmitting an excitation signal from the GW transducer 180C and receiving it by the GW transducers 180B, 180D, the reflected echoes are transmitted to the GW transducer 180B, 180D that is closer to the damage, e.g. GW transducer 180B if the damage is on the left side of both, or GW transducer 180D if the damage is on the right side of both GW transducers 180B, 180D (GW transducers 180B, 180D are arranged as shown in FIGS. 1 and 5). faster in the time trace in Defects, such as pits or corrosion/erosion patches, typically increase in size over time. Therefore, the amplitude of the echo reflected by the defect increases over time. A permanently installed system therefore makes it possible to monitor changes in amplitude following the flight time. Monitoring A-scan changes after baseline subtraction and digital filtering, for example, reduces analysis complexity and increases the reliability of test results. Next to baseline subtraction, additional digital signal processing tools or machine learning algorithms can be used for feature extraction or pattern recognition, which further increases the confidence level and changes earlier in time. useful for detecting.

FIG. 23 shows a simplified pipe and instrumentation diagram (PID) corresponding to an example corrosion/erosion monitoring system as shown in FIG. 1, according to some embodiments disclosed herein. . Simplified PID 2300 includes multiple probe assemblies shown as circles numbered 19-84. For example, three different/separate probe assemblies 150A, 150B, 150C are shown. Of course, the number of probe assemblies within PID 2300 can be any number as desired. In one example, a human operator/inspector can focus the inspection on a down-selected list of TMLs, as described herein. These down-selected TMLs may represent more efficient candidate measurement locations to capture the global corrosion behavior of the entire asset while still being able to inspect localized corrosion. For example, significant amounts of time/energy and cost are saved by down-selecting the number of TMLs so that only the probe assemblies with the highest probability of detecting localized corrosion are inspected by a human operator/inspector. can do. Rather than checking all of the probe assemblies 19-84, or randomly checking less than all of the probe assemblies 19-84, the down-selected TML is determined by which TML is measured. This is the best identification. In some examples, an inspector can measure wall thickness at numbered locations on the simplified PID 2300 using a handheld or other manual device. In another example, some type of rig or harness is pre-installed at numbered locations on the simplified PID 2300 to allow the inspector to measure wall thickness at each thickness measurement location. may be done. In yet another example, the tester may be an automated machine that takes measurements at specific time intervals at the downselected TML. Even in automated measurement systems, downselecting the TML reduces the amount of processing power and network bandwidth consumed by the measurement data generated by the measurement device at each numbered location on the simplified PID 2300. This is advantageous because it reduces For example, some large industrial facilities may have thousands of probe assemblies, which can generate vast amounts of data. Additionally, once any localized corrosion is identified and repaired, the human operator can indicate its extent so that any model can be updated to reflect the new wall thickness value. . Additionally, in some instances, if localized corrosion is incorrectly identified, supervised human input to machine learning or neural networks running on a digital analytics platform can alert and model that accordingly. can be improved.

FIG. 13A shows an example piping with a sensor 1301 installed on the pipe in accordance with one or more aspects of the features disclosed herein. The pipe may have liquid flow in the direction indicated by the arrow. During testing, one approach may be to test each and all of sensors 1-6 shown in FIG. 13A and take measurements therefrom. In another example, a random selection of sensors may be tested and measured. In accordance with some of the systems and methods disclosed herein, in another example, multiple thickness monitoring locations (TML) indicated on each sensor 1-6 may be intelligently considered. Often, a smaller/narrower set of TMLs may be down-selected for inspection. Further, in accordance with some of the systems and methods disclosed herein, TMLs may be grouped based on one or more criteria in the process of downselecting TMLs. The down-select criteria, in one simplified example, may identify and exclude sensors that have historically measured only gross corrosion in that area (eg, sensors 1 and 3). Thus, by down-selecting, the system 100 avoids using clustering, but instead uses grouping to identify some as unnecessary for evaluating the health of machine components. Downselect the sensor. Therefore, time and resources are saved. In contrast, some conventional systems have attempted to reduce risk by adding more TMLs and inspecting those TMLs. However, the risk-based inspection (RBI) approach described in various aspects of this disclosure provides a superior process and system. The RBI method also uses models that take into account other criteria such as the type of fluid transported within the piping system, the temperature inside and outside the pipe/component, the elbow/configuration of the piping component, and other criteria. May be used. For example, measurements at elbows may be weighted to be more likely to be selected as part of a downselect within a group, because historically, locations near elbows in pipes have been more turbulence and friction, and therefore higher potential for corrosion and acidity.

Referring to FIG. 13B, probe assembly 1302 may include a tether device that captures precise spot measurements of component thickness. In another embodiment, the probe assembly may include a tether device to capture precise spot measurements and area monitoring. For example, the device of FIG. 13B or an equivalent device may be used to capture area monitoring of the thickness of a pipe component. In yet another embodiment, the probe assembly can include a wireless device that captures accurate spot measurements without necessarily being in direct contact with the piping component requiring thickness monitoring. The probe assembly includes a thickness monitoring ultrasonic transducer, an area monitoring ultrasonic transducer, and/or a thickness monitoring ultrasonic transducer configured to verify global corrosion within the piping system (e.g., confirming that no localized corrosion is detected). One or more of these combinations may be provided.

FIG. 13B is a diagram of exemplary piping with sensors installed. Sensor 1302 may be any of a variety of types of sensors configured to measure the thickness of the piping at or near its point of installation on the pipe. The sensor 1302 is typically installed in a permanent location and remains attached to the pipe for an extended period of time (e.g., the life of that circuit of piping, more than 5 years, more than 3 years, or other period of time). . Although the sensor 1302 shown in FIG. 13B is installed outside the piping and tethered with wires, in some examples according to one or more aspects of the present disclosure, the sensor is untethered and one Data can be wirelessly communicated to or above wireless receiver/transceiver devices. Additionally, although the sensors displayed in FIG. 13B are shown in a linear pattern along the length of the pipe, this disclosure contemplates sensors installed in any of several different patterns. For example, the density of installed sensors may be based on the direction of gravity and the type of material being transported within the piping. For example, in one example, assume that the piping of FIG. 13B is transporting liquid from left to right along the length of the pipe, where the bottom of the pipe is the portion of the pipe where sensor 1302 is installed. In such instances, a sensor installed on the pipe would be sensitive to internal conditions within the pipe (e.g. , more liquid contacting the bottom of the pipe than the top of the pipe) may be distributed around the circumference of the piping, taking into account that the inner part of the pipe may be exposed to a greater possibility of deterioration.

FIG. 14 is an example network architecture of an industrial facility with sensors, communication components, and other components in accordance with various aspects of the present disclosure. Data analysis platform 112 may be communicatively coupled to one or more networked components via a network, such as local area network 1408. For example, data analysis platform 112 may output to visualization platform 114 to generate one or more of the example graphs included herein. The monitoring system may include a software platform 112 for remotely monitoring and analyzing historical measurements collected by multiple sensors attached to pipes and components. The monitoring system may include analytical tools for monitoring, diagnosing, and/or predicting areas that are candidates for localized corrosion (e.g., because the system was unable to confirm global corrosion for the area). Can be done. By using the analysis system disclosed herein, the TML can be optimized to reduce, ie down-select, the number of measurement locations without compromising risk, among other things.

In another example, data analysis platform 112 may trigger an alert generated at remote alert device 1410. Remote alert device 1410 may provide for immediate inspection of one or more components, or may result in certain piping components being prioritized for subsequent inspection of the facility.

As measurements and other data are collected by the system 1400, the data may be stored in a data store 1406 that is communicatively coupled to and accessible to the data analysis platform 112. In some examples, data may be stored in computer memory 1404, but the amount of computer memory required may be large. Alternatively, in some examples, a model 1412, such as a machine learning artificial neural network, may be stored in computer memory 1404 for execution by processor 1402, while historical data and other data may be stored in data store 1406. obtain. In some examples, the data store may be moved to platform 112, but for illustrative purposes is shown communicating with platform 112 via local area network 1408.

FIG. 15 is an illustration of grouping multiple sensors (eg, probe assemblies) in one embodiment of the present disclosure. Each probe assembly can be assigned a unique TML identifier (TML ID), as shown in FIG. The TML ID may be any unique character, symbol, or other identifier that uniquely identifies each TML (i.e., probe assembly). In FIG. 15, the thick rectangular box around the ID number of the selected TML indicates the grouping of probe assemblies. At 1502, on March 7, 2007, the system grouped probe assemblies 4, 5, 6, and 7 into a group based on one or more rules. At 1504, on March 7, 2008, the graphical representation of data stored in computer memory 1404 shows that system 1400 adjusted the grouping to include/exclude one or more TMLs. At 1504, the model may recommend that the probe assembly corresponding to TML ID number 4 should no longer be part of the group ID corresponding to the bold rectangular box at 1504. As a result, the one or more probe assemblies downselected for that group ID may also change. Finally, at 1506, on March 7, 2009, the graphical depiction shows that the system 1400 further adjusts the grouping, now grouping probe assemblies 5 and 6 into the first group ID, probe assemblies 7 and 8 into a different/separate second group ID. As a result, the downselect and risk profile changes for the entire system 100, as described below in FIG. 16.

In one example, grouping of TMLs into group IDs may be done in one of several different ways. For example, the initial grouping of each circuit of components in a facility may be based on measured data levels. For each date a measurement was taken by the probe assembly, a new group may be triggered if the probe assembly satisfies any of the following conditions: (i) if the probe assembly is the first TML in the circuit; (ii) ) If the (absolute) difference between the measurement and the preceding TML measurement is greater than about 0.5 to about 3.0 standard deviations of all measurements on that day, then the value of this parameter becomes more conservative. (iii) the nominal wall thickness measurement of the TML is equal to the nominal wall thickness measurement of the preceding TML, which may be decreased for aggressive grouping or increased for more aggressive grouping; If the comparison is different, or if the TML has only one measurement historically (across all dates). In another example, TML grouping may be performed in a multi-step process. In the first step, a group of connected components (e.g., circuits) is obtained on a particular date (or any other predetermined time period, e.g., within a one-hour window, within the same week, or other). All measurements can be compared to determine the number of pairs (or tuples) measured on a particular date. In one example, any TML pair that has less than a predetermined percentage (e.g., 70%, 80%, 60%, 75%, or other percentage) of total measurements within that survey year (or other time period) is removed. be done. Next, the minimum measurement value of all TMLs may be identified, and all TMLs paired with that TML in an earlier (eg, first) step are assigned to the same group ID. Other examples of rules for grouping TMLs will be apparent to those skilled in the art after reviewing the entire disclosure herein.

Additional other example rules for grouping TMLs are contemplated in this disclosure. For example, in some rules, groupings may be reassigned based on TML pairing percentages. For component circuits with at least two measurement dates, TML pairs that are grouped together at least a predetermined threshold percentage number of times may be kept in the same group, but TMLs that do not meet this threshold are can be individually assigned to distinct groups using more than one rule. In yet another example, measurement dates that do not have sufficient TML may be dropped. For each component circuit, the system 1400 may, in some examples, only consider measurement dates that have at least a predetermined threshold percentage of the maximum number of TMLs for any given date. TMLs that occur on dates that do not meet this threshold may be individually assigned to separate groups.

In some examples, system 1400 discards (e.g., drops) seemingly invalid measurements based on the lack of historical data and performs TML analysis based on one or more of the rules described herein. You may proceed to regroup the . The threshold used is a hyperparameter that can be adjusted based on the diversity and quality of the dataset. This adjustment may be made at the end of the process during data confirmation and validation. In one example, the threshold percentage may be set to 75%, but for some TMLs, previous measurements may not have been taken in years. In some embodiments, a hypergrid may be generated and used to adjust parameters and/or hyperparameters of system 1400. In some examples, the threshold setting may be strongly correlated to the number of TML measurements that the system 1400 collects for each TML ID. Accordingly, the threshold may be adjusted up or down based on how much data is made available to the system 1400.

16 and 17 illustrate graphical plots of various data collected and/or analyzed by system 1400. FIG. 16A shows a probability plot graph of measurements (normal is 95%) with percentages on the Y-axis and measurements on the X-axis. The system 1400 defaults to assuming that global corrosion is detected unless the plot shows that the tail does not extend vertically, as shown near the top of the graph in FIG. 16A. Data analysis platform 112 may perform one or more steps to verify that the probe assembly pipe wall thickness measurements are global corrosion and not localized corrosion. For example, in some embodiments, validation generates a probability plot of all pipe wall thickness measurements associated with the piping system, and then groups the plotted pipe wall thickness measurements by nominal thickness; To confirm that the corrosion is likely not global corrosion, this may be done by identifying non-linear relationships in the probability plot of pipe wall thickness measurements grouped by nominal thickness. On the other hand, if the graph shows a linear relationship, the TML corresponding to those data points in the graph is indicative of overall corrosion. This technique is an advance over systems that may have used standard deviations to construct normal probability plots. Furthermore, the verification step provides further assurance that the system 1400 is operating to accurately detect gross corrosion and accordingly down-select the appropriate probe assemblies installed on components within the facility. Add. System 1400 should not generate alerts (eg, from device 1410) for global corrosion because it is widespread and typically not the primary target during inspection. Rather, overall corrosion is considered in the scheduling and planning of bulk replacement of components within the facility.

Referring to FIGS. 16B and 16C, these graphs illustrate the relationship between the risk of falsely identifying gross corrosion and the amount of thickness measurement locations (TML). Although the amount of risk asymptotes to a threshold minimum amount of risk 1602 regardless of the number of measurement locations, FIG. 16B shows that the level of risk illustrated for the amount of probe assembly (i.e., TML) shows that TML decreases as more TML is added. On the other hand, the effect of the systems and methods disclosed herein is illustrated in FIG. 16C, which shows the shift in the curve showing the illustrated level of risk versus the amount of TML after downselecting. show. 16B and 16C are described in more detail below in connection with the method steps illustrated in the flowchart of FIG. 22. On the other hand, FIG. 17 is a graph showing the cumulative thickness distribution of a tube with naphthenic acid corrosion in an existing system known in the art.

FIG. 18A is a corrosion sensor analysis graph showing TML measurements by date for a particular circuit ID (or asset ID). The X-axis corresponds to TML identifiers. For practical purposes, probe assemblies installed on a piping system may be assigned identifiers sequentially or in any other order along the circuit formed by the piping system. Each TML may have an ID indicating its upstream or downstream location on the pipe. Other data cleaning and/or scrubbing of TML based on location data may be performed to harmonize/standardize measurement data for analysis. Each TML may be assigned a nominal thickness from the time the pipe is first installed. One or more publicly available databases (eg, the Meridian database) may provide data including nominal thickness measurements and specifications. On the other hand, as the legend on the right side of FIG. 18A indicates, measurements can be taken over a period of time so that historical data over at least several years (ie, a long period of time) can be stored and analyzed. In this example, approximately 25 years of wall thickness measurement data is stored, analyzed, and plotted in FIG. 18A. Graphical plot 1802 in FIG. 18A corresponds to measurements taken on August 3, 2015. Meanwhile, other plots in the graph correspond to thickness measurements taken for each TML on corresponding dates going back approximately 25 years (eg, an extended period of time).

Data analysis platform 112 may set one or more hyperparameters for model 1412 corresponding to the graph plotted in FIG. 18A. Hyperparameters are typically set before the training/learning process begins on the model. In contrast, values for other parameters are derived through model training. In FIG. 18A, a graphical user interface for adjusting the grouping_sensitivity hyperparameter is displayed at the top. Visual platform 114 may include graphical tools/sliders to which hyperparameters may be adjusted. In FIG. 18A, the grouping_sensitivity hyperparameter is shown set to a "standard" setting. Meanwhile, in FIG. 18B, which shows another view of model 1412, the grouping_sensitivity hyperparameter is shown set to a "medium" setting. As a result, the number of groups is only 61 in FIG. 18B instead of 97 groups in FIG. 18A. Furthermore, the graph 1812 plotted in FIG. 18B differs slightly from the graph 1802 of FIG. 18A due to changes in hyperparameter settings and TML selection methods. Furthermore, if the grouping_sensitivity hyperparameter was set to "high" in FIG. 18C, the graph 1822 plotted in FIG. 18C would be even different from FIGS. 18A and 18B. The number of groups is approximately 75, but the total number of TMLs remains constant at 155.

The grouping_sensitivity hyperparameter refers to the sensitivity or aggressiveness of TML grouping and may be applied at the initial grouping stage. In some examples, assigning a TML to a new group when the (absolute) difference between the measurement and the previous TML measurement is greater than one standard deviation (SD) of all measurements for that day. Can be done. This threshold may be adjusted for more conservative or aggressive grouping. A threshold of less than 1 SD results in a grouping that is more sensitive to changes in measurements and results in a more conservative grouping. On the other hand, a threshold greater than 1 SD makes the grouping less sensitive to changes in measurements, leading to more aggressive grouping (eg, higher grouping ratios). In one example, as shown in FIG. 18C, five different grouping_sensitivity may be implemented with decreasing sensitivity from most conservative to most aggressive as follows: High (0. 5SD), standard (1SD), medium (1.5SD), low (2SD), and very low (3SD). In another example, more or fewer groups than the five groupings described above may be used to provide finer or coarser granularity. Since the grouping_sensitivity hyperparameter is applied in the initial grouping step as in the case of hyperparameters, all subsequent grouping steps can be re-executed based on the initial grouping result, i.e. the entire grouping cycle is , repeated five times, once for each of the five grouping sensitivity levels.

In particular, Figures 18A, 18B, and 18C (collectively referred to as "Figure 18") illustrate multiple TML selection methods that may be applied to measurements to optimize data point grouping and plotting. Enumerate. Figure 18 shows three optimization functions: median_TML_within_group ID, minimum_average_TML_within_group ID, and minimum_variation_from_mean. (minimum variation from the average) is listed, but Other optimization functions may be used according to one or more aspects of this disclosure. For example, the TML_position optimization function can be used and if one TML is selected, the TML in the center of the group is selected. If two TMLs are selected, the group is split into two subgroups and the TML in the center of each subgroup is selected, and so on. Other examples of TML selection methods are envisioned herein. For example, the optimization function may be a minimum_average_TML_within_groupID optimization function. In the minimum_average_TML_within_groupID method to determine which TML to choose from each group, the method selects the TML with the lowest average measurement (across dates) within each group. For example, in one exemplary system that uses the minimum_average_TML_within_groupID optimization function, the system calculates the average measurement for each TML (across dates) and groups the TMLs within each group by average measurement (e.g., in ascending order). The first n TMLs can be ranked and selected based on the number (n) of TMLs selected from each group. Similarly, the median_TML_within_groupID optimization function is similar to the minimum_average_TML_within_groupID optimization function, but is based on the median instead of the minimum average.

In another example, the optimization function may be a minimum_variation_from_mean optimization function. In the minimum_variation_from_mean_method for determining which TML to select from each group, the method selects the TML with the minimum mean variation from the mean measurements of the group. For example, in one exemplary system using the minimum_variation_from_mean optimization function, the system may calculate an average group measurement for each date. Then, for each TML, calculate the absolute difference from the mean for each date, and calculate the average variation (eg, absolute difference from the mean) for each TML. The minimum_variation_from_mean optimization function then ranks the TMLs in each group by mean variation (e.g., in ascending order) and selects the first n TMLs based on the number (n) of TMLs selected from each group. Select.

After completing grouping, system 1400 determines the TML candidate selection method and desired number of probe assemblies per group. The number of candidates per group may be another hyperparameter. By default, the system 1400 may use the greater of 1% or 1 for each TML group. If more than one TML candidate is selected, the TML group may be divided into equally sized subgroups while maintaining TML order. System 1400 can then apply a TML candidate selection method based on one or more scenarios described herein.

FIG. 19A is a graph 1902 showing the measured TC thickness of a component in millimeters over time. Alternatively, the graphs may include temperature calibration (in degrees Celsius), temperature coefficient (e.g. 1%), corrosion rate ST (mm/year), corrosion rate LT (mm/year), remaining component life (years), remaining half-life. The period (years) and actual thickness (mm) may also be indicated.

Additionally, FIG. 19B is a modified graph showing FSH (percentage) values versus thickness values (in millimeters or other units). The graph also shows the thickness range of gate A 1904 and gate B 1906. Alternatively, the graph may illustrate mV values instead of FSH. Additionally, in some examples, the graph may be displayed as HF instead of being modified.

FIG. 20A is a graph 2002 of an acid battery installation showing measured TC thickness in millimeters of components over time. Alternatively, the graphs may include temperature calibration (in degrees Celsius), temperature coefficient (e.g. 1%), corrosion rate ST (mm/year), corrosion rate LT (mm/year), remaining component life (years), remaining half-life. The period (years) and actual thickness (mm) may also be indicated. Additionally, FIG. 20B is a modified graph of an acid battery installation showing FSH (percentage) values versus thickness values (in millimeters or other units). The graph also shows the thickness range of gate A 2004 and gate B 2006. Alternatively, the graph may illustrate mV values instead of FSH. Additionally, in some examples, the graph may be displayed as HF instead of being modified.

FIG. 21 shows a simplified example of an artificial neural network 2100 on which machine learning algorithms may be implemented. FIG. 21 is a diagram illustrating an example of nonlinear processing using an artificial neural network; other forms of nonlinear processing may be used to implement machine learning algorithms in accordance with the features described herein. .

In FIG. 21, each of the input nodes is connected to a first set of processing nodes. The external source 2102 provided to the input node may be a metric from the results through the steps of the methods disclosed herein. Each of the first set of processing nodes is connected to each of the second set of processing nodes. Each of the second set of processing nodes is connected to each of the output nodes. Although only two sets of processing nodes are shown, any number of processing nodes may be implemented. Similarly, although only four input nodes, five processing nodes, and two output nodes per set are shown in FIG. 21, any number of nodes per set may be implemented. The data flow in FIG. 21 is depicted from left to right, and data may be input to input nodes, flow through one or more processing nodes, and output by output nodes. Input to the input node may originate from an external source 2102. Output 2104 may be sent to a feedback system 2106 and/or a data store. Feedback system 2106 can send output to input nodes for successive processing iterations with the same or different input data.

In one example method of using feedback system 2106, the system can use machine learning to determine the output. Outputs may include leak area boundaries, multi-sensor detection events, confidence values, and/or classification outputs. The system may use suitable machine learning models including xg boosted decision trees, autoencoders, perceptrons, decision trees, support vector machines, regression, and/or neural networks. Neural networks may include feedforward networks, radial basis networks, recurrent neural networks, long/short-term memories, gated recurrent units, autoencoders, variational autoencoders, convolutional networks, residual networks, Kohonen networks, and/or other types. may be any suitable type of neural network, including. In one example, output data in a machine learning system may be represented as a multidimensional array, an extension of a two-dimensional table (such as a matrix) to data with higher dimensions.

A neural network may include an input layer, several hidden layers, and an output layer. Each layer may have its own weight. The input layer may be configured to receive as input one or more feature vectors described herein. The intermediate layer may be a convolutional layer, a pooling layer, a dense (fully connected) layer, and/or other types. The input layer can pass input to the intermediate layer. In one example, each intermediate layer can process output from the previous layer and then pass the output to the next intermediate layer. The output layer may be configured to output classification or actual values. In one example, layers within a neural network may use activation functions such as sigmoid functions, Tanh functions, ReLu functions, and/or other functions. The neural network may also include a loss function. The loss function may measure the number of missed positives in some examples. Alternatively, the number of false positives may be measured. A loss function may be used to determine the error when comparing the output value and the target value. For example, when training a neural network, the output of the output layer may be used as a prediction and compared to the target value of the training instance to determine the error. The error can be used to update the weights in each layer of the neural network.

In one example, a neural network may include techniques for updating weights in one or more of the layers based on error. Neural networks may use gradient descent to update weights. Alternatively, the neural network may use an optimizer to update the weights at each layer. For example, the optimizer may use various techniques or combinations of techniques to update the weights at each layer. If appropriate, the neural network may include mechanisms to prevent overfit regularization (such as L1 or L2), dropouts, and/or other techniques. Neural networks may also increase the amount of training data used to prevent overfitting.

In one example, FIG. 21 illustrates nodes that can perform various types of processing, such as discrete calculations, computer programs, and/or mathematical functions implemented by a computing device. For example, an input node may comprise logical inputs from different data sources, such as one or more data servers. A processing node may include parallel processes running on multiple servers within a data center. The output node may also be a logical output that is ultimately stored in a resulting data store, such as the same or different data server as the input node. In particular, the nodes need not be distinct. For example, two nodes in any two sets may perform exactly the same processing. The same node may be repeated for the same or different sets.

Each of the nodes may be connected to one or more other nodes. A connection can connect the output of a node to the input of another node. Connections may be correlated with weighting values. For example, one connection may be weighted as more important or significant than another, thereby influencing the degree of further processing as the input traverses the artificial neural network. Such connections may be modified such that artificial neural network 2100 may learn and/or be dynamically reconfigured. Although nodes are shown in FIG. 21 as having connections to only consecutive nodes, connections may be formed between any nodes. For example, one processing node may be configured to send output to a previous processing node.

Input received at an input node may be processed through processing nodes, such as a first set of processing nodes and a second set of processing nodes. Processing may result in an output at an output node. Processing may include multiple steps or sequences, as indicated by the connections from the first set of processing nodes and the second set of processing nodes. For example, a first set of processing nodes may be a coarse data filter and a second set of processing nodes may be a more detailed data filter.

Artificial neural network 2100 may be configured to accomplish decision making. As a simplified example for illustration, artificial neural network 2100 may be configured to detect faces in photos. The input node may be provided with a digital copy of the photo. The first set of processing nodes may each be configured to perform a particular step for removing non-facial content, such as large contiguous sections of red color. The second set of processing nodes may each be configured to look for a rough approximation of the face, such as face shape and skin tone. Multiple subsequent sets can further refine this process, each looking for additional, more specific tasks, and each node performing some form of operation that does not necessarily have to be performed in furtherance of that task. Execute processing. Artificial neural network 2100 may then predict locations on the face. Predictions may be accurate or inaccurate.

Feedback system 2106 may be configured to determine whether artificial neural network 2100 made the correct decision. The feedback may include an indication of correct answers and/or an indication of incorrect answers and/or a degree of accuracy (eg, a percentage). For example, in the facial recognition example provided above, the feedback system 2106 is configured to determine whether a face was correctly identified and, if so, what percentage of the face was correctly identified. obtain. The feedback system may already know the correct answer so that it can train the artificial neural network 2100 by indicating whether it made the correct decision. The feedback system can include human input, such as an administrator telling artificial neural network 2100 whether it made the correct decision. The feedback system may provide feedback (e.g., an indication that a previous output was correct or incorrect) to the artificial neural network 2100 via an input node, or may send such information to one or more nodes. obtain. The feedback system may additionally or alternatively be coupled to storage such that the output is stored. The feedback system may not have a correct answer at all, but instead the feedback may be based on further processing, for example, the feedback system may be such that the artificial neural network 2100 uses the results of a manually programmed system. A system programmed to identify faces may be provided to allow comparison of results.

Artificial neural network 2100 can be dynamically modified to learn and provide better input. For example, based on previous inputs and outputs and feedback from feedback system 2106, artificial neural network 2100 may modify itself. For example, processing at the nodes may vary and/or connections may be weighted differently. Following the example provided earlier, the face prediction may have been inaccurate because the photos provided to the algorithm were colored to make all faces appear red. Therefore, a node that excludes sections of the photo that contain large neighboring sections in red may be considered unreliable, and connections to that node may be weighted significantly less. Additionally or alternatively, nodes may be reconfigured to process photos differently. Modifications may be predictions and/or inferences by artificial neural network 2100, and thus artificial neural network 2100 may change its nodes and connections to test hypotheses.

Artificial neural network 2100 need not have a set number of processing nodes or set of processing nodes, but its complexity may be increased or decreased. For example, artificial neural network 2100 may determine that one or more processing nodes are unnecessary or should be reused, and may discard or reconfigure the processing nodes accordingly. As another example, artificial neural network 2100 may determine that further processing of all or a portion of the input is required and add additional processing nodes and/or sets of processing nodes based thereon.

The feedback provided by feedback system 2106 may be merely reinforcement (e.g., providing an indication of whether the output is accurate or inaccurate, giving a number of points to the machine learning algorithm, etc.) , or specific (e.g., providing accurate output). For example, a machine learning algorithm may be asked to detect faces in a photo. Based on the output, the feedback system provides a score (e.g., 75% accuracy, an indication that the guess was accurate, etc.) or a specific response (e.g., specifically identifying where the face was located). can be shown. In one example, a human operator/examiner can focus the inspection on a down-selected list of TMLs. If any localized corrosion is identified and repaired, the human operator can indicate the amount so that the model 1412 can be updated to reflect the new wall thickness value. Additionally, in some instances, localized corrosion may be incorrectly identified in the system 1400 and supervised human input to machine learning or neural networks performed in the digital analytics platform 112 may The alerts and models can be refined accordingly.

Artificial neural network 2100 may be supported or replaced by other forms of machine learning. For example, one or more of the nodes of artificial neural network 2100 may be a decision tree, an associative rule set, logical programming, a regression model, a cluster analysis mechanism, a Bayesian network, a propositional formula, a generative model, and/or other algorithm or algorithm. A form of decision may be implemented. Artificial neural network 2100 may implement deep learning.

FIG. 22 is a flowchart illustrating example steps of a method 2200 performed in accordance with some embodiments disclosed herein. Method 2200 may be performed by system 1400 when computer-executable instructions stored on a non-transitory computer-readable medium are executed by a processor. The method 2200 can, among other things, down-select among probe assemblies installed in piping systems of industrial facilities. As a result, the system and method for optimized asset health monitoring allows representative measurement locations to be identified through down-selecting and the remaining probe assemblies to be used regularly on piping systems and other components within an industrial facility. This is improved since it can be ignored during inspection.

22, in step 2202, the system stores in computer memory 1406 communicatively coupled to processor 1402 historical pipe wall thicknesses collected over a period of time from probe assembly 150A installed on a piping system within an industrial facility. Memorizes measured values. At step 2204, data analysis platform 112 may set one or more hyperparameters, such as, but not limited to, a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, a group_size hyperparameter, and/or a combination thereof. Once the hyperparameters are set, in step 2206 the system 1400 trains the model using at least the historical pipe wall thickness measurements stored in computer memory 1406 and the hyperparameter values stored in computer memory 1404. You may start.

At step 2208, the model stored in computer memory 1404 may group the first set of probe assemblies from among the multiple probe assemblies installed in the piping system. As described in this disclosure, such as with respect to FIG. 15, several methods are provided in which grouping can be performed in a model. After grouping the probe assemblies, data analysis platform 112 may assign a unique group identifier (group ID) to each set of probe assemblies. The unique group ID can be any identifier that the system 1400 can use to uniquely reference a group of probe assemblies.

At step 2210, data analysis platform 112 selects an optimization function for operation of system 1400 based at least on the trained model. Without limitation, the median_TML_within_groupID optimization function, the minimum_average_TML_within_groupID optimization function, the minimum_variation_from_mean optimization function, and/or the TML_position(TM A number of example optimization functions are described in this disclosure, including the L position) optimization function. The decision to select a particular optimization function leads to subsequent identification and measurement steps. For example, in steps 2212A, 2212B, 2212C, and 2212D (collectively "steps 2212"), system 1400 determines the pipe wall thickness of the piping system based on the model stored in computer memory 1404 and the selected optimization function. Identify probe assemblies corresponding to each group ID for monitoring. In some examples, system 1400 can identify a single probe assembly for an entire group ID representing the area being measured. In other examples, multiple probe assemblies may be identified as representing a group ID. The number of TMLs assigned to be downselected from a group may be based on one or more rules. This is based on the value of the maximum standard deviation of the groups, in one example, if any group has a maximum standard deviation of 0.25 or less, one TML is selected from the group and the maximum SD is 0.25. , the number of TMLs selected increases by 1 (ie, if 0.25-0.5, two probe assemblies can be selected from the group, etc.). Additionally, the 0.25 step value can be modified to adjust the sensitivity of TML selection. Decreasing the 0.25 value will lead to more TMLs being selected in each group (i.e. a conservative approach) and increasing the 0.25 value will lead to fewer TMLs being selected in each group. leading to being selected (i.e., proactive approach).

At step 2214, during testing, the system 1400 may exclude the probe assemblies identified from each group ID and ignore all remaining probe assemblies within each group ID. The system may measure the wall thickness of each probe assembly identified for each group ID, but excludes other probe assemblies within the group ID. Accordingly, the system down-selects among multiple probe assemblies installed in the piping system. At least one advantage of downselecting the number of probe assemblies used during testing is the resulting time savings. For example, a human inspector who may have previously inspected each probe assembly can now inspect measurements with a reduced number of probe assemblies without substantially increasing the risk of missing dangerous localized corrosion. can. In one example, in step 2214, the system 1400 can output a human readable report listing probe assemblies for which wall thickness measurements should be manually inspected by a human inspector. The output may be ordered in any of a variety of ways, such as based on highest risk of localized corrosion, based on geographic convenience from the human inspector's known starting location, or other orders. obtain.

For example, when the amount of risk is graphed against the number of measurement locations taken, as shown in FIGS. 16B and 16C, the amount of risk increases even though the number of measurement locations increases. , asymptotically approaches the threshold minimum risk amount 1602. Importantly, as the amount of measurements decreases, the delta change in risk increases at an increasing rate, as shown by graph 1604, or in other words, decreases the number of probe assemblies sampled. can increase the risk to a dangerous amount. However, the system 1400 and method 2200 disclosed herein shifts the graph from the initial risk graph 1606 to a more preferred risk graph 1608. Therefore, down-select the number of probe assemblies required to be actively checked during inspection by identifying what is statistically most likely to be gross corrosion/deterioration to the pipe wall. Doing so results in reduced testing time/cost while simultaneously maintaining (or even reducing) the risk profile.

Finally, in step 2216 of FIG. 22, the thickness monitoring controller 130 may receive and transmit probe assembly pipe wall thickness measurements from each group ID for inspection. Thickness monitoring controller 130 may send measurement data (and any other data) to data store 1406 for historical recording and analysis and to data analysis platform 112 for analysis and generation of visualizations. can. For example, wall thickness measurements indicate that a particular segment of pipe within a plumbing system has deteriorated other than general corrosion, such that it has risen to the level of dangerous localized corrosion and must be replaced within a specified period of time. It can be shown that the person is suffering from In another example, pipe wall thickness measurements may be performed on one or more of pipes, tanks, vessels, and/or pipelines in a facility.

While particular embodiments have been shown and described with respect to the drawings, it is contemplated that various modifications may be devised by those skilled in the art without departing from the spirit and scope of the claims. Accordingly, this disclosure and the appended claims are not intended to be limited to the particular embodiments illustrated and discussed with respect to the drawings, but modifications and other embodiments may be made within the scope of this disclosure and the accompanying drawings. It will be understood that it is intended to be included within. Furthermore, while the foregoing description and associated drawings describe example embodiments in the context of certain example combinations of elements and/or features, different combinations of elements and/or features may be of interest to the present invention. It is to be understood that alternative embodiments may be provided without departing from the scope and scope of the appended claims. Furthermore, the above description describes how to enumerate the performance of several steps. Unless otherwise stated, one or more steps within a method may not be required, one or more steps may be performed in a different order than stated, and one or more steps may be performed in a different order than stated. may be formed substantially simultaneously. The various aspects are capable of other embodiments and of being practiced or carried out in a variety of different ways. It is to be understood that the expressions and terminology used herein are for purposes of description and are not to be considered limiting. Rather, the words and terms used herein are to be given their broadest interpretation and meaning. The use of "including" and "comprising" and variations thereof is meant to include the subsequently listed items and their equivalents, as well as additional items and their equivalents.

Claims (20)

A method for down-selecting among probe assemblies installed in a piping system, the probe assembly configured for pipe wall thickness monitoring, the method comprising:
setting a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter for the model before training the model;
the model executed on a processor groups the first set of probe assemblies based at least on historical pipe wall thickness measurements collected from the probe assemblies installed in the piping system over a period of time; And,
assigning a unique group ID to each set of probe assemblies;
selecting, by the model, an optimization function among a plurality of optimization functions for the model after training the model;
identifying, by the model, a single probe assembly corresponding to each group ID for pipe wall thickness monitoring of the piping system;
transmitting pipe wall thickness measurements of the single probe assembly from each group ID for inspection by a thickness monitoring controller associated with the piping system;
including methods. 2. The probe assembly of claim 1, wherein the probe assembly comprises at least a resistance temperature detector, a thickness monitoring ultrasonic transducer, and an area monitoring ultrasonic transducer configured to detect localized corrosion within the piping system. Method described. verifying that the single probe assembly pipe wall thickness measurements are global corrosion and not localized corrosion;
generating a probability plot of all pipe wall thickness measurements associated with the piping system;
grouping the plotted pipe wall thickness measurements by nominal thickness;
3. The method of claim 2, further comprising: failing to identify a nonlinear relationship in a probability plot of pipe wall thickness measurements grouped by nominal thickness to confirm global corrosion. During testing, all remaining probes within each group ID except for the single probe assembly from each group ID in order to reduce the number of test samples measured without compromising the risk profile of the piping system. 2. The method of claim 1, further comprising downselecting by ignoring the assembly. The grouping of the first set of probe assemblies further comprises at least inspection information provided to a system and historical pipe wall thickness measurements collected over a period of time from the probe assemblies installed in the piping system. The method according to claim 1, based on. The plurality of optimization functions include a median TML within a group (median_TML_within_group) ID, a minimum average TML within a group (minimum_average_TML_within_group) ID, and a minimum variation_from_me 2. The method of claim 1, comprising: an). 7. The method of claim 6, wherein the plurality of optimization functions include TML position (TML_position). 2. The method of claim 1, wherein the piping system comprises a tank, and a first of the probe assemblies is configured to measure wall thickness of the tank. 5. The pipe wall thickness monitoring includes measuring the thickness of a pipe wall at a particular probe assembly, the pipe wall being located in one or more of a pipe, a tank, a vessel, and a pipeline. The method described in 1. 10. The method of claim 9, wherein the pipe wall thickness monitoring includes analyzing original wall thickness, wall thickness loss over time, calibration errors, and measurement position repeatability errors by the probe assembly. storing historical pipe wall thickness measurements collected over time from the probe assembly installed in the piping system in a computer memory communicatively coupled to a processor;
2. The method of claim 1, further comprising: training a model using at least the historical pipe wall thickness measurements stored in the computer memory by the processor. 12. The method of claim 11, wherein the model includes an artificial neural network. A system for detecting localized corrosion on multiple components transporting material over a distance, the system comprising:
a plurality of probe assemblies attached to one or more of the components, each of the plurality of probe assemblies corresponding to a unique identifier;
a data store configured to store historical wall thickness measurements collected over a period of time from measurements performed by the probe assembly;
a model trained on historical wall thickness measurements in the data store using hyperparameters including at least a grouping sensitivity hyperparameter;
A monitoring device comprising a processor and a memory storing computer-executable instructions, the computer-executable instructions, when executed by the processor, causing the system to:
grouping the first set of probe assemblies based on the model;
assigning a unique group ID to each set of probe assemblies;
selecting an optimization function from a plurality of optimization functions based on the model;
identifying a probe assembly for each group ID for wall thickness monitoring of the component based on the model and the selected optimization function, each group ID corresponding to an identified probe assembly; a step corresponding to a unique identifier;
a monitoring device configured to perform a step including: outputting a list of the unique identifiers corresponding to an arbitrary group ID;
A system equipped with. The plurality of probe assemblies are identified from each group ID, the plurality of probe assemblies comprising at least a thickness monitoring ultrasonic transducer and an area monitoring ultrasonic transducer configured to detect localized corrosion to the component. comprises two or more probe assemblies of a plurality of probe assemblies, and a memory of the monitoring device stores computer-executable instructions that, when executed by the processor, cause the system to To,
transmitting wall thickness measurements of the probe assembly from each group ID for inspection by a thickness monitoring controller associated with the component;
during testing, ignoring all remaining probe assemblies within each group ID except for two or more probe assemblies from each group ID;
14. The system of claim 13, further comprising: verifying that wall thickness measurements of two or more probe assemblies from each group ID fail to identify gross corrosion. 14. The system of claim 13, wherein probe assembly wall thickness measurements from a first group ID include wall thicknesses of pipe components in the probe assembly. 14. The system of claim 13, wherein probe assembly wall thickness measurements from a first group ID include wall thicknesses of tank components in the probe assembly. The hyperparameters include a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter, and the plurality of optimization functions include a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter; roup) 14. The system of claim 13, comprising an ID, a minimum_average_TML_within_group ID, a minimum_variation_from_mean, and a TML position. a non-transitory computer-readable medium storing computer-executable instructions, the computer-executable instructions, when executed by a processor, causing the system to:
storing in a computer memory communicatively coupled to the processor historical pipe wall thickness measurements collected over a period of time from probe assemblies installed in the piping system;
setting hyperparameters for the model;
training a model by the processor using at least historical pipe wall thickness measurements stored in the computer memory;
grouping the first set of probe assemblies by the model running on the processor;
assigning a unique group ID to each set of probe assemblies;
selecting an optimization function from a plurality of optimization functions based on the model;
identifying a probe assembly corresponding to each group ID for pipe wall thickness monitoring of the piping system based on the model and the selected optimization function;
transmitting pipe wall thickness measurements of the probe assembly from each group ID for inspection by a thickness monitoring controller associated with the piping system; a non-transitory computer-readable medium for down-selecting among probe assemblies; The hyperparameters include at least one of a grouping_sensitivity hyperparameter, a threshold_measurements hyperparameter, and a group_size hyperparameter, and the plurality of optimization functions TML median (median_TML_within_group) ID,
19. The non-transitory computer-readable medium of claim 18, comprising a minimum average TML within group ID, a minimum variation from mean, and a TML position. further storing computer-executable instructions, which, when executed by the processor, cause the system to:
19. The non-transitory computer-readable method of claim 18, wherein the non-transitory computer-readable method of claim 18 causes the non-transitory computer-readable method to perform the steps comprising excluding the probe assembly identified from each group ID and ignoring all remaining probe assemblies within each group ID during testing. Medium.

Images