Systems and methods for inspecting an object for a defect

Abstract

There is provided a system for inspecting a metal object. The system includes a magnetic field generator to generate a magnetic field to become incident upon the metal object. The magnetic field produces a magnetic response by the metal object. The system also includes a sensor module to detect the magnetic response and generate magnetic response data. The sensor module and the metal object are moveable relative to one another along a direction of motion. The system also includes a controller in communication with the sensor module. The controller may determine based on the magnetic response data whether the metal object has a defect. A method for inspecting a metal object for a defect is also provided.

Description

[0001] SYSTEMS AND METHODS FOR INSPECTING AN OBJECT FOR A DEFECT

[0002] CROSS-REFERENCE TO RELATED APPLICATION

[0003] [1] This application claims the benefit of, and priority from, United States Provisional Patent Application No. 63 / 776,899, filed on March 24, 2025, which is incorporated herein by reference in its entirety.

[0004] TECHNICAL FIELD

[0005] [2] This specification relates to systems and methods for inspecting an object for a defect, and in particular to systems and methods for inspecting a metal object for a defect.

[0006] BACKGROUND

[0007] [3] Some objects may contain a defect as a result of their manufacturing process.

[0008] Objects may also develop defects during the course of their use or operation. These defects may degrade the properties or performance of these objects, or lead to their functional failure. Detecting such defects may allow for mitigation measures to be put in place against performance degradation or unexpected functional failures.

[0009] SUMMARY

[0010] [4] According to an aspect of the present specification there is provided a system for inspecting a metal object, the system comprising: a magnetic field generator to generate a magnetic field to become incident upon the metal object, the magnetic field to produce a magnetic response by the metal object; a sensor module to detect the magnetic response and generate magnetic response data, the sensor module and the metal object being moveable relative to one another along a direction ofmotion, the sensor module comprising a first sensor array comprising a first set of sensors and a second sensor array comprising a second set of sensors, the first sensor array spaced from the second sensor array along a longitudinal direction being along the direction of motion, the first set of sensors being offset relative to the second set of sensors along a lateral direction being transverse to the direction of motion; and a controller in communication with the sensor module, the controller to determine based on the magnetic response data whether the metal object has a defect.

[0011] [5] The system and the metal object may be moveable relative to one another along the direction of motion.

[0012] [6] The system may further comprise a transportation mechanism to propel the system relative to the metal object along the direction of motion.

[0013] [7] The magnetic field may comprise one or more of: a first magnetic field that is along the longitudinal direction; and a second magnetic field that is along the lateral direction.

[0014] [8] The magnetic field may movably adhere the system to the metal object.

[0015] [9] The sensor module may comprise a magnetic sensor.

[0016]

[0010] The sensor module may further comprise one or more additional sensor types comprising one or more of an ultrasound sensor, a camera, a laser scanner, and an eddy current sensor, each additional sensor type having a corresponding additional sensor type data.

[0017]

[0011] Each additional sensor type data may comprise corresponding depth data.

[0018]

[0012] The controller may be further to modulate at least one of the magnetic response data and the one or more additional sensor type data based on one or more of the remaining of the magnetic response data and the one or more additional sensor type data.

[0013] To modulate the at least one of the magnetic response data and the one or more additional sensor type data the controller may be to exclude the at least one of the magnetic response data and the one or more additional sensor type data.

[0019]

[0014] The controller may comprise one or more of: a data acquisition module to capture the magnetic response data; a data processing module to process the magnetic response data to generate processed data; and an inspection controller module to receive the processed data, and control one or more of: operation of the data acquisition module; and operation of the data processing module.

[0020]

[0015] The processed data may comprise a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of the sensor module relative to the metal object; and the controller may be to determine based on the data structure whether the metal object has the defect.

[0021]

[0016] The inspection controller module may be to control, based on one or more of the processed data and whether the metal object has the defect, one or more of: motion of the system relative to the metal object; and operation of the sensor module.

[0022]

[0017] The data processing module may comprise a machine learning model to determine based on the magnetic response data whether the metal object has the defect.

[0023]

[0018] The machine learning model may comprise an autoencoder model.

[0024]

[0019] The machine learning model may comprise a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model.

[0025]

[0020] The data processing module may be to determine a position of the system relative to the metal object associated with the magnetic response data, the position being the position of the system along the direction of motion.

[0021] According to another aspect of the present specification there is provided a method of inspecting a metal object for a defect, the method comprising: generating, using a magnetic field generator, a magnetic field to become incident upon the metal object; moving the metal object and a sensor module relative to one another along a direction of motion; detecting, using the sensor module, a magnetic response of the metal object to the magnetic field, to generate magnetic response data; processing the magnetic response data into a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of the sensor module relative to the metal object; and determining, at a controller and based on the data structure, whether the metal object has the defect.

[0026]

[0022] Two or more of the magnetic field generator, the sensor module, and the controller may be part of a system for inspecting the metal object; and the moving the metal object and the sensor module may comprise moving the metal object and the system for inspecting the metal object relative to one another.

[0027]

[0023] The detecting the magnetic response may comprise detecting the magnetic response using the sensor module comprising a first sensor array comprising a first set of sensors and a second sensor array comprising a second set of sensors, the first sensor array spaced from the second sensor array along a longitudinal direction being along the direction of motion, the first set of sensors being offset relative to the second set of sensors along a lateral direction being transverse to the direction of motion.

[0028]

[0024] The system for inspecting the metal object may comprise a transportation mechanism; and the moving the system and the metal object relative to one another may comprise using the transportation mechanism to propel the system relative to the metal object along the direction of motion.

[0029]

[0025] The generating the magnetic field may comprise generating the magnetic field comprising one or more of: a first magnetic field that is along the longitudinal direction; and a second magnetic field that is along the lateral direction.

[0026] The method may further comprise movably adhering the system to the metal object using the magnetic field.

[0030]

[0027] The sensor module may comprise a magnetic sensor.

[0031]

[0028] The sensor module may further comprise one or more additional sensor types comprising one or more of an ultrasound sensor, a camera, a laser scanner, and an eddy current sensor, each additional sensor type having a corresponding additional sensor type data.

[0032]

[0029] Each additional sensor type data may comprise corresponding depth data.

[0033]

[0030] The method may further comprise the controller modulating at least one of the magnetic response data and the one or more additional sensor type data based on one or more of the remaining of the magnetic response data and the one or more additional sensor type data.

[0034]

[0031] Modulating at least one of the magnetic response data and the one or more additional sensor type data may comprise the controller excluding the at least one of the magnetic response data and the one or more additional sensor type data.

[0035]

[0032] The controller may comprise: a data acquisition module to capture the magnetic response data; a data processing module to process the magnetic response data into the data structure; and an inspection controller module to receive the data structure, and control one or more of: operation of the data acquisition module; and operation of the data processing module.

[0036]

[0033] The method may further comprise: the inspection controller module controlling, based on one or more of the data structure and whether the metal object has the defect, one or more of: motion of the system relative to the metal object; and operation of the sensor module.

[0037]

[0034] The determining whether the metal object has the defect may comprise the data processing module using a machine learning model to determine based on the data structure whether the metal object has the defect.

[0038]

[0035] The machine learning model may comprise an autoencoder model.

[0036] The machine learning model may comprise a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model.

[0039]

[0037] The determining whether the metal object has the defect may comprise the data processing module determining a position of the system relative to the metal object associated with the magnetic response data, the position being the position of the system along the direction of motion.

[0040]

[0038] According to yet another aspect of the present specification there is provided a mobile inspection system for inspecting metal objects, the system comprising: a transportation mechanism to propel the mobile inspection system; a magnetic field generator to produce at least one magnetic field proximal to a metal object; and a sensing system to detect, capture, and process the magnetic response of the metal object to the magnetic field.

[0041]

[0039] The transportation mechanism may further comprise an adhesion mechanism whereby an attractive force is generated between the mobile inspection system and the metal object.

[0042]

[0040] The magnetic field generator may further comprise a magnet producing a magnetic field which is disposed longitudinally along the metal object in the direction of the motion produced by the transportation mechanism.

[0043]

[0041] The magnetic field generator may further comprise a magnet producing a magnetic field which is disposed in a transverse direction, substantially orthogonal to the direction of the motion produced by the transportation mechanism.

[0044]

[0042] The magnetic field generator may further comprise at least two magnets, producing a magnetic field which is disposed in both longitudinal and transverse directions to the direction of motion produced by the transportation mechanism.

[0045]

[0043] The adhesion mechanism may further comprise an alignment mechanism, the alignment mechanism to maintain the lateral position of the mobile inspectionsystem as the transportation mechanism propels the mobile inspection system along the metal object.

[0046]

[0044] The magnetic field generator may further be used to perform at least one of the functions of the adhesion mechanism and the alignment mechanism.

[0047]

[0045] The mobile inspection system may further comprise: a vehicle controller to align the mobile inspection system with the metal object in the lateral direction and to control the transportation mechanism; a communication system to send and receive information from a wayside station; a camera system to capture visual information; and an energy storage system to provide electrical power to the transportation mechanism.

[0048]

[0046] One or more of: the metal object is a train rail, the transportation mechanism comprises an electric motor coupled to a wheel system, the communication system comprises a wireless radio, the energy storage system comprises a battery, the alignment mechanism comprises a permanent magnet attached to a pivoting bogie, and the sensing system comprises an array of digital magnetic sensors communicating with a programmable logic device.

[0049]

[0047] According to yet another aspect of the present specification there is provided a system for inspecting metal objects, the system comprising: a magnetic field to produce a magnetic response by an object; a sensor module to detect the magnetic response; a data acquisition module to capture the measurement data by the sensor module; a data processing module to process the measurement data; and an inspection controller module receiving the output of the data processing module, the inspection controller to adjust at least one of the operation of the data acquisition module and the operation of the data processing module.

[0050]

[0048] The data processing module may further comprise at least one of a lateral localization system and a longitudinal localization system, the lateral localization system to determine the lateral position of the magnetic response, the longitudinal localization system to determine the longitudinal position of the magnetic response.

[0049] The inspection controller may further be to adjust at least one of the rate, gain, noise level, lateral phasing, longitudinal phasing, sampling duration, filtering, and activation state of the sensor module in response to the output of the data processing module.

[0051]

[0050] The magnetic field may further be produced by a magnetic field generator comprising at least one permanent magnet.

[0052]

[0051] The magnetic field may further be produced by a magnetic field generator comprising at least one electromagnet.

[0053]

[0052] The inspection controller may further be to adjust the longitudinal phasing of the sensor module to increase or decrease the longitudinal resolution of a second portion of the sensor module in response to the measurement data produced by a first portion of the sensor module.

[0054]

[0053] The increasing or decreasing of the longitudinal resolution may further be to perform at least one of: increase or decrease the time between the measurements made by sensors comprising the second portion of the sensor module, and increase or decrease the vehicle speed during the detection of the magnetic response by the second portion of the sensor module.

[0055]

[0054] The inspection controller may further be to produce a classification signal comprising information relevant to the magnetic response of the metal object, the classification signal to be received by an operator, and to be stored in a database.

[0056]

[0055] According to yet another aspect of the present specification there is provided a method for inspecting metal objects, the method comprising: generating a magnetic field incident upon a metal object; detecting the magnetic response of a metal object to the incident magnetic field; processing the data produced by the magnetic response into an at least two-dimensional data structure; analyzing the at least two- dimensional data structure to produce a classification output for the condition of the metal object subject to the incident magnetic field.

[0056] The detecting the magnetic response may further comprise reading data from a plurality of sensors disposed in a sensor module, the sensor module comprising: a first portion of a sensor module disposed laterally across a metal object in a first direction; and a second portion of a sensor module disposed laterally across a metal object, the second portion offset from the first portion along the surface of a metal object.

[0057]

[0057] The reading data from a plurality of sensors may further comprise recording simultaneous electrical signals produced by a plurality of sensors.

[0058]

[0058] The recording simultaneous electrical signals may further be achieved by a parallel processing module, the parallel processing module comprising a series of input terminals such that at least one of a plurality of digital communication buses and a plurality of analog signals can be attached to the parallel processing module independently.

[0059]

[0059] The detecting the magnetic response may further comprise: detecting a variation in the magnetic field proximal to the first portion of the sensor module; localizing the detected variation in the lateral and longitudinal directions along the surface of the metal object; and adjusting the sampling rate of the second portion of the sensor module to improve the quality of the combined dataset produced by reading data from the first portion and the second portion.

[0060]

[0060] The adjusting the sampling rate may further comprise: measuring the longitudinal speed of the sensor module along the metal object; computing the desired relative position of additional measurements on the surface of the metal object; and commanding a plurality of sensors to perform measurements at the desired positions.

[0061]

[0061] The adjusting the sampling rate may further comprise: measuring the longitudinal speed of the sensor module along the metal object; computing the desired relative position of additional measurements on the surface of the metal object; and adjusting the longitudinal speed of the sensor module.BRIEF DESCRIPTION OF THE DRAWINGS

[0062]

[0062] Some example implementations of the present specification will now be described with reference to the attached Figures, wherein:

[0063]

[0063] Fig. 1 A shows a schematic representation of an example system for inspecting a metal object.

[0064]

[0064] Fig. 1 B shows a profile view of an example mobile inspection system located on a metal object.

[0065]

[0065] Fig. 2 shows a profile view of an example mobile inspection system comprising additional components.

[0066]

[0066] Fig. 3 shows an example magnetic field generator located above a metal object, in addition to two example sensor modules disposed proximally to the metal object.

[0067]

[0067] Fig. 4 shows an example sensor module comprising a first portion and a second portion, disposed above a metal object containing a defect.

[0068]

[0068] Fig. 5 shows an example phased measurement comprising three separate measurements conducted by three portions of a sensor module.

[0069]

[0069] Fig. 6 shows an example comprising two sensor modules connected to parallel processing units.

[0070]

[0070] Fig. 7A shows a flowchart of an example method for inspecting a metal object for a defect.

[0071]

[0071] Fig. 7B is a flowchart showing another example inspection method.

[0072]

[0072] Fig. 8 is a flowchart showing an example phased measurement method.

[0073] DETAILED DESCRIPTION

[0074]

[0073] In this specification, elements may be described as “to” perform one or more functions, “configured to” perform one or more functions, or “configured for” such functions. In general, an element that is to perform, configured to perform, orconfigured for performing a function is enabled to perform the function, or is suitable for performing the function, or is adapted to perform the function, or is operable to perform the function, or is otherwise capable of performing the function.

[0075]

[0074] It is understood that for the purpose of this specification, language of “at least one of X, Y, and Z” and “one or more of X, Y and Z” can be construed as X only, Y only, Z only, or any combination of two or more items X, Y, and Z (e.g., XYZ, XY, YZ, ZZ, and the like). Similar logic can be applied for two or more items in any occurrence of “at least one ...” and “one or more...” language.

[0076]

[0075] Unless the context requires otherwise, throughout this specification the word “comprise” and variations thereof, such as, “comprises” and “comprising” are to be construed in an open, inclusive sense, that is as “including, but not limited to.”

[0077]

[0076] As used in this specification, the singular forms “a,” “an,” and “the” include plural referents unless the content clearly dictates otherwise. It should also be noted that the term “or” is generally employed in its broadest sense, that is as meaning “and / or” unless the content clearly dictates otherwise.

[0078]

[0077] Non-destructive inspection systems may be used to detect defects and other anomalies present in metal objects. For example, magnetic non-destructive inspection systems may be used to detect corrosion and fatigue in load-bearing longitudinal structures such as rails and pipelines. The capability of the inspection system to detect an individual defect may depend on the location of the defect, the size of the defect, the depth of the defect within the metal object under inspection, the rate at which the inspection system is moved across the region containing the defect, the sensitivity of the inspection system to the particular type of defect, and the performance of the inspection system itself. Improvements in performance, appropriate selection of the type of inspection system, appropriate utilization of the inspection system, and operator skill may all increase the chances of success.

[0079]

[0078] Fig. 1A shows a schematic representation of an example system 100a for inspecting a metal object 110. Fig. 1A is shown from the perspective of a top plan view. Moreover, in Fig. 1A the various components of system 100a are shown asschematic, functional modules. It is contemplated that in some examples, the relative sizes and positions of the components of system 100a may be different than those shown in Fig. 1A. Metal object 110 may comprise a rail, pipe, sheet, and the like. While Fig. 1A shows system 100a as being wider than metal object 110, it is contemplated that in some examples the relative sizes and dimensions of system 100a and metal object 110 may be different than those shown in Fig. 1A.

[0080]

[0079] System 100a comprises a magnetic field generator 112 to generate a magnetic field to become incident upon metal object 110. This magnetic field may produce a magnetic response by metal object 110. In some examples, magnetic field generator 112 may comprise a permanent magnet, an electromagnet, and the like. Examples of the magnetic field and response are described in greater detail in relation to Fig.

[0081] 3.

[0082]

[0080] System 110a also comprises a sensor module 114 to detect the magnetic response and generate corresponding magnetic response data. In some examples, sensor module 114 and metal object 110 may be moveable relative to one another. Moreover, in some examples, this movement may be along a direction of motion 116. While Fig. 1 A shows a one-directional arrow as depicting direction of motion 116, it is contemplated that in some examples the movement may be in both directions, e.g. forward and backward, along direction of motion 116. In addition, in some examples where the metal object is elongated, such as rails or pipes, the direction of motion may be along the longitudinal axis of the metal object.

[0083]

[0081] In some examples, metal object 110 may be moved relative to sensor module 114 or system 110a. Moreover, in some examples sensor module 114 or system 110a may be moved relative to metal object 110. In some such examples, system 100a may comprise a transportation mechanism 118 to propel system 100a relative to metal object 110 along direction of motion 116. Some examples of the transportation system are described in greater detail in relation to Figs. 2 and 3. In Fig. 1 A transportation system 118 in shown in dashed line to signify that in some examples system 100a need not comprise a transportation system.

[0082] It is contemplated that in some examples the magnetic field or sensor module 114 may move relative to metal object 110. It is also contemplated that in some examples system 100a and metal object 110 may be moveable relative to one another along direction of motion 116. System 100a also comprises a controller 120 in communication with sensor module 114. Controller 120 may determine based on the magnetic response data whether metal object 110 has a defect 122. Example approaches to the determination of whether metal object 110 has defect 122 are described in greater detail in relation to Figs. 1B-8.

[0084]

[0083] In system 100a, sensor module 114 comprises a first sensor array 124 and a second sensor array 126. Sensor array 124 comprises a first set of sensors 128. Sensor array 126 comprises a corresponding set of sensors, which may also comprise one or more sensors 128. In sensor module 114, sensor array 124 is spaced from sensor array 126 by a distance 130 along a longitudinal direction being along direction of motion 116. Moreover, the first set of sensors of sensor array 124 are offset relative to the second set of sensors of sensor array 126 along a lateral direction 132 being transverse to direction of motion 116. In Fig. 1A the first and second sets of sensors are offset from one another by an offset distance 134. In addition, while Fig. 1A shows lateral direction 132 being perpendicular to direction of motion 116, it is contemplated that in some examples lateral direction 132 need not be exactly perpendicular to direction of motion 116, and may deviate from perpendicular.

[0085]

[0084] This offset arrangement of sensor arrays 124 and 126 allows for higher resolution measurements across the width of metal object 110. Each sensor 128 has a physical size. This size may be determined by the size of the various components of the sensor package, such as the sensor itself, the substrate (e.g. printed circuit board (PCB)) containing the sensor, connection or wiring ports or outlets, and the like. This physical size of the sensors places a limit on how closely the sensors can be positioned beside one another along a target dimension of the metal object, such as the width of the object, and the like. This distance is depicted as a minimum packing distance 136 shown in Fig. 1A. By organizing sensors 128 into two arrays124 and 126 whose sensors are offset laterally by offset distance 134, the effective coverage of sensors 128 across metal object 110 may be increased, thereby increasing the effective sensing resolution of sensors 128 across metal object 110. In some examples, offset distance 134 may be smaller than minimum packing distance 136.

[0086]

[0085] In order to combine the sensor readings from arrays 124 and 126, controller 120 may track the position on metal object 110 along direction of motion 116 where sensors 128 of each of arrays 124 and 126 make their measurements. For example, sensors 128 of first sensor array 124 may make their measurements when they are closest to defect 122. Controller 120 tracks that position and the sensor data from sensor array 124. Then system 100a moves along direction of motion 116 and sensors 128 of second sensor array 126 also make their measurements when sensors 128 of second sensor array 126 are closest to defect 122. Then controller 120 can combined the measurements made by sensor arrays 124 and 126 at the same position on metal object 110 (e.g. the position of defect 122) to obtain a higher effective sensing or measurement resolution along lateral direction 132, i.e. across the width of metal object 110. This combining of sensor data from sensor arrays 124 and 126 may also be described as interleaving the sensor data. This combining or interleaving may provide an example of phased measurements of metal object 110. Other examples of phased measurements are described in greater detail in relation to Figs. 1B-8.

[0087]

[0086] Distance 130 between sensor arrays 124 and 126 may be based on the physical size and packaging of arrays 126 and 126, sensor module 114, and system 110a. Moreover, distance 130 may also be related to the sensor data gathering and processing speed of controller 120 as well as the optimal or desirable movement speed of system 100a along direction of motion 116. A relatively smaller distance 130 would mean that either sensor data gathering and processing speed of controller 120 may need to be relatively higher, or the movement speed of system 100a long direction of motion 116 may need to be relatively slower.

[0087] While Fig. 1A shows two sensor arrays 124 and 126 each having four sensors 128, it is contemplated that in some examples sensor module 114 may have a different number of sensors per sensor array, or a different number of sensor arrays. In some examples, the positions or arrangement of the sensors or sensor arrays may also be different than those shown in Fig. 1 A.

[0088]

[0088] In some examples, the magnetic field generated by magnetic field generator 112 may comprise one or more of a first magnetic field that is along the longitudinal direction and a second magnetic field that is along the lateral direction. Different types of magnetic field directions may allow for different types of measurement of metal object 110. For example, a magnetic field along the longitudinal direction may allow for magnetic flux leakage inspection. In addition, such a magnetic field may allow for different sensing modes at the leading edge of the system (e.g. eddy current sensing) as compared to the trailing edge (e.g. DC-based detection).

[0089] Similarly, a magnetic field that is along the lateral direction may provide different distributions of magnetic flux and eddy currents in metal object 110, that allow for corresponding sensing modes.

[0090]

[0089] Furthermore, in some examples, the magnetic field generated by magnetic field generator 112 may also be used to movably adhere system 100a to metal object 110. Such examples may include situations where gravity or mechanical coupling are not sufficient, or not used, to movably couple system 100a to metal object 110.

[0091]

[0090] As described above, in some examples sensor module 114 comprises magnetic sensors 128. It is also contemplated that in some examples system 100a, or sensor module 114, may comprise one or more additional sensor types. Such sensor types may include an ultrasound sensor, a camera, a laser scanner, an eddy current sensor, and the like. Each of these additional sensor types may have a corresponding additional sensor type data. In other words, each of these sensor types may measure or collect corresponding sensor data.

[0092]

[0091] Moreover, in some examples, each sensor type data includes corresponding depth data. In some examples where the sensor is capable of depth measurement or measurement at different depths, depth data may be part of the sensor datacaptured or measured by the sensor. For example, an ultrasound sensor may be able to make measurements at different depths. In other examples, a depth indication may be assigned to the sensor data. For example, a visual camera may be able to image only the surface of metal object 110. As such, this type of camera image data may be assigned a corresponding depth indication.

[0093]

[0092] In addition, in some examples, the depth of a measurement made by a sensor may be dependent on other operating or sensing parameters of system 100a. For example, some properties or responses of metal object 110 may be time-dependent, which makes the sensor measurement a function of the time that elapses between excitation of metal object 110 and measurement of the metal object’s response. Eddy current and some types of magnetic and acoustic sensing my fall within this time-dependent category. For example, in the case of an acoustic sensing (e.g. ultrasound) a shorter time interval between excitation and sensing may correspond to a shallower depth in metal object 110, whereas a longer time interval between excitation and sensing may correspond to a deeper depth in metal object 110.

[0094]

[0093] The time interval between excitation and sensing may depend on different sensing or operating parameters of system 100a, such as the speed of sampling of the sensor data, the speed of movement of system 100a relative to metal object 110, and the like. In such examples, controlling the time interval between excitation and sensing (e.g. by controlling the sampling rate or the speed of movement of system 100a) may allow for controlling the depth within metal object 110 that is being sensed to examined. This depth data, in turn, may be included as part of the sensor data.

[0095]

[0094] Moreover, in some examples, controller 120 may modulate one or more of the sensor data types based on one or more of the other, or remaining, sensor data types. The modulated data types may include one or more of the magnetic response data and the additional sensor type data. In some examples, this type of modulation may include determining the degree of weight or importance that is given to the modulated sensor data. Furthermore, in some examples, modulating may include excluding the modulated sensor data from further analysis or determination of defect122. For example, if visual (camera) sensor data indicates that there is a pit on the surface of metal object 110, controller 120 may modulate data from sensors that rely on continuous, high-quality contact with the surface of metal object 110, such as acoustic sensors, and the like.

[0096]

[0095] In addition, in some examples, controller 120 may comprise one or more of a data acquisition module 138, a data processing module 140, and an inspection controller module 142. Modules 138, 140, and 142 may be functional modules or physical modules of controller 120. It is contemplated that in some examples two or more of modules 138, 140, and 142 may be combined together into the same physical or functional module. It is also contemplated that in some examples, the function of all of modules 138, 140, and 142 may be performed by a single physical or functional module that is, or is a part of, controller 120. In Fig. 1A modules 138, 140, and 142 are shown in dashed lines to indicate that in some examples controller 120 need not have separate or distinct modules 138, 140, or 142, and the functions of one or more of modules 138, 140, or 142 may be performed by controller 120 as a whole.

[0097]

[0096] In some examples, data acquisition module 138 may capture the magnetic response data. Moreover, in some examples, data processing module 140 may process the magnetic response data to generate processed data. Furthermore, in some examples, inspection controller module 142 may receive the processed data and control one or more of the operation of data acquisition module 138 and the operation of data processing module 140. Such controlling of the operation of data acquisition module 138 may include controlling from which sensors data is collected, sampling rates, use of physical noise or cut-off filters, and the like. In addition, such controlling of the operation of data processing module 140 may include modulations of data from different sensor types, and the like.

[0098]

[0097] In some examples, inspection controller module 142 may control, based on one or more of the processed data and whether the metal object has the defect, one or more of: the motion of system 100a relative to metal object 110, and the operation of sensor module 114. For example, if an eddy current sensor detects a surface defect,module 142 may control system 100a to back-up or slow down to allow a visual (camera) sensor to take high-quality photos of the surface defect. Module 142 may also control the operation of sensor module 114, for example by determining which sensors to use, the operating parameters of those sensors, and the like. Functions and features of modules 138, 140, and 142 are described in greater detail in relation to Figs. 1B-8.

[0099]

[0098] In addition, in some examples, data processing module 140 may determine a position of system 100a relative to metal object 110 associated with the magnetic response data. This position may be the position of system 100a along direction of motion 116. In other words, module 140 may correlate the magnetic response data to the position on metal 110 where that response was measured.

[0100]

[0099] In some examples, the processed data generated by module 140 may include a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of sensor module 114 or system 100a relative to metal object 110. In some such examples, controller 120 may then determine based on this data structure whether metal object 110 has defect 122. Tracking the position associated with the magnetic response may allow system 100a to determine where defect 122 is located along metal object 110, and also to perform interleaving and phased measurement operations.

[0101]

[0100] It is also contemplated that in some examples sensor data from different or additional sensor types may be included in the data structure. In such examples, the data structure may have one or more additional dimensions beyond two dimensions, with each additional dimension corresponding to sensor data of an additional sensor type.

[0102]

[0101] Moreover, in some examples, data processing module 140 may comprise a machine learning model to determine based on the magnetic response data whether metal object 110 has defect 122. In cases where system 100a comprises additional sensor types, data from those additional sensor types may also be used by the machine learning model to determine if object 110 has defect 122.

[0102] In some examples, this machine learning model may comprise an autoencoder model. This type of model can learn from sensor data as system 100a operates and gathers sensor data. In comparison to classifier models which require preexisting labeled training datasets, the use of an autoencoder may allow system 100a to gather and learn from its own sensor data without being limited by the need for a preexisting labeled defect dataset to train on.

[0103]

[0103] In addition, in some examples, the machine learning model may comprise a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model. This type of hierarchical structure may comprise two or more constituent machine learning models.

[0104]

[0104] In some examples, one or more of the constituent machine learning models may perform one or more of the following: determination of whether a metal object may contain a defect based on a single sensor type data; determination of what type of defect may be present based on a single sensor type data; correlation of data from all sensor types based on presence and type of defect; assessment of defect presence based on data from all sensor types; determination of defect characteristics based on data from all sensor types and environment; and the like.

[0105]

[0105] As discussed above, a hierarchical machine learning model may include multiple layers or multiple constituent machine learning models. For example, a hierarchical model may include a first layer that uses a single sensor type to assess presence of a defect. A second layer (e.g. classification) may be possible for pre-trained types. As system 100a continues to operate and gather sensor data, this type of a pretrained classifier may become possible based on a growing library based on segmented data from the field. A third layer may unify all the sensors into one output, and attempt to blend in high level knowledge about metal object conditions. A fourth layer may use the unified dataset to judge whether a high or low confidence level has been achieved, potentially with statistical significance. A fifth level may blend in environmental data including temperature, moisture, time of day, weather, and system 100a chassis vibration to judge whether the detection statistics alignwith historical data and metal object conditions. Machine learning models with other types, structure, and functions are also contemplated.

[0106]

[0106] It is contemplated that in some examples system 100a may include the features and functions described in relation to the systems and methods described in association with Figs. 1B-8. It is also contemplated that in some examples the systems and methods described in relation to Figs. 1 B-8 may include the features and functions described in association with system 100a.

[0107]

[0107] Fig. 1B shows a representation of a mobile inspection system 100b disposed on top of a metal object 200. The mobile inspection system 100b comprises a plurality of modules, which may serve to support both the vehicular and inspection functions of the mobile inspection system 100b. The mobile inspection system 100b is propelled by the transportation mechanism 101 and adhered to the metal object 200 by the adhesion mechanism 102. Together, these two modules may support the mobility of the mobile inspection system 100b.

[0108]

[0108] It is contemplated that in some examples inspection system 100b need not comprise adhesion mechanism 102. For example, if system 100b is built using a four-wheeled rail carriage platform, an adhesion mechanism may not be necessary, as the weight of system 100b may keep system 100b on the rails. In examples where gravity (i.e. weight of system 100b) may not be sufficient to moveably couple system 100b to the metal object, adhesion mechanism 102 may be used to assist with that moveable coupling.

[0109]

[0109] The mobile inspection system 100b also comprises a magnetic field generator 103 to produce a magnetic field 201 (shown in Fig. 3) which is incident on the metal object 200. A sensing system 104 is located proximally to the magnetic field generator 103 as determined by the strength and distribution of the magnetic field 201. In some examples, sensing system 104 may also be described as a sensor module. The mobile inspection system of Fig. 1B also comprises a data acquisition module 105, a data processing module 106, and an inspection controller module 107. In some examples, one or more of data acquisition module 105, data processing module 106, and inspection controller module 107 may be described as,be contained in, or be a functional module of a controller. These modules may support the acquisition of data from the sensor module 300 (shown in Fig. 3), processing of that data into an appropriate format, and control of the inspection process.

[0110]

[0110] Fig. 2 shows a representation of a mobile inspection system 100b comprising a plurality of modules which may support the vehicle functions of the mobile inspection system 100b. Fig. 2 shows several components in relation to system 100b which are not shown in Fig. 1 B. It is contemplated that while these components are shown in the example of system 100b shown in Fig. 2, in other examples of system 100b one or more of these components need not be present. The vehicle controller 500 may be used to control the speed, lateral alignment, longitudinal position, battery state of charge, and temperature of the mobile inspection system 100b. The communication system 501 may be used to communicate with a wayside station, a user, or other external devices. Commands, signals, and data may all be sent and received using the communication system 501. In some examples, system 100b may comprise a camera 502. Camera 502 may face along the direction of motion of system 100b. In other words, camera 502 may be a forward-facing camera. In some examples, camera 502 may allow system 100b to image the larger scale condition or defects of the metal object. Moreover, in some examples, camera 502 may be used to allow for some autonomy or autonomous functions for system 100b.

[0111]

[0111] The energy storage system 503 may be used to provide electrical power to the transportation mechanism 101 , as well as to provide electrical power to all of the other electrical components present in the mobile inspection system. The electric motor 504 is a component of the transportation mechanism 101 and may use large amounts of energy from the energy storage system 503 in order to propel the mobile inspection system 100b along the metal object 200 at high velocity. The wheel system 505 may support both lateral and longitudinal positioning of the mobile inspection system and may be directly coupled to the electric motor 504.

[0112]

[0112] In some examples, system 100b may comprise a sensor frame 506. Sensor frame 506 may also be described as a sensor chassis. In some examples, some orall of the sensors of the sensor module, or the sensor module itself, may be secured to sensor frame 506. While mechanically coupled to the rest of system 100b, in some examples sensor frame 506 may be at least partially vibrationally de-coupled from the rest of system 100b. This vibrational de-coupling may be achieved using non-rigid attachments to the rest of system 100b, vibration dampers, and the like. Such vibrational de-coupling may at least partially protect the sensors, and the accuracy of their measurements, from vibrations or other mechanical perturbations affecting the rest of system 100b.

[0113]

[0113] Fig. 3 shows a schematic representation of magnetic field generator 103 producing a magnetic field 201 which is incident on metal object 200 and wherein a magnetic response 202 has been induced in the metal object 200 by the magnetic field 201. Two representations of a sensor module 300 are also shown, the first being located graphically above the metal object, and the second being located graphically below the metal object. “Above” and “below” in this context may also be described as a first side of the metal object, and a second side of the metal object being opposite the first side. These representations may be understood to reflect a plurality of physical locations relative to the magnetic field 201. Indeed, it is contemplated that multiple sensor modules may be present in proximity to the metal object to facilitate a complete inspection. In this context, “proximity” may comprise a sensor module or array being close enough to detect its target effect. In examples where the sensor array comprises magnetic sensors, “proximity” may comprise the magnetic sensors being close enough to one or more of metal object 200 and magnetic response 202, to detect magnetic response 202. These sensor modules may be located in positions wherein they may detect the magnetic response 202 of the metal object 200. Each sensor module is comprised of a plurality of sensors 301. Sensor module 300 may comprise at least two sensors 301 , but in some examples sensor module 300 may comprise at least 16 sensors, and in other examples may comprise at least 64 sensors.

[0114]

[0114] In this description “sensor module” is generally used to refer to a collection of sensors, or collectively to the sensing module or functionality, of a system. Withinthe sensor module, the individual sensors may be grouped into different groupings. These groupings may be described as sensor arrays, portions of sensor modules or sensor arrays, sets of sensors, and the like. Occasionally “sensor module” may be used in place of, or interchangeably with, “sensor array” or “set of sensors”.

[0115]

[0115] The magnetic response 202 of the metal object 200 may be understood to be comprised of electrical currents, electrical fields, magnetic flux, or magnetic fields. Indeed, in some examples the magnetic response 202 may be comprised of a plurality of fields, currents, and fluxes. The representation of the magnetic response 202 shown in Fig. 3 may be understood to contain electrical currents in a circulating pattern which diminishes with distance from the magnetic field generator 103. It is contemplated that in some examples different sensors 301 may be disposed to perform at different distances from the magnetic field generator 103 due to the variation in the amplitude of the magnetic response 202.

[0116]

[0116] In some examples, the magnetic response 202 of the metal object 200 may also be accompanied by additional features which may be detected by a plurality of different mechanisms, including visual, ultrasonic, eddy current, x-ray sensors, and the like. A camera, ultrasonic transducer, eddy current sensor, or x-ray emitter and detector may be used as part of, or in addition to, the sensors 301 to measure such features.

[0117]

[0117] Furthermore, in some examples, sensor module 300 may comprise a plurality of different types of sensors 301 that are placed at locations where they may be optimally suited to detect the magnetic response 202. Sensors 301 which are able to detect large field amplitudes may be placed closer to the magnetic field generator where the magnetic response 202 from the metal object 200 is high, and other sensors 301 which are able to detect small field amplitudes may be placed farther away from the magnetic field generator 103 where the magnetic response 202 from the magnetic field generator 103 is low.

[0118]

[0118] In addition, some examples of the magnetic field generator 103 involve relative motion between the magnetic field 201 and the metal object 200. This may be accomplished by movement of the magnetic field generator 103, by movement of themagnetic field 201, by movement of system 100b, or by movement of the metal object 200. In one such example, the magnetic field generator is affixed to the mobile inspection system 100b as the mobile inspection system 100b translates over the metal object 200. In another example, the magnetic field generator 103 produces a modulating magnetic field 201 similar to the response that would be produced by a translating magnetic field 201 without modulation, which generates a magnetic response 202 in the metal object 200.

[0119]

[0119] Furthermore, the magnetic field generator 103 may comprise a single device, or it may comprise multiple devices arranged throughout the mobile inspection system 100b. The magnetic field generator 103 of Fig. 3 is shown as a single device wherein the magnetic field 201 is produced at a single location. It is also contemplated that in some examples, a particular type of defect 203 (shown in Fig.

[0120] 4) may benefit from multiple devices being utilized in order to generate a magnetic response 202. Additional examples may also comprise two devices producing a closed flux path, wherein the magnetic field 201 is polarized in one direction in one device, and in the opposite direction in the other device. Such examples may be used when a higher magnetic field or a different field distribution is needed.

[0121]

[0120] Fig. 4 shows a representation of sensor module 300, comprised of sensors 301 which may be separated into a plurality of portions. These portions may comprise at least a first portion 302 and a second portion 303. As the mobile inspection system 100b translates over the metal object 200, it may be understood that the first portion 302 of sensor module 300 may interact proximally with a defect 203 located in the metal object 200. In this context interacting proximally may comprise coming within sensing or detection range of, coming within optimal or substantially maximal sensing or detection range of, coming physically near to, coming physically and optimally near to, coming physically and substantially maximally near to, passing over or across, and the like. The first portion may thus be disposed to produce a signal corresponding to the defect 203. It may further be understood that at a later point in time the second portion 303 of the sensor module 300 may interact proximally with the same defect 203, and that the temporal relationship between thetwo points in time at which the first portion 302 and the second portion 303 of sensor module 300 interact proximally with the defect 203 may be calculated if the displacement between the first portion 302 and the second portion 303 of sensor module 300 and the velocity of the mobile inspection system 100b are both known.

[0122]

[0121] Moreover, in some examples the mobile inspection system 100b may utilize the displacement between the first portion 302 and the second portion 303 in combination with the velocity of the mobile inspection system 100b along the metal object 200 to perform a phased measurement of the magnetic response 202 in relation to the defect 203. While Fig. 4 shows first portion 302 and second portion 303 as being different in size, it is contemplated that in some examples first portion 302 and second portion 303 may be the same size or may comprise the same number of sensors. Moreover, in some examples first portion 302 and second portion 303 may be described as a first set of sensors and a second set of sensors respectively. In such examples sensor module 300 may also be described as comprising a first sensor array having the first set of sensors and a second sensor array comprising the second set of sensors.

[0123]

[0122] Fig. 5 shows a representation of an example phased measurement method. In the phased measurement method of Fig. 5, a plurality of portions of the sensor module 300 interact proximally with the defect 203, and the combined signals of each portion are used to improve the spatial resolution of the overall measurement of the defect in the longitudinal direction along the metal object 200. This improvement in spatial resolution may increase the quality of the digital representation of the defect 203 recorded by the mobile inspection system 100b. In addition, the improvement in spatial resolution may improve the ability of the mobile inspection system 100b to calculate the physical size of the defect 203, and it may also reduce the likelihood of magnetic noise producing false signals which may appear as defects.

[0124]

[0123] Furthermore, in some examples the mobile inspection system 100b may be used to inspect a metal object 200 which does not have a smooth surface. In such examples, the wheel system 505 (shown in Fig. 2) may transmit vibration or otherdisturbances to the sensor module 300 as the mobile inspection system 100b translates along the metal object 200. Such vibrations or disturbances may produce large variations in the measurements produced by sensor module 300. This phenomenon is known as “lift-off’ to those skilled in the art. In contemplated examples the phased measurement method of Fig. 5 may be used to detect the liftoff phenomenon while still producing relevant signals to the defect 203.

[0125]

[0124] Fig. 6 depicts two examples of the sensor module 300 comprising differing methods of communication between the parallel processing module 400 and the sensor module 300. In some examples, module 400 may be contained in, or a functional module of, data acquisition module 105. In the example depicted on the left comprising digital communication bus 401, at least a portion of the sensor module 300 is linked to the parallel processing module 400 using a common bus. It is to be understood that sensor module 300 may be linked to parallel processing module 400 using multiple digital communication busses 401. In the example depicted on the right comprising analog signals 402, each sensor 301 of sensor module 300 produces an individual analog output which may be linked to the parallel processing module 400.

[0126]

[0125] In examples comprising digital communication busses 401, sensors 301 may be addressed, selected, or otherwise communicated with individually so as to permit reading data from each sensor. The sensor module 300 may therefore be comprised of a series of different groups of sensors 301 , which may be read from simultaneously on different digital communication busses, or in sequence on a single digital communication bus 401. The groups of sensors may be physically arranged such that simultaneous reads and sequential reads may occur in specific physical locations. Furthermore, sensors which are arranged on a single digital communication bus may be timed to interact at specific intervals with the data acquisition module 105. This may be understood to support the aforementioned functionality of phased measurement per the example shown in Fig. 5.

[0127]

[0126] In examples comprising analog signals 402, sensors are individually related to the data acquisition module. This may permit the data acquisition module 105 toread arbitrary subsets of the sensors 301 simultaneously or in sequence. The data acquisition module 105 may therefore be responsible for determining the timing of each interaction with each sensor 301 and therefore may be the primary interface by which a phased measurement is conducted.

[0128]

[0127] The systems described herein may carry out an inspection process utilizing a magnetic field translated along a metal object. Such an inspection process may comprise a variety of methods, which may permit the system to detect defects in such a metal object, generally by reading data from at least one array of sensors, and processing the data.

[0129]

[0128] Fig. 7A shows a flowchart of an example method 600 for inspecting a metal object for a defect. At box 605, a magnetic field generator may generate a magnetic field to become incident upon a metal object. At box 610, the metal object and a sensor module may be moved relative to one another along a direction of motion. At box 615, a sensor module may be used to detect a magnetic response of the metal object to the magnetic field, to generate magnetic response data.

[0130]

[0129] Moreover, at box 620, the magnetic response data may be processed into a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of the sensor module relative to the metal object. Furthermore, at box 625 a controller may be used to determine, based on the data structure, whether the metal object has the defect. Further details and examples of the functions and components mentioned in relation to boxes 605-625 are described in greater detail in relation to Figs. 1A-6, 7B, and 8.

[0131]

[0130] In some examples, two or more of the magnetic field generator, the sensor module, and the controller may be part of a system for inspecting the metal object. Examples of such a system include systems 100a and 100b. Moreover, in some examples moving the metal object and the sensor module may comprise moving the metal object and the system for inspecting the metal object relative to one another.

[0131] In addition, in some examples, detecting the magnetic response comprises detecting the magnetic response using the sensor module comprising a first sensor array comprising a first set of sensors and a second sensor array comprising a second set of sensors. The first sensor array may be spaced from the second sensor array along a longitudinal direction being along the direction of motion.

[0132] Moreover, the first set of sensors may be offset relative to the second set of sensors along a lateral direction being transverse to the direction of motion. An example of such an arrangement of sensor arrays is shown in Fig. 1 A and described in greater detail in relation thereto.

[0133]

[0132] Furthermore, in some examples, the system for inspecting the metal object may comprise a transportation mechanism. In such examples, moving the system and the metal object relative to one another may comprise using the transportation mechanism to propel the system relative to the metal object along the direction of motion.

[0134]

[0133] In some examples, the magnetic field that is generated may comprise one or more of a first magnetic field that is along the longitudinal direction and a second magnetic field that is along the lateral direction. Moreover, in some examples method 600 may further comprise movably adhering the system to the metal object using the magnetic field. In addition, in some examples, the sensor module may comprise a magnetic sensor.

[0135]

[0134] It is contemplated that in some examples, the sensor module may also include one or more additional sensor types comprising one or more of an ultrasound sensor, a camera, a laser scanner, an eddy current sensor, and the like. Each additional sensor type may have a corresponding additional sensor type data. In some examples, one or more of the additional sensor type data may include corresponding depth data.

[0136]

[0135] Moreover, in some examples, method 600 may further include the controller modulating at least one of the magnetic response data and the one or more additional sensor type data based on one or more of the remaining of the magnetic response data and the one or more additional sensor type data. In some examplesthis modulation may include excluding the at least one of the magnetic response data and the one or more additional sensor type data.

[0137]

[0136] In some examples, the controller may include a data acquisition module to capture the magnetic response data, and a data processing module to process the magnetic response data into the data structure. The controller may also include an inspection controller module to receive the data structure, and control one or more of the operation of the data acquisition module and the operation of the data processing module.

[0138]

[0137] In addition, in some examples, method 600 may further include the inspection controller module controlling, based on one or more of the data structure and whether the metal object has the defect, one or more of the motion of the system relative to the metal object and the operation of the sensor module. In some examples, determining whether the metal object has the defect may include the data processing module determining the position of the system relative to the metal object associated with the magnetic response data. This position may be the position of the system along the direction of motion.

[0139]

[0138] It is also templated that in some examples determining whether the metal object has the defect may include the data processing module using a machine learning model to determine based on the data structure whether the metal object has the defect. In some examples this machine learning model may include an autoencoder model. Furthermore, in some examples, the machine learning model many include a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model.

[0140]

[0139] It is contemplated that in some examples method 600 may include the features and functions described in relation to the systems and methods described in association with Figs. 1A-6, 7B, and 8. It is also contemplated that in some examples the systems and methods described in relation to Figs. 1A-6, 7B, and 8 may include the features and functions described in association with method 600.

[0140] Fig. 7B depicts a flowchart of an example method 700 for carrying out an inspection process and the like. Method 700 may be used to train a machine learning model, and to use that model to classify, or otherwise detect, defects in relevant examples of metal objects. At box 705 the mobile inspection system is placed on a metal object upon which an inspection is to be performed. In some examples the metal object may comprise a rail, a plate, a pipe, and the like.

[0141]

[0141] Moreover, the metal object may comprise a series of rails forming a track extending in the longitudinal direction, which may further comprise two parallel rails affixed to transverse members forming a support structure. In examples wherein the metal object comprises a track, the mobile inspection system may transport itself along the track, either along a single rail, or along both rails. Furthermore, in other examples wherein the metal object comprises a pipe or a plate, it is contemplated that a series of pipes or plates may be assembled in a manner that would permit the mobile inspection system to transport itself along the series of pipes or plates, such as down the interior or exterior of a pipeline.

[0142]

[0142] In addition, the mobile inspection system may be understood to be in an operating mode following box 705. Such an operating mode may involve longitudinal motion along the metal object, data being sent from a sensor module to a parallel processing module, adhesion forces being generated between the mobile inspection system and the metal object, communications being sent and received, and vehicular functions being performed. It is contemplated that in some examples the generation of magnetic adhesion forces need not be present, if moveable coupling is accomplished using a different method such as gravity or weight, moveable mechanical coupling, and the like.

[0143]

[0143] At box 710, the mobile inspection system may utilize a transportation mechanism to move along the metal object until it has traversed a region of the metal object comprising a defect. Data may be gathered and processed throughout the duration between box 705 and box 710. Moreover, the data gathered throughout this duration may be understood to pertain to “normal”, “healthy”, or “non-defect” regions of the metal object wherein no significant defects are present. Significance may bedetermined by rules defined in software by the user, by the noise floor of the sensor module, or by a trained machine learning model, and the like.

[0144]

[0144] At box 715, when a defect has been traversed, the defect may produce a magnetic response, which in turn may be detected by the sensor module, sent as data to the data acquisition device, and recorded. Such a magnetic response may comprise a variation in the induced current circulating on or in the metal object, a variation in the magnetic flux passing through the sensors, a variation in the distribution of the magnetic field due to a metallurgical or physical phenomena, and the like. Induced current circulating on or in the metal object may be referred to as “eddy current”. Variations in a relevant magnetic field or current may involve increases, decreases, changes in frequency, or characteristic spikes, shapes, and the like.

[0145]

[0145] At box 720, the data may be pre-processed. This may comprise filtering, transforming, modifying, augmenting, sorting, or otherwise altering the data to produce a dataset which is useful as an input to the remaining processes in method 700. In some examples, pre-processing may comprise applying a fast Fourier transform to generate information relating to the amplitudes and frequencies present in the data. In further examples, pre-processing may comprise applying a filter to the data to remove noise or lift-off signals. Still other examples may involve applying transformations to data in either the spatial or frequency dimensions, such as transposing data to prepare phased measurements to reflect physical locations on the metal object, or such as applying a variable gain to the frequency domain to maximize detectable characteristics.

[0146]

[0146] At box 720, the data may be fed directly into a machine learning model, such as an autoencoder or a similar unsupervised model type. Such a model may be used to perform initial detection of an anomaly or may be trained on data not containing an anomaly, such as a defect.

[0147]

[0147] At box 725, the pre-processed data may be further arranged into an at least two- dimensional array. Such an array may comprise dimensions relating to time, spatial location of sensors, frequency, magnetic field direction, and the like. In someexamples, the two-dimensional array may comprise a spectrogram, wherein one dimension corresponds to frequency, and the other dimension corresponds to time. In further examples one dimension may correspond to individual sensor locations, and the other dimension may correspond to time. Still other examples may comprise an array with three dimensions, wherein one dimension corresponds to time, another dimension corresponds to individual sensor locations, and another dimension corresponds to frequency. It is contemplated that arrays with even higher dimensions may be used to combine additional information with the magnetic field information, such as data from the transportation module, or data from environmental conditions. Additionally, it is contemplated that multiple arrays may be formed in order to represent additional data in two-dimensional arrays, instead of forming an array with more dimensions.

[0148]

[0148] At box 750, the at least two-dimensional array may be assembled with relevant metadata such as a timestamp, geospatial coordinates, environmental conditions, and visual information. Timestamps may be used to mark data in order to enable phased measurements to be assembled from measurements conducted at different times. Geospatial coordinates may be used to indicate the location of a specific defect or a specific area of a metal object. Environmental conditions may be used to correlate temperature, humidity, air quality and other factors with different types of defects, as well as specific mechanisms for degradation of metal objects. Visual information may be used to provide visual confirmation of the location in which the measurement was conducted. It is contemplated that a plurality of such types of metadata may be used to improve the context in which the magnetic responses of the metal object are detected.

[0149]

[0149] In addition, at box 750, sensor data from other sensing means may be added to the dataset to produce a multi-modal dataset. This addition will share metadata corresponding to location, temperature, humidity, air quality, and other characteristics such that the multi-modal dataset represents defects and non-defects relating to the same segment of track.

[0150] At box 755, the array may be segmented such that data related to defects and data relating to non-defect areas are separated. The segmentation process may be accomplished by an algorithm, a trained machine learning model, and the like. It is also contemplated that in some examples segmentation may be at least partially performed by an operator. In some examples, initial segmentation may be provided by an operator or an algorithm such that a high degree of confidence is present in the training dataset. In additional examples, the training dataset may be generated using known defects with known characteristics. After at least one training process has been completed, a machine learning model may be used to generate segmented datasets in addition to the initial dataset. It is contemplated that in some examples machine learning models may be continuously trained and improved as new datasets are incorporated into the training data.

[0150]

[0151] At box 755, the array may also be segmented using a plurality of signals produced by an unsupervised machine learning model, such as an autoencoder. For example, visual data from a camera module may be used to segment magnetic data, or magnetic data from the magnetic sensors may be used to segment ultrasound data.

[0151]

[0152] At box 760, a model training process may be initiated using the segmented data from box 755. This training process may utilize the data pertaining to defect areas and non-defect areas to train a machine learning model to correctly classify new data provided to it as an input. In some examples, the training process may utilize a “loss” metric or another heuristic to report on the success level of the training process.

[0152]

[0153] At box 765, the machine learning model trained by the training process of box 760 may be validated using verified datasets. These datasets may comprise data gathered from known defect and non-defect areas of a metal object, labeled data, or a subset of the dataset provided to the training process, and the like. In some examples, the labeled data may have been labeled by another machine or system, by a human operator, and the line.

[0154] At box 730, the machine learning model trained by the process of box 760 and validated by the process of box 765 may be applied to the data generated by the mobile inspection system. This may comprise running the machine learning model, streaming data from the sensing system, uploading data from the data acquisition device to the model, and the like. In some examples, the machine learning model may run in real time during the inspection process. In such examples the processes of boxes 750, 755, 760, and 765 may occur in parallel or offline, whereas the process of box 730 may directly follow the process of box 725.

[0153]

[0155] At box 735, the machine learning model may generate an output comprising an identified defect, along with relevant metadata such as a timestamp, geospatial coordinates, environmental conditions, visual information, and the like. In some examples the output may comprise the name of the defect type, in addition to a severity level. In further examples, the output may comprise a report incorporating all of the available data and metadata correlated with the identified defect.

[0154]

[0156] At box 770, the classification options for the machine learning model may be updated. The options may remain the same, such as a list of the major types and classes of defects expected in a metal object, or they may be altered, extended, or reduced to reflect the underlying data available to the mobile inspection system. Such classes of defect may be manually determined, or they may be automatically identified from the underlying data using an algorithm, heuristic, or the like.

[0155]

[0157] At box 775, the system may execute user-defined rules which may relate to the available classification options, detected defects, metadata generated by the inspection system, and the like. Such rules may comprise vehicle actions taken by the inspection system, including deceleration or acceleration, and they may further comprise actions taken by the inspection controller module, including alterations to the timing of sensor data, changes to the longitudinal resolution of the inspection process, and the like.

[0156]

[0158] At box 780, the system may send one or more notifications to the operator. Such notifications may comprise the type and location of detected defects, parameters related to the health of the mobile inspection system, and the like. In someexamples, notifications may comprise a report on a detected defect in a metal object.

[0157]

[0159] At box 790, the database of defects and normal areas may be updated to reflect new information. This information may include inputs to the training process, outputs from the machine learning model, segmented data, metadata pertaining to the type and frequency of defects, and the like. This information may be provided as an input to the process of box 750, such that the input dataset used for further training of machine learning models may contain additional information and improved context.

[0158]

[0160] Furthermore, it is contemplated that method 700 may be altered to perform some processes offline, in parallel, or asynchronously with the active inspection process itself. In some examples, the training process may be conducted externally to the mobile inspection system, using data gathered from previous inspections. In other examples, the inspection process may not need to occur in order to iterate on model training using new datasets. It should be recognized that method 700 shows a logical arrangement of processes which relate to each other by way of inputs and outputs, wherein some processes may be active in real time during an inspection, and others may be conducted offline or separately.

[0159]

[0161] As discussed previously, in some examples the inspection system may conduct a phased measurement wherein data is assembled from the sensor module measuring proximal locations on a metal object at separate points in time. In this context “proximal locations” may include locations on the metal object that are measured by the sensor module. In some examples, proximal locations may be those locations on the metal object that are the target areas of the sensors of the sensor module. In some examples, such target areas may include an area that is closest to the sensor module, an area that is within the optimal or substantially maximal sensing range of the sensor module, an area on which the sensor module is aimed or focused, and the like. Fig. 8 depicts a flowchart of an example phased measurement method 800, wherein a first portion and a second portion of a sensor module are used to measure locations on a metal object. At box 805, the mobile inspection system may traverse a defect present in a metal object and may detect amagnetic response of the defect to the magnetic field produced by the magnetic field generator. It should be recognized that the sensor module may comprise more than two portions, and that method 800 may be extended to relate to examples wherein a plurality of portions are available.

[0160]

[0162] At box 810, the first portion of the sensor module may be used to perform a measurement at a position proximal to a metal object. The first portion may comprise a first line of sensors in the sensor module, such that these sensors are the first to encounter areas of the metal object as the mobile inspection system moves along the metal object. This information may then be used to assess duration and lateral position.

[0161]

[0163] At box 815, the duration of the signal detected during the measurement process of box 810 may be assessed. In some examples, the duration may be used to calculate the longitudinal size of the detected defect by utilizing a known longitudinal velocity.

[0162]

[0164] At box 840, the lateral position of the signal detected during the measurement process of box 810 may be assessed. In some examples, the lateral position may be used to select a second portion of the sensor module to conduct a phased measurement. In other examples, the lateral position may be recorded and used to notify the operator.

[0163]

[0165] At box 820, the longitudinal position of the defect may be assessed based on available metadata, such as geospatial coordinates, an estimate of the distance travelled by the inspection system relative to a landmark, and the like.

[0164]

[0166] At box 825, the longitudinal velocity of the inspection system may be altered to support the phased measurement process. In some examples, the longitudinal velocity may be too high to perform a phased measurement on a defect with a minimal detected duration. In such cases, the longitudinal velocity may be reduced.

[0165]

[0167] At box 830, a measurement may be conducted using the second portion of the sensor module at a predetermined time corresponding to the point at which the second portion of the sensor module is located at a specific physical location relativeto the physical location at which the measurement was conducted by the first portion of the sensor module. In some examples, the measurement using the second portion of the sensor module may be conducted in real time as the mobile inspection system traverses a defect. In other examples, the measurement using the second portion of the sensor module may be conducted at a later point in time, such as on a return trajectory or direction along the metal object.

[0166]

[0168] At box 850, the data comprising a phased measurement may be assembled. In some examples, the data generated by the first portion of the sensor module and the data generated by the second portion of the sensor module may be collated such that they are spatially related and interleaved.

[0167]

[0169] At box 855, the dataset generated by the phased measurement process may be pre-processed. Pre-processing may comprise filtering, transforming, modifying, augmenting, sorting, or otherwise altering the data to produce a dataset which is useful as an input to the model training process in method 700. In some examples, pre-processing may comprise applying a fast Fourier transform to generate information relating to the amplitudes and frequencies present in the data. In further examples, pre-processing may comprise applying a filter to the data to remove noise or lift-off signals.

[0168]

[0170] At box 860, the dataset corresponding to the phased measurement process may be generated, wherein data is collated and wherein a spatial axis along the metal object has been substituted for time in the spectrogram.

[0169]

[0171] It is also contemplated that in some examples, method 700 and the other methods described herein may be stored in non-transitory computer-readable storage media as machine-readable or computer-readable instructions. Such instructions, upon execution by a processor, may cause the processor to perform method 700 and the other methods described herein.

[0170]

[0172] It should be recognized that features and aspects of the various examples provided herein may be combined into further examples that also fall within the scope of the present disclosure.

Claims

1. Claims:

1. A system for inspecting a metal object, the system comprising:a magnetic field generator to generate a magnetic field to become incident upon the metal object, the magnetic field to produce a magnetic response by the metal object;a sensor module to detect the magnetic response and generate magnetic response data, the sensor module and the metal object being moveable relative to one another along a direction of motion, the sensor module comprising a first sensor array comprising a first set of sensors and a second sensor array comprising a second set of sensors, the first sensor array spaced from the second sensor array along a longitudinal direction being along the direction of motion, the first set of sensors being offset relative to the second set of sensors along a lateral direction being transverse to the direction of motion; and a controller in communication with the sensor module, the controller to determine based on the magnetic response data whether the metal object has a defect.

2. The system of claim 1 , wherein the system and the metal object are moveable relative to one another along the direction of motion.

3. The system of any one of claims 1 to 2, further comprising a transportation mechanism to propel the system relative to the metal object along the direction of motion.

4. The system of any one of claims 1 to 3, wherein the magnetic field comprises one or more of:a first magnetic field that is along the longitudinal direction; anda second magnetic field that is along the lateral direction.

5. The system of any one of claims 1 to 4, wherein the magnetic field movably adheres the system to the metal object.

6. The system of any one of claims 1 to 5, wherein the sensor module comprises a magnetic sensor.

7. The system of claim 6, wherein the sensor module further comprises one or more additional sensor types comprising one or more of an ultrasound sensor, a camera, a laser scanner, and an eddy current sensor, each additional sensor type having a corresponding additional sensor type data.

8. The system of claim 7, wherein each additional sensor type data comprises corresponding depth data.

9. The system of claim 8, wherein the controller is further to modulate at least one of the magnetic response data and the one or more additional sensor type data based on one or more of the remaining of the magnetic response data and the one or more additional sensor type data.

10. The system of claim 9, wherein to modulate the at least one of the magnetic response data and the one or more additional sensor type data the controller is to exclude the at least one of the magnetic response data and the one or more additional sensor type data.

11. The system of any one of claims 1 to 10, wherein the controller comprises one or more of:a data acquisition module to capture the magnetic response data;a data processing module to process the magnetic response data to generate processed data; andan inspection controller module to receive the processed data, and control one or more of:operation of the data acquisition module; andoperation of the data processing module.

12. The system of claim 11 , wherein:the processed data comprises a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of the sensor module relative to the metal object; andthe controller is to determine based on the data structure whether the metal object has the defect.

13. The system of any one of claims 11 to 12, wherein the inspection controller module is to control, based on one or more of the processed data and whether the metal object has the defect, one or more of:motion of the system relative to the metal object; andoperation of the sensor module.

14. The system of any one of claims 11 to 13, wherein the data processing module comprises a machine learning model to determine based on the magnetic response data whether the metal object has the defect.

15. The system of claim 14, wherein the machine learning model comprises an autoencoder model.

16. The system of any one of claims 14 and 15, wherein the machine learning model comprises a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model.

17. The system of any one of claims 11 to 16, wherein the data processing module is to determine a position of the system relative to the metal object associated with the magnetic response data, the position being the position of the system along the direction of motion.

18. A method of inspecting a metal object for a defect, the method comprising:generating, using a magnetic field generator, a magnetic field to become incident upon the metal object;moving the metal object and a sensor module relative to one another along a direction of motion;detecting, using the sensor module, a magnetic response of the metal object to the magnetic field, to generate magnetic response data;processing the magnetic response data into a data structure having at least two dimensions, a first dimension capturing the magnetic response data and a second dimension capturing a corresponding position of the sensor module relative to the metal object; anddetermining, at a controller and based on the data structure, whether the metal object has the defect.

19. The method of claim 18, wherein:two or more of the magnetic field generator, the sensor module, and the controller are part of a system for inspecting the metal object; andthe moving the metal object and the sensor module comprises moving the metal object and the system for inspecting the metal object relative to one another.

20. The method of any one of claims 18 to 19, wherein the detecting the magnetic response comprises detecting the magnetic response using the sensor module comprising a first sensor array comprising a first set of sensors and a second sensor array comprising a second set of sensors, the first sensor array spaced from the second sensor array along a longitudinal direction being along the direction of motion, the first set of sensors being offset relative to the second set of sensors along a lateral direction being transverse to the direction of motion.

21. The method of any one of claims 19 to 20, wherein:the system for inspecting the metal object comprises a transportation mechanism; andthe moving the system and the metal object relative to one another comprises using the transportation mechanism to propel the system relative to the metal object along the direction of motion.

22. The method of any one of claims 20 to 21 , wherein the generating the magnetic field comprises generating the magnetic field comprising one or more of:a first magnetic field that is along the longitudinal direction; anda second magnetic field that is along the lateral direction.

23. The method of any one of claims 19 to 22, further comprising movably adhering the system to the metal object using the magnetic field.

24. The method of any one of claim 18 to 23, wherein the sensor module comprises a magnetic sensor.

25. The method of claim 24, wherein the sensor module further comprises one or more additional sensor types comprising one or more of an ultrasound sensor, a camera, a laser scanner, and an eddy current sensor, each additional sensor type having a corresponding additional sensor type data.

26. The method of claim 25, wherein each additional sensor type data comprises corresponding depth data.

27. The method of claim 26, further comprising the controller modulating at least one of the magnetic response data and the one or more additional sensor type data based on one or more of the remaining of the magnetic response data and the one or more additional sensor type data.

28. The method of claim 27, wherein modulating at least one of the magnetic response data and the one or more additional sensor type data comprises the controller excluding the at least one of the magnetic response data and the one or more additional sensor type data.

29. The method of any one of claims 19 to 28, wherein the controller comprises:a data acquisition module to capture the magnetic response data;a data processing module to process the magnetic response data into the data structure; andan inspection controller module to receive the data structure, and control one or more of:operation of the data acquisition module; andoperation of the data processing module.

30. The method of claim 29, further comprising:the inspection controller module controlling, based on one or more of the data structure and whether the metal object has the defect, one or more of:motion of the system relative to the metal object; andoperation of the sensor module.

31. The method of any one of claims 29 to 30, wherein the determining whether the metal object has the defect comprises the data processing module using a machine learning model to determine based on the data structure whether the metal object has the defect.

32. The method of claim 31, wherein the machine learning model comprises an autoencoder model.

33. The method of any one of claims 31 and 32, wherein the machine learning model comprises a plurality of constituent machine learning models organized in a hierarchical structure whereby an output of a first constituent machine learning model is used as an input of a second constituent machine learning model.

34. The method of any one of claims 29 to 33, the determining whether the metal object has the defect comprises the data processing module determining a position of the system relative to the metal object associated with the magnetic response data, the position being the position of the system along the direction of motion.

Images