EP4168196A1

EP4168196A1 - Detection and location of powder spreading anomalies using acoustic emission measurements

Info

Publication number: EP4168196A1
Application number: EP21736635.0A
Authority: EP
Inventors: Damien Jonathan Julien COURAPIED; Daniel André Jean CORNU; Rémi Robert GIRAUD
Original assignee: Safran SA
Current assignee: Safran SA
Priority date: 2020-06-19
Filing date: 2021-06-09
Publication date: 2023-04-26
Also published as: FR3111574A1; US20230219140A1; WO2021255367A1; FR3111574B1

Abstract

Device (100) for manufacturing a part using a method of selective fusion or selective sintering on a powder bed (103) comprising a build plate (101) having a working surface (102) parallel to a first direction (X) and to a second direction (Y), on which surface the part is intended to be manufactured, a wiper (110) which is placed on the working surface and capable of moving and spreading the powder in the first direction on the working surface, characterised in that it further comprises: - at least two acoustic sensors (121, 122) which are fixed and spaced in the second direction on the wiper and capable of detecting an acoustic signal; - a laser range finder (130) pointing in the first direction and capable of determining a position of the wiper in the first direction; and - a control system (140) capable of detecting an anomaly (150) on the basis of the acoustic signal and of determining a position of the anomaly on the basis of the acoustic signal detected by the acoustic sensors and a position signal measured by the laser range finder.

Description

Title of the invention: DETECTION AND LOCATION OF ANOMALIES OF POWDER SPREADING BY ACOUSTIC EMISSION MEASUREMENTS

Technical area

The present invention relates to the general field of the manufacture of parts by additive manufacturing by selective sintering or selective melting on a powder bed, and more particularly to the detection and localization of anomalies during the spreading of the powder.

Prior art

The manufacturing processes by selective melting or selective sintering on a powder bed comprise a step during which is deposited and spread by a scraper, on a production plate, a first layer of powder of a metal, of a metal alloy or of ceramic of controlled thickness; then a step consisting in heating a predefined zone of the powder layer. These steps are then repeated for each additional layer of powder deposited and spread.

The spreading of the powder by the scraper is an important step in the production. Indeed, the quality of the powder layer applied partly determines the quality of the part resulting from the manufacture. A bad powder spreading can, if it is not corrected, lead to a stop of the machine carrying out the manufacture and / or to the manufacture of an unconfirmed part, or even to material defects not detected in the part thus manufactured.

In order to remedy this, SA Schevchik et al. proposed in the article "Acoustic emission for in situ quality monitoring in additive manufacturing using spectral convolutional neural networks" to use the sound waves emitted during manufacturing to detect anomalies to quantify the rate of porosity in the material and thus determine the material quality after manufacture. L. Scime and J. Beuth have proposed in the article "A multi-scale convolutional neural network for autonomous anomaly detection and classification in a laser powder bed fusion additive manufacturing process" to acquire images during manufacturing to detect anomalies when spreading the powder. These images also make it possible to classify the anomalies according to their type: irregular spreading, incomplete spreading, bumps or hollows in the powder, etc.

However, these anomalies detections do not make it possible to precisely locate the anomalies during the spreading of the powder, nor to quantify the seriousness of these anomalies.

It is therefore desirable to have a system allowing precise localization of an anomaly detected during the spreading of the powder. It is also desirable to have a system for quantifying the severity of the anomaly detected.

Disclosure of the invention

The invention relates to a device for manufacturing a part by a method of selective melting or selective sintering on a powder bed comprising a production plate having a work surface, parallel to a first direction and to a second direction, on which the part is intended to be manufactured, a scraper placed on the work surface and able to move and spread the powder in the first direction on the work surface, characterized in that it further comprises:

- at least two acoustic sensors fixed and spaced in the second direction on the scraper and able to detect an acoustic signal;

a laser range finder oriented in the first direction and able to determine a position of the scraper in the first direction; and

a control system capable of detecting an anomaly as a function of said acoustic signal and of determining a position of the anomaly as a function of the acoustic signal detected by said acoustic sensors and of a position signal measured by said laser range finder. The device of the invention makes it possible to detect and precisely locate a contact between the scraper and the bed of powder during the spreading of the powder by the scraper, the contact possibly being an anomaly. In fact, during contact between the powder and the scraper, acoustic waves are emitted and propagate in the scraper. The acoustic sensors detect this acoustic signal, and the position of this contact in the second direction is determined by the time lag between the detection of the acoustic signal by the first sensor and the detection of the same acoustic signal by the second acoustic sensor. Thanks to the laser range finder, it is possible to determine the actual position of the scraper in the first direction. The control system is then able to precisely locate the contact and therefore the anomaly on the work surface. He is also able to assess whether this contact is indeed an anomaly.

The device of the invention can also detect anomalies resulting from an irregular spreading of the powder (hollow or hole in the powder bed). In fact, in this case, no acoustic wave will be detected by the acoustic sensors, which means that no contact has taken place between the scraper and the powder. It will then be up to the control system to determine if this lack of contact is normal or if it is the result of poor spreading of the powder.

The anomaly is more particularly localized in a system with three coordinates in the first, second and third directions. Thus the first coordinate which corresponds to the first direction is determined by virtue of the position signal measured by the laser range finder. The second coordinate which corresponds to the second direction is determined by the time lag between the detections of the acoustic signal from the two acoustic sensors. The third coordinate which corresponds to a third direction is determined by the number of layers of powder spread by the scraper.

According to another particular characteristic of the invention, the laser range finder is fixed to the scraper.

According to another particular characteristic of the invention, the laser range finder is fixed to the scraper by gluing or by a mechanical connection. According to another particular characteristic of the invention, each acoustic sensor is fixed to the scraper by gluing or by mechanical connection.

According to another particular characteristic of the invention, the manufacturing device comprises two acoustic sensors each fixed to one end of the scraper.

According to another particular characteristic of the invention, the control system is a machine learning system.

The machine learning system is then able to give the precise localization of the anomalies and thanks to the acoustic signals detected by the acoustic sensors, it is able to learn to classify the anomalies according to their severity according to the characteristics of these signals. These characteristics are for example the amplitude, the duration or the waveform of the signal.

Thanks to the classification of the anomalies by the machine learning system, it is also possible to know whether the anomaly results in damage to the scraper or not, and whether it is necessary to stop the manufacture of the part to repair or change the scraper. .

According to another particular characteristic of the invention, the machine learning system is a neural network, and more particularly is a convolutional neural network.

According to another particular characteristic of the invention, the machine learning system is a system of decision trees or a forest of decision trees.

According to another particular characteristic of the invention, the machine learning system is based on probabilistic modeling.

According to another particular characteristic of the invention, the machine learning system is based on methods using kernels.

According to another particular characteristic of the invention, the machine learning system is based on a gradient amplification algorithm.

Neural networks have the advantage of allowing supervised learning. They will be able to do all the work of extracting the data and description of anomalies. During the training phase, the classification error will be minimized in order to optimize the classification parameters. In addition, the specific architecture of the neural network makes it possible to extract signatures of the anomaly of different complexities, from the simplest to the most complex. In addition, the automatic extraction and prioritization of signatures, which adapt to the given problem, is one of the strengths of convolutional neural networks.

Another object of the invention is a method for detecting an anomaly implemented by a device according to the invention comprising the following steps:

- detection of an acoustic signal by the at least two acoustic sensors;

- determination of the position of the anomaly, the position of the anomaly being determined in the first direction by the position signal measured by the laser range finder, in the second direction by a time shift of the acoustic signal detected by the at least two acoustic sensors and in a third direction by a number of layers of powder spread by the scraper; and

- classification of the anomaly according to its severity.

Determining the position of the anomaly in three directions makes it possible to locate the anomaly in a three-dimensional space, and not only in a two-dimensional space as in the prior art. This may possibly be able to correct the anomaly quickly. In addition, the classification of the anomaly according to its seriousness also makes it possible to know whether a production stop is necessary or not, if the scraper is damaged or even if the final part will contain acceptable material defects or not.

Thanks to the detection of anomalies during the spreading of the powder by the scraper, it also becomes possible to detect hot spots linked to the elevation by increasing the thickness of the powder layer, detachment of the part or supports. compared to the build plate or even cracks in the part caused by the lack of powder spreading. Another subject of the invention is a method of manufacturing a part by selective melting or selective sintering on a powder bed comprising a step of detecting anomalies by the detection method according to the invention.

Brief description of the drawings

Other characteristics and advantages of the present invention will emerge from the description given below, with reference to the appended drawings which illustrate exemplary embodiments thereof without any limiting nature.

[Fig. 1] FIG. 1 represents a device for manufacturing a part by selective melting or selective sintering on a powder bed according to one embodiment of the invention.

[Fig. 2] FIG. 2 represents a scraper of the manufacturing device according to one embodiment of the invention.

Description of the embodiments

FIG. 1 represents a device 100 for manufacturing a part by selective melting or selective sintering on a powder bed comprising a production plate 101 having a work surface 102 on which the part is manufactured. The device 100 also comprises a scraper 110 which makes it possible to spread the powder 103 in the X direction by moving on the work surface 102. The scraper 110 thus moves in a plane formed by the axes (X, Y) and the build plate 101 as well as the work surface 102 are parallel to this plane. The number of layers of powder deposited and spread defines the thickness of the final part along the Z axis.

Following the spreading of the powder by the scraper 110, areas 160 of the working surface 102 are heated by an energy source, such as a laser beam or an electron beam, so as to sinter or merge powder in these areas 160.

In order to detect anomalies 150 resulting from contact or from an absence of contact between the powder and the scraper 110, the device 100 also comprises a laser range finder 130 and two acoustic sensors 121 and 122. The laser range finder 130 is fixed to one end 170 of the device 100 and is oriented towards the scraper 110 in the X direction. The range finder 130 makes it possible to measure the position of the scraper 110 in the X direction, which makes it possible to determine the position of a possible anomaly 150 in the same direction X.

The two acoustic sensors 121 and 122 are fixed on the scraper 110. They are spaced from each other in the Y direction. When contact takes place between the scraper 110 and the powder 103 spread on the work surface 102, acoustic waves are emitted. and propagate in the scraper 110. By propagating in the scraper, these acoustic waves are detected by the two acoustic sensors 121 and 122. Thanks to the data transmitted by the acoustic sensors 121 and 122, a control system 140 identifies or not a anomaly. These same data also make it possible to determine the position of the anomaly 150 in the direction Y. Indeed, thanks to the time shift between the detections of the acoustic wave by the two sensors 121 and 122, it is possible to go back to the position of the contact. and therefore the anomaly 150 in the Y direction.

The control system 140 determines the position of the anomaly 150 in the X direction, which is the position of the scraper 110 in this same X direction measured by the laser range finder 130.

The determination of the position of the anomaly 150 in the Z direction by the control system 140 is given by the number of layers of powder 103 deposited and spread on the production plate 101.

The laser range finder 130 and the acoustic sensors 121 and 122 can be attached to the device 100 by gluing or by a mechanical connection.

According to a particular characteristic of the invention, the acoustic sensors 121 and 122 are placed respectively at 1/3 and 2/3 of the length of the scraper 110 in the direction Y.

If no wave is detected by the two acoustic sensors 121 and 122, this means either that there is no more powder to spread or that the scraper 110 has reached the end of its travel, or that there is a hole in the powder bed 103 which may be an anomaly. It will be the control system 140 which will determine if the absence of signals acoustics is or is not an anomaly. Likewise, the position of this hole in the powder 103 can be determined by virtue of the time shift between the two acoustic sensors 121 and 122.

The control system 140 can be a machine learning system. The system 140 can thus learn to classify the acoustic waveforms detected by the acoustic sensors 121 and 122 contacts, to determine whether the contact, or the absence of contact is an anomaly 150 and then classify this anomaly 150 according to its gravity. Indeed, an anomaly 150 may be the result of a gas bubble, a hole in the spread powder, an elevation of the powder, an incomplete spread of the powder, traces in the spreading or even irregular spread. These anomalies do not present the same level of severity. Usually, the most serious anomalies result from powder elevation, traces in spreading, and uneven spreading in the (X, Y) plane of the work surface 102. Classification according to the severity of the anomalies allows the control system 140 to give or predict the risk of damage to the scraper, to the final part and to decide whether or not production should be interrupted.

The classification of the detected anomalies 150 can be done according to the characteristics of the detected waves: waveform, amplitude, duration, etc ...

According to one particular characteristic of the invention, the machine learning system is chosen from among a convolutional neural network, a system based on probabilistic modeling, a system based on a kernel algorithm, decision trees, a forest of 'decision trees or a gradient implementation system. A convolutional neural network will, for example, make it possible to have an automatic extraction and prioritization of the contacts and non-contacts detected in order to classify the anomalies according to their severity.

FIG. 2 represents a scraper 210, and more particularly the arrangement of the acoustic sensors 221 and 222 and of the laser range finder 230 according to another embodiment of the invention. The two acoustic sensors 221 and 222 are still fixed on the scraper 210 and spaced apart in the Y direction. So as to obtain a significant time shift to determine the position of the anomaly 250 according to the direction Y, the acoustic sensors 221 and 222 are each placed on one end of the scraper 210.

The laser range finder 230 is fixed on the scraper 210 and is always oriented in the direction X to determine the position of the scraper 210 and of the anomaly 250 in this same direction X.

According to a particular characteristic of the invention, the laser range finder 230 is placed in the middle of the scraper 210 between the two acoustic sensors 221 and 222.

As before, the two acoustic sensors 221 and 222, as well as the laser range finder 230 can be attached to the scraper 210 by gluing or by a mechanical connection.

In the event of contact between the scraper 210 and the powder at the instant to, acoustic waves propagate in the scraper 210. The acoustic sensor 221 detects these waves at the instant t ₀ + a, while the acoustic sensor 222 detects these same waves at the instant to + a + b. It is the temporal shift between the two instants of detection b which makes it possible to determine the position of the contact (and of the possible anomaly 250) in the direction Y, the positions of the two acoustic sensors 221 and 222 being known.

As indicated above, the stopping of the propagation of acoustic waves in the scraper can also be detected and the possible anomaly at the origin of this stopping will also be localized thanks to the time lag between the acoustic sensors.

In order to achieve a more precise localization of the anomaly in the Y direction, the manufacturing device can also include more than three acoustic sensors. Indeed, if the acoustic sensors are too far apart, one of the sensors could not perceive the acoustic wave propagating in the scraper, or could perceive a wave already disturbed by the arrival of a new wave. The use of three or four sensors makes it possible to remedy this, because it is thus possible to place the sensors at closer intervals, while maintaining a sufficient spacing between at least two sensors to detect a significant time shift for the determination of the position. of the anomaly. According to a particular characteristic of the invention, the device comprises three acoustic sensors placed respectively at 1/4, 2/4 and 3/4 of the length of the scraper in the Y direction.

Another object of the invention is a method for detecting an anomaly during the spreading of the powder by the scraper, the method being implemented by one of the devices presented above.

The method firstly comprises the detection of an acoustic signal by the acoustic sensors, then the determination of the position of the anomaly, and finally the classification of the anomaly according to its severity.

The position of the anomaly is determined:

- in the first direction X by the position signal measured by the laser range finder,

- in the second direction Y by a time shift of the acoustic signal detected by the acoustic sensors, and

- in the third direction Z by the number of layers of powder spread by the scraper.

It is the control system which determines whether the detected acoustic wave or the absence of a wave really results from an anomaly, then which determines the position of this anomaly and finally which classifies this anomaly according to its severity.

Another subject of the invention is a method of manufacturing a part by selective melting or selective sintering on a powder bed comprising a step of detecting anomalies by the detection method described above.

The detection of an anomaly, the determination of its position in space and its classification makes it possible to know if this anomaly must be corrected, and if so, we can do it quickly by knowing its precise position; or if the production must be stopped. This also makes it possible to predict any defects in the final part.

Claims

[Claim 1] Device (100) for manufacturing a part by a selective melting process or selective sintering on a powder bed (103) comprising a build plate (101) having a work surface (102), parallel to a first direction (X) and to a second direction (Y), on which the part is intended to be manufactured, a scraper (110, 210) placed on the work surface and able to move and spread the powder (103) in the first direction on the work surface, characterized in that it further comprises:

- at least two acoustic sensors (121, 122, 221, 222) fixed and spaced in the second direction (Y) on the scraper and able to detect an acoustic signal;

- a laser range finder (130, 230) oriented in the first direction (X) and able to determine a position of the scraper in the first direction (X); and

- a control system (140) capable of detecting an anomaly (150, 250) as a function of said acoustic signal and of determining a position of the anomaly as a function of the acoustic signal detected by said acoustic sensors and of a measured position signal by said laser range finder.

[Claim 2] A device according to claim 1, wherein the laser range finder is attached to the scraper.

[Claim 3] Device according to claim 1 or 2, in which each acoustic sensor is fixed to the scraper by gluing or by a mechanical connection.

[Claim 4] Apparatus according to any one of claims 1 to 3, wherein the control system is a machine learning system.

[Claim 5] A device according to claim 4, wherein the machine learning system is a neural network.

[Claim 6] A device according to claim 5, wherein the neural network is a convolutional neural network.

[Claim 7] Apparatus according to claim 4, wherein the machine learning system is a decision tree system or a forest of decision trees.

[Claim 8] A method of detecting an anomaly (150, 250) implemented by a device (100) according to any one of claims 1 to 7, comprising the following steps:

- detection of an acoustic signal by the at least two acoustic sensors

(121, 122, 221, 222);

- determination of the position of the anomaly, the position of the anomaly being determined in the first direction (X) by the position signal measured by the laser range finder, in the second direction (Y) by a time shift of the acoustic signal detected by the at least two acoustic sensors and in a third direction (Z) by a number of layers of powder spread by the scraper; and - classification of the anomaly according to its severity.

[Claim 9] A method of manufacturing a part by selective melting or selective sintering on a powder bed comprising a step of detecting anomalies by the detection method according to claim 8.