Method for learning to recognise a compliant weld bead

Abstract

The invention relates to a method for learning to recognise a compliant weld deposited along a welding path, the method being implemented by means of a welding facility comprising: - a welding head that is movable along the welding path and capable of depositing a weld bead on the welding path; - means for acquiring at least one item of data that influences at least one characteristic of the weld bead; - at least one means for measuring the position of the welding head; - a computer unit in which at least one computer program is implemented, which computer unit is able to receive and store the at least one item of data acquired and associated with the measured position of the welding head.

Description

Learning process for recognizing a conforming weld bead Technical field of the invention

[0001] The invention relates to the field of welding. In particular, the invention relates to the field of recognizing a conformal weld deposited along a welding path. Technical background

[0002] Tube welding, buttering or additive manufacturing are operations that can be carried out automatically using a welding head.

[0003] A welding path may, for example, be formed at a junction between two joined tubes or directly at a specific surface, such as at least part of one edge of a part for the purpose of buttering. This welding path, circular or flat, is filled with molten metal.

[0004] In the case where two tubes are to be welded together, this filling is carried out in several passes, each pass corresponding to a 360° rotation around the components, for example metallic, of the welding electrode.

[0005] Each pass deposits molten metal corresponding to at least one weld bead. Each pass deposits the weld beads and superimposes them on top of each other.

[0006] When all passes have been made and the weld path is filled, the weld is complete.

[0007] Welds intended for certain sectors, such as nuclear power or hydrocarbon transport, must comply with strict specifications.

[0008] As things stand, the weld is analyzed by non-destructive or destructive testing to detect possible defects once the weld is completed, i.e. when all the passes have been made and the weld path has been filled.

[0009] When a defect is detected, that is to say when at least one bead constituting the weld is considered non-compliant, it is then currently necessary to remove it and redo it. This corrective operation may require grinding the weld to a thickness corresponding to the location of the defect, then making several passes to refill the welding path.

[0010] These correction operations are tedious and burden production.

[0011] There is therefore a need to optimize these correction operations and to be able to establish the conformity of each weld bead deposited after each pass before the weld is completed.

[0012] The invention thus aims to ensure the conformity of the weld bead deposited after each pass.

[0013] To this end, there is proposed firstly a method for learning to recognize a conforming weld deposited along a welding path, the method being implemented by means of a welding installation comprising:– a welding head movable along the welding path and capable of depositing a weld bead in the welding path,– means for acquiring at least one piece of data influencing at least one characteristic of the weld bead,– at least one means for measuring the position of the welding head,– a computer unit in which at least one computer program is implemented, capable of receiving and storing the at least one piece of data acquired and associated with the measured position of the welding head,said method comprising:– an operation of moving the welding head along the welding path,– an operation of depositing a weld bead in the welding path,– an operation of acquiring at least one piece of data influencing at least one characteristic of the weld bead, this acquisition operation being carried out simultaneously with the operation of depositing a weld bead,– an operation of measuring the position of the welding head, this measurement operation being carried out simultaneously with the operation of acquiring the at least one piece of data,– an operation of storing the acquired data and associated with the measured positions of the welding head, in the computer unit,said previous steps being repeated so as to deposit several weld beads, all of the weld beads deposited corresponding to a weld,the method further comprising:– an operation of controlling the weld,

[0014] – an operation of evaluating the controlled weld so as to establish a conforming weld, when the conforming weld has been established, the method further comprises:

[0015] – an operation of inserting, into the computer unit, the data acquired and associated with the measured positions of the welding head representative of each weld bead deposited,

[0016] the learning method being characterized in that the preceding steps are repeated to deposit several compliant welds and to constitute a set of data representative of the compliant welds, and in that it further comprises an operation of calibrating the at least one computer program with the set of data representative of the compliant welds inserted into the computer unit to obtain a calibrated computer program, the calibrated computer program being configured to recognize a compliant weld.

[0017] The learning method according to the invention is divided into a learning phase and a monitoring phase. In the learning phase, the method refines a predictive model based on operational data obtained as described later for a so-called compliant weld (hereinafter referred to as a subset of operational data). In the monitoring phase, the learning method calculates an approval score based on a deviation or gap of the operational data from the subset of operational data observed during monitoring compared to a prediction result obtained by the predictive model. Then, from this learning score, it is possible to inform, for example, the operator performing the welding of the status of the current weld, and to display associated information.

[0018] Preferably, the calibration operation, corresponding to the learning phase of the method, consists of a calibration operation of the predictive model aimed at predicting the conformity of a weld based on the subset of operational data associated with the corresponding conforming weld. The calibration operation is preferably implemented by a computer.

[0019] The predictive model is advantageously configured to determine an approval score representative of a compliant weld. The predictive model is advantageously of the anomaly detection type.

[0020] Among said “operational data set”, each conforming weld is associated with a subset of data representative of the corresponding conforming weld.

[0021] Each subset of operational data of a weld is associated with at least two distinct operational data each influencing at least one characteristic of a weld bead of the corresponding conforming weld.

[0022] Advantageously, each of said two operational data is associated with a weighting index, and at least the position of the welding electrode corresponding to the data, thus forming structured operational data.

[0023] It will be understood that each structured data item comprises an association of at least the corresponding data item, a weighting index associated with said data item and the position of the welding electrode corresponding to said data item. For example, in the case of said two operational data items, these may form two structured operational data items, namely a first structured operational data item and a second structured operational data item, the first structured operational data item associating the first operational data item, a weighting index corresponding to the first operational data item and the position of the welding electrode corresponding to the first operational data item, and the second structured operational data item associating the second operational data item, a weighting index corresponding to the second operational data item and the position of the welding electrode corresponding to the second operational data item.

[0024] Advantageously, the weighting index of each of said two operational data may be dependent on a variation in the value of the corresponding data.

[0025] Advantageously, the weighting index of each of said two operational data can be adjusted according to the approval score of the predictive model.

[0026] Regardless of the approval score, the learning method adjusts the weighting indices of each of said two operational data, and determines a second approval score. This step of adjusting the weighting indices is repeated for a number of cycles N.

[0027] Furthermore, for each approval score is associated with a prediction error representative of a deviation of at least one of said two operational data, the prediction error is obtained from the prediction result of the prediction model.

[0028] The prediction error of a cycle is advantageously used to adjust the weighting indices of each of said two operational data for the following cycle.

[0029] Iterating over several cycles advantageously allows for refining the learning of the predictive model.

[0030] It will be understood that the predictive model is trained from said structured operational data corresponding to a compliant weld.

[0031] Once the predictive model has been trained, the method comprises a phase of monitoring a weld in progress, the monitoring phase comprising a step of acquiring in real time said subset of operational data of said weld in progress.

[0032] Preferably, the monitoring phase comprises determining in real time, from the trained predictive model, the approval score of the current weld based on said subset of operational data of said current weld.

[0033] An approval score obtained within a predetermined value range ensures that the current weld is compliant, while an approval score outside this value range identifies a deviation or gap in at least one piece of data from said subset of operational data of said current weld.

[0034] When such a deviation or gap is identified, the monitoring phase includes a step of displaying information representative of the deviation or gap. The operator supervising the current weld can then carry out an operation to check and correct the current weld.

[0035] The learning process thus guarantees the successful conformity of the current weld. It follows that no subsequent control operation is necessary. In other words, it will be understood that the weld does not require analysis by non-destructive or destructive control to detect possible defects once the weld is completed, unlike known methods.

[0036] A conforming weld is a weld that meets predetermined criteria. For example, the predetermined criteria may be determined by specifications or by one or more specific standards. Predetermined criteria include the absence of certain defects such as spheroidal blowholes, blowhole nests, aligned blowholes, metallic inclusions, connection defects, fusion gaps, bonding gaps, and weld bead thicknesses that are too thin or too thick compared to a reference value.

[0037] Conforming welds means a plurality of welds each of which satisfies predetermined criteria distinct from another weld in that plurality of welds.

[0038] A welding head means a fusible electrode or a refractory electrode used with or without welding wire. For example, the welding head may be connected to a welding generator which is typically a power supply generator.

[0039] It should be noted that the at least one piece of data influencing at least one characteristic of the weld bead may take the form of a value or several values ​​which may be intrinsic to the elements used for the removal of the weld bead and / or be representative of one or more measured, programmed and / or calculated parameters, and / or linked to the environment.

[0040] For example, characteristics of the weld bead include its thickness, surface condition, volume, position in the welding path, mechanical properties and chemical composition.

[0041] It should be noted that the learning method does not include a step of inserting, into the computer unit, data representative of non-compliant welds, for example welds comprising a weld bead comprising a defect.

[0042] Various additional characteristics may be provided alone or in combination:– the at least one data item is the voltage and / or the current delivered by a welding generator connected to the welding head;– the at least one data item is the speed of movement of the welding head;– the at least one data item is the ambient temperature;– the at least one data item is the ambient humidity;– the at least one data item is the wind speed;– the at least one data item is the temperature of the weld bead;– the at least one data item is the volume or mass flow rate of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air;– the at least one data item is the pressure of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air;– the at least one data item is the temperature of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air;– at least one data item is the profile of the weld bead and / or the profile of the adjacent areas.;

[0043] Secondly, a method for automated recognition of a conforming weld bead deposited or in the process of being deposited along a welding path is proposed, this method using the lessons of a prior learning method as previously described, the automated recognition method being implemented by means of a welding installation comprising:– a welding head movable along the welding path and capable of depositing a weld bead in the welding path,– means for acquiring at least one piece of data influencing at least one characteristic of the weld bead,– at least one means for measuring the position of the welding head,– a computer unit in which at least one computer program is implemented and in which the acquired data and associated with the measured positions of the welding head are inserted,the computer program being calibrated by means of said learning method to ensure the conformity of the weld bead being deposited,said automated recognition method comprising:– an operation of moving the welding head along the welding path,– an operation of depositing a weld bead in the welding path,– an operation of acquiring at least one piece of data influencing at least one characteristic of the weld bead, this acquisition operation being carried out simultaneously with the operation of depositing a weld bead,– an operation of measuring the position of the welding head, this measurement operation being carried out simultaneously with the operation of acquiring the at least one piece of data,– an operation of detecting a deviation of the at least one piece of data acquired and associated with its measured position with respect to the data representative of the conforming welds defined according to the calibrated computer program,when said deviation is detected, the automated recognition method further comprises an alert operation for the attention of an operator.,

[0044] It should be noted that the above automated learning and recognition methods can be implemented in the context of depositing a weld between at least two components, for example metallic, defining the welding path or even in the context of depositing a weld directly on a specific surface and in a predetermined direction defining the welding path, or for example during buttering or during the implementation of additive manufacturing.

[0045] By detection of a deviation, we mean the detection of data, considered alone or in combination with others, acquired and associated with its measured position which would be sufficiently far from the representative data of the welds qualified as compliant according to the definition of conformity defined by the calibrated computer program. This data is then sufficiently far to be considered as being outside the scope of coverage of the representative data of the compliant welds defined according to the calibrated computer program. Detailed description of the invention

[0046] In the following, a method for learning to recognize a conforming weld bead according to an exemplary embodiment of the invention will be described.

[0047] This learning process is described in the context of welding two metal tubes. However, it should be noted that this learning process may be suitable for welding between two metal components of different shapes and / or nature.

[0048] This process uses a welding installation.

[0049] The welding installation comprises a welding head and a device for feeding the metal to be welded. For example, the head may be a consumable electrode connected to a welding generator, which is typically a power supply generator, and which creates an electric arc intended to melt the metal to be welded.

[0050] The installation comprises a moving device, on which the welding head is mounted. The moving device is capable of allowing the welding head to be moved.

[0051] The installation includes means for measuring the position of the welding head. The position of the welding head is understood to mean its coordinates in a given reference frame. The reference frame is materialized by a mark on one of the two tubes to be welded.

[0052] The welding system allows two tubes to be welded together. First, two tubes are joined together. The junction between these two tubes forms a circular welding path. To fix these tubes together, molten metal is deposited in the welding path.

[0053] The installation is therefore able to deposit several weld beads in the welding path.

[0054] The installation includes means of acquiring data influencing a characteristic of the weld bead.

[0055] The installation includes a means of measuring the position of the welding head.

[0056] The installation comprises a computer unit. At least one computer program is implemented in the computer unit. The computer unit is capable of receiving and storing the data acquired by the acquisition means and associated with the positions measured by a measuring means.

[0057] More precisely, the computer unit stores in real time the data acquired and associated with the positions of the welding head.

[0058] Data influencing a characteristic of the weld bead is systematically associated with the corresponding position of the welding head.

[0059] The learning phase involves starting a first welding operation between two tubes. The first welding operation involves several passes to deposit several weld beads in order to fill the welding path with molten metal. Thus, the first welding operation involves making a weld, called a complete weld, between two metal tubes.

[0060] Thus, the learning method includes an operation of moving the welding head along the welding path. This operation is carried out by means of the moving device.

[0061] The method comprises an operation of depositing a weld bead in the welding path. At least one weld bead is deposited per pass.

[0062] The operation of depositing a weld bead is carried out as the head moves along the welding path.

[0063] The method comprises an operation of measuring several data influencing at least one characteristic of the weld bead. The method simultaneously comprises a measurement of the position of the welding head in the reference frame.

[0064] The method comprises an operation of storing the data acquired and associated with the measured positions of the welding head, this storing being carried out in the computer unit.

[0065] In other words, several data influencing at least one characteristic of the weld bead are measured and associated with the corresponding position of the welding electrode. Finally, these data are stored in the computer unit.

[0066] The previous steps are repeated so as to deposit several weld beads, all of the weld beads deposited corresponding to one weld.

[0067] When the welding of the tubes is completed, that is, when the weld is finished or the weld path is filled with weld beads after several passes, a destructive or non-destructive control operation of the weld is carried out. Then, an evaluation operation of the controlled weld is carried out in order to establish a compliant weld. This operation can be carried out manually. An X-ray, for example, can be carried out to obtain an image of the depth of the weld. Other techniques can be used.

[0068] The X-rays are analyzed by an operator who visually detects welding defects.

[0069] When the operator does not detect any welding defects, the weld inspected and evaluated is considered to be a compliant weld. In this case, the operator enters into the computer unit the data acquired and associated with the measured positions of the welding head representative of each weld bead deposited.

[0070] The computer unit then associates with each weld bead contained in the conforming weld the data acquired and associated with the measured positions of the welding head.

[0071] The previous steps are repeated to deposit several conforming welds and to constitute a representative data set of the conforming welds.

[0072] Then, a calibration operation of the at least one computer program with the set of data representative of the compliant welds inserted into the computer unit is carried out. This calibration operation aims to obtain a calibrated computer program, the calibrated computer program then being configured to recognize a compliant weld.

[0073] In the following, the data influencing at least one characteristic of the weld bead will be described in detail.

[0074] Advantageously, data influencing at least one characteristic of the weld bead are the voltage and current delivered by the welding generator. The current is measured at the terminals of the generator. The voltage is measured on the displacement device.

[0075] The current and voltage determine the quality of the weld. More precisely, the weld pool is likely to fail to melt the surrounding metal or, on the contrary, to collapse.

[0076] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the speed of movement of the welding electrode. This speed is measured by a sensor which continuously measures the speed of movement of the welding electrode.

[0077] Travel speed is an important factor in weld bead removal. High travel speeds cause significant mechanical stresses that can cause cracks to appear in the weld bead.

[0078] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the ambient temperature. This temperature is provided by a sensor capable of measuring the ambient temperature.

[0079] Ambient temperature affects weld quality. When the temperature is below a certain temperature, welding defects may occur.

[0080] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the ambient humidity. The ambient humidity is provided by a sensor capable of measuring it.

[0081] Ambient humidity affects weld quality. Humidity measurement, along with temperature measurement, allows the dew point to be determined. The dew point can affect weld quality.

[0082] Advantageously, one data influencing at least one characteristic of the welding bead is the wind speed at the welding electrode. This speed is measured by means of a speed sensor.

[0083] Wind speed near the welding electrode can impact weld quality. It can cause porosity, oxidation, and blowholes in the weld.

[0084] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the temperature of the weld bead. This temperature is measured by means of a temperature sensor.

[0085] The temperature of the weld bead has an impact on the quality of the weld.

[0086] Advantageously, a piece of data influencing at least one characteristic of the weld bead is the volume or mass flow rate of a shielding gas intended to prevent oxidation of the molten metal by the oxygen in the air.

[0087] The flow rate of the shielding gas has an impact on the quality of the weld. In particular, it is the cause of porosity, oxidation and blowholes in the weld.

[0088] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the pressure of the shielding gas.

[0089] The pressure of the shielding gas has an impact on the quality of the weld.

[0090] Advantageously, one piece of data influencing at least one characteristic of the weld bead is the temperature of the shielding gas.

[0091] The temperature of the shielding gas has an impact on the quality of the weld.

[0092] Advantageously, data influencing at least one characteristic of the weld bead is the profile of the weld bead and / or the profile of the adjacent areas. Adjacent areas are understood to mean areas located near the weld bead. These profiles are, for example, captured using a profilometer.

[0093] The profile provides useful information about the quality of the weld.

[0094] Other data influencing at least one characteristic of the weld bead could be recorded by means of any known element configured to capture at least one video stream or one sound recording.

[0095] Once the learning phase is completed, it becomes possible to recognize weld beads that conform to an early stage of the welding operation. Indeed, the computer unit having been enriched with the lessons provided by the learning process, and the computer program having been calibrated, it is possible to implement an automated process for recognizing a conforming weld bead.

[0096] Thus, the invention relates to a method for automated recognition of a conforming weld bead. This method comprises an operation of moving the welding electrode along the welding path.

[0097] The method comprises simultaneously an operation of depositing a weld bead in the welding path. The method comprises an operation of measuring at least one data item influencing at least one characteristic of the weld bead and simultaneously an operation of measuring the position of the welding electrode in the reference frame.

[0098] The method comprises an operation of detecting a deviation of the at least one acquired data item and associated with its measured position with respect to the data representative of the compliant welds defined by the calibrated computer program. The detection operation is therefore notably carried out using the calibrated computer program. When the deviation of the at least one acquired data item and associated with its measured position with respect to the data representative of the compliant welds is detected, the automated recognition method further comprises an operation of alerting an operator.

[0099] Thus the operator can suspend the welding operation at an early stage, i.e. before the welding operation is completed and thus remove the thin thickness of weld characteristic at least in part of the non-compliant, or possibly non-compliant, weld bead, for which a deviation of the at least one acquired data item and associated with its measured position with respect to the data representative of the compliant welds has been detected.

Claims

Method for learning to recognize a conforming weld deposited along a welding path, the method being implemented by means of a welding installation comprising:– a welding head movable along the welding path and capable of depositing a weld bead in the welding path,– means for acquiring at least one piece of data influencing at least one characteristic of the weld bead,– at least one means for measuring the position of the welding head,– a computer unit in which at least one computer program is implemented, capable of receiving and storing the at least one piece of data acquired and associated with the measured position of the welding head,said method comprising:– an operation of moving the welding head along the welding path,– an operation of depositing a weld bead in the welding path,– an operation of acquiring at least one piece of data influencing at least one characteristic of the weld bead,this acquisition operation being carried out simultaneously with the operation of depositing a weld bead,– an operation of measuring the position of the welding head, this measurement operation being carried out simultaneously with the operation of acquiring the at least one piece of data,– an operation of storing the acquired data and associated with the measured positions of the welding head, in the computer unit,said previous steps being repeated so as to deposit several weld beads, all of the deposited weld beads corresponding to a weld,the method further comprising:– an operation of checking the weld,– an operation of evaluating the checked weld so as to establish a compliant weld,when the compliant weld has been established, the method further comprises:– an operation of inserting, in the computer unit,data acquired and associated with the measured positions of the welding head representative of each deposited weld bead, the learning method being characterized in that the preceding steps are repeated to deposit several compliant welds and to constitute a set of data representative of the compliant welds, and– in that it further comprises an operation of calibrating the at least one computer program with the set of data representative of the compliant welds inserted into the computer unit to obtain a calibrated computer program, the calibrated computer program being configured to recognize a compliant weld., Method according to the preceding claim in which the at least one data item is the voltage and / or the intensity delivered by a welding generator connected to the welding electrode. Method according to any one of the preceding claims in which the at least one data is the speed of movement of the welding electrode. Method according to any one of the preceding claims in which the at least one data item is the ambient temperature. Method according to any one of the preceding claims in which the at least one data item is the ambient humidity. Method according to any one of the preceding claims in which the at least one data item is the wind speed. Method according to any one of the preceding claims in which the at least one data item is the temperature of the weld bead. Method according to any one of the preceding claims in which the at least one data is the volume or mass flow rate of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air. Method according to any one of the preceding claims in which the at least one data is the pressure of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air. Method according to any one of the preceding claims in which the at least one data is the temperature of a shielding gas intended to prevent oxidation of the molten metal by oxygen in the air. Method according to any one of the preceding claims in which the at least one data item is the profile of the cord and / or the profile of the adjacent zones. Automated method for recognizing a conforming weld bead deposited or being deposited along a welding path, this method using the teachings of a prior learning method according to any one of the preceding claims, the automated recognition method being implemented by means of a welding installation comprising:– a welding head movable along the welding path and capable of depositing a weld bead in the welding path,– means for acquiring at least one piece of data influencing at least one characteristic of the weld bead,– at least one means for measuring the position of the welding head,– a computer unit in which at least one computer program is implemented and in which the acquired data and associated with the measured positions of the welding head are inserted,the computer program being calibrated by means of said learning method to ensure the conformity of the weld bead being deposited,said automated recognition method comprising:– an operation of moving the welding head along the welding path,– an operation of depositing a weld bead in the welding path,– an operation of acquiring at least one piece of data influencing at least one characteristic of the weld bead, this acquisition operation being carried out simultaneously with the operation of depositing a weld bead,– an operation of measuring the position of the welding head, this measurement operation being carried out simultaneously with the operation of acquiring the at least one piece of data,– an operation of detecting a deviation of the at least one piece of data acquired and associated with its measured position with respect to the data representative of the conforming welds defined by the calibrated computer program,when said deviation is detected, the automated recognition method further comprises an alert operation for the attention of an operator.,

Images