Automatic drilling optimization method

Abstract

A method for optimizing the execution of an automatic drilling system controlled by a numerical NC machine, wherein the NC machine performs the following steps: - identifying at a time instant t each drilling performed by an automatic drilling tool; - storing the theoretical position of each drilling performed at the time instant t: - calculating a learned position using a machine learning model and based on the theoretical position stored at the time instant t; - estimating an intermediate position by applying a trend statistical function, if the difference between the intermediate position and the learned position is below a preconfigured threshold, using the intermediate position to position the drilling tool to the next drilling and performing drilling at the next time instant t+1.

Description

Technical Field

[0001] This invention relates to the field of automated drilling and countersunk hole drilling processes specifically for the aircraft manufacturing and assembly industry.

[0002] More specifically, the present invention relates to a method for optimizing the execution of an automatic drilling and / or countersunk hole drilling tool controlled by an automatic control system. Background Technology

[0003] Numerical control (NC) (also known as computer numerical control, CNC) is the automatic control of machining tools (such as drilling machines, countersinking machines, boring machines, lathes, etc.) by a computer. NC machines can process materials (metals, plastics, wood, ceramics, or composite materials) to meet specifications by following coded programming instructions and without human operation.

[0004] NC-like systems are used in any process that can be described as motion and manipulation. These processes include laser cutting, welding, friction stir welding, ultrasonic welding, flame and plasma cutting, bending, spinning, drilling, stapling, gluing, fabric cutting, sewing, tape and fiber placement, wiring, pick and place, and sawing. A specific application of NC machines or robots is in aircraft manufacturing.

[0005] Today, when programming CNC machines for drilling and / or countersunk holes, position, motion, and related processes are executed and simulated in a theoretical manner.

[0006] In CNC systems, the position of the tool is defined by a set of instructions called the part program. When parts are produced in an automated cell working on an NC machine, significant variations are theoretically observed due to the interplay of tolerances, processes, and production capacity. Methods for adapting to these variations are integrated into the NC machine's functionality, and various approaches exist for achieving this adaptation. However, the most common method is called rotational translation, which adapts the theoretical drilling diagram to the actual conditions measured by the machine. However, this adaptation to reality results in non-optimal motion to reach the target (e.g., drilling), leading to corresponding performance losses.

[0007] Therefore, there is a great need to provide a method to improve the productivity of automatically controlled drilling tools by bringing theory and practice closer together in the execution of the drilling motion and positioning process. Summary of the Invention

[0008] The present invention addresses the aforementioned problems and overcomes the limitations of the prior art as previously explained by providing an automated drilling optimization method that optimizes the performance of drilling and / or countersunk hole drilling processes performed by an automated control system (i.e., a CNC machine or robot).

[0009] One aspect of the present invention relates to a method for optimizing the execution of an automatic drilling tool controlled by a CNC machine.

[0010] This invention is applicable to the positioning of any type of machining tool (e.g., for drilling, countersunk holes, removing adhesive or sealant-type products, inspections performed by sensors, rotary tools, or any other operations required for manufacturing part assemblies) operating under automated control similar to an NC system. Preferably, but not exclusively, the invention is applied to the optimization of drilling operations in structural surfaces (e.g., machined / assembled aircraft cabins).

[0011] This invention has many advantages over the prior art, which can be summarized as follows:

[0012] - By using data generated in previous executions, this invention allows for the optimization of the production process by finding a learning point that brings theory and practice closer together.

[0013] - The learned position calculated by applying machine learning reduces empty motion and optimizes the machine localization process.

[0014] - This optimization of movement and positioning has increased machine productivity while preserving all machine processes that ensure the quality and safety of operation.

[0015] These and other advantages will become apparent from the detailed description of the invention. Attached Figure Description

[0016] For the purpose of aiding in understanding the features of the invention, and to supplement this specification, the following drawings are appended as an integral part of the invention, which has the following illustrative and non-limiting features:

[0017] Figure 1 The diagram shows the correction error distribution of a benchmark model used to evaluate the machine learning model employed in the automated drilling optimization method.

[0018] Figure 2 The diagram shows the correction error distribution of the first machine learning model used in the automated drilling optimization method.

[0019] Figure 3 The diagram shows the correction error distribution of the second machine learning model used in the automated drilling optimization method.

[0020] Figure 4 A schematic diagram summarizing the automated drilling optimization method according to a preferred embodiment is shown. Detailed Implementation

[0021] Embodiments of the invention can be implemented on various architectural platforms, operating systems, server systems, devices, systems, or applications. Any particular architectural layout or implementation presented herein is provided for illustrative and understanding purposes only and is not intended to limit aspects of the invention.

[0022] A preferred embodiment of the automated drilling optimization method uses production data from previously executed automated drilling and machine learning operations to better align theoretical positioning with actual conditions by returning to the learned position, thereby increasing the productivity of the automated drilling system. The learned position is calculated from previously executed position statistics and ultimately transmitted to the part program (or computer numerical control, CNC, file) to provide position values ​​for optimizing the drilling process.

[0023] In a preferred embodiment, each drilling operation has a unique identifier ID, defined by the CNC machine that executes the part program and controls the automatic drilling tool. Each drilling execution also has a unique identifier, allowing historical executions to be tracked via the CNC machine. The final position of each execution is recorded by the NC machine.

[0024] If multiple past executions are retrieved at the final borehole position level, a trend function can be found that allows prediction of the theoretically optimal position for executing the next borehole. This trend function (e.g., a regression algorithm) provides an intermediate position between theory and reality, which will be closer to reality due to the learned position calculated based on past executions as input to a machine learning model.

[0025] If only the theoretical position is used, the NC machine's part program first moves the machine to the drilling point determined by the theoretical position, then checks whether the drilling conditions are met at this point to ensure the quality and safety of the operation. If the quality and safety requirements are met, the drilling tool is instructed to perform / drill at the determined (theoretical) position. If neither quality nor safety requirement is met, the NC machine performs a second movement to adjust the tool's drill head to the drilling point. This results in wasted time and productivity because the first movement results in a wasted movement (i.e., a movement that adds no value to the drilling process). This adjustment for the second movement typically occurs in 90% of executions.

[0026] The primary objective of the automated drilling optimization method is to avoid adjustments for the second motion by using an intermediate position instead of a theoretical position calculated via a trend function. This intermediate position is a learned position, which is also stored by the NC machine, allowing it to be used as input for training a machine learning model. The use of the learned position reduces idle motion and optimizes the positioning process of the machining tool used for drilling and / or countersunk holes. The use of the learned position also allows for a reduction in the time elapsed from positioning to drilling commencement.

[0027] To evaluate the proposed method for using machine learning (ML) models to optimize the drilling process, a baseline model is proposed, such as... Figure 1 As shown. In a given drilling operation t (corresponding to the new hole to be drilled), the following assumptions are made:

[0028] Theory_I t =Correction_I t-1

[0029] Theory_J t =Correction_J t-1

[0030] Theory_K t =Correction_K t-1

[0031] Correction_I, Correction_J, and Correction_K are the values ​​that the machine learning model aims to predict in order to reduce the number of iterations required by the automated drilling system to correct the borehole position. Correction_I, Correction_J, and Correction_K are the components (real numbers) of the correction normal vector calculated by the NC machine for each borehole (drilling at operations t-1, t, t+1, ..., where t represents time). The normal vector is a vector (I, J, K) orthogonal to the surface of the borehole at the borehole point or position. Theoretical_I, Theoretical_J, and Theoretical_K are the theoretical values ​​for the borehole position, i.e., the theoretical components of the normal vector defined before the method optimizes.

[0032] The NC machine calculates the error generated by the automated drilling system for each t:

[0033] |Correction_I t -Theory_I t |=|Correction_I t -Correction_I t-1 |

[0034] Then, based on the calculated error, a distribution of the benchmark correction error that is superior is established, such as... Figure 1 As shown in the figure, the graph illustrates the dispersion of values, primarily moving from 0 to 0.01.

[0035] Figure 2 The first ML model is shown, which is trained to predict the normal vector (I, J, K) for each individual drilling operation t. Each normal vector predicted by ML modeling is called a correction vector, and the components of the correction vector are correction_I, correction_J, and correction_K, the values ​​of which are in sexagesimal degrees of angular motion of the tool's drill bit.

[0036] This first ML model combines three smaller models: the I model, the J model, and the K model, such as Figure 2 As shown, each small model focuses on one of the three dimensions of the correction vector.

[0037] The small I model takes a dataset containing the following values ​​as input:

[0038] [Theory_X] t Theory_Y t Theory_Z t TDrill t Quarter t ]

[0039] Furthermore, the smaller I model will correct the first component _I t The value is returned as output.

[0040] The same procedure applies to small J-models and small K-models, thus returning the second component correction _J of the correction vector. t and the third component correction_K t .

[0041] Please note:

[0042] -X t Y t and Z t This indicates the spatial coordinates at which the hole is placed in the learning position.

[0043] -Theory_X t Theory_Y t Theory Z t These are the theoretical coordinates for placing the holes. This is before applying machine learning to optimize the method.

[0044] -I, J, and K are the components of the normal vector (orthogonal to the surface of the borehole) for a given borehole (at the borehole location).

[0045] -TDrill is the identifier or name to be applied to the NC machine (given for reference only).

[0046] -Quarter is a name given for reference only to distinguish the part from the part used in the process applied to a given NC machine.

[0047] The NC machine calculates the corrected prediction error for each drilling operation t and establishes the distribution of the first correction error. For example... Figure 2 As shown, the values ​​mainly move from 0 to 0.001, which means that the first ML model achieves an accuracy of 10^-3, better than... Figure 1 The accuracy of the reference model shown is 10^-2.

[0048] Figure 3 The second ML model is shown, which is trained to predict the correction vector (I, J, K) for each individual borehole operation t.

[0049] I, J, and K are values ​​in sexagesimal units, which are the angular movements that the drill bit of the tool will make.

[0050] This second ML model combines three smaller models: the I model, the J model, and the K model, such as Figure 3 As shown, each small model focuses on one of the three dimensions of the correction vector.

[0051] The small I model takes a dataset containing the following values ​​as input:

[0052] [Theory_X] t Theory_Y t Theory_Z t TDrill t Quarter t Correction t-1 Correction_J t-1 Correction_K t-1 ]

[0053] The smaller I model will correct _I t The value is returned as output.

[0054] The same applies to small J-models and small K-models.

[0055] The NC machine calculates the corrected prediction error for each drilling operation t and establishes the distribution of the second correction error. For example... Figure 3 As shown, the values ​​mainly move from 0 to 0.0001, which means the second ML model achieves a further improvement in accuracy of 10^-4, better than... Figure 1 The accuracy of the reference model shown is 10^-2.

[0056] Figure 4 This diagram summarizes the operation of an automated drilling tool controlled by an NC or CNC machine. The NC machine checks if the drilling process is stable, i.e., whether the difference between the execution / operation (drilling a hole / drilling a countersunk hole) is less than a pre-set threshold (percentage). If stable, the NC machine then applies a statistical trend function (e.g., a regression algorithm) to estimate the most favorable (optimal) location, which reduces the time between positioning and drilling. This regression is applied to each drilling operation performed on each surface to be drilled. Each hole has an identifier ID for this application. In the machine learning of each identified hole, the theoretical location and those learned locations obtained in previous drilling operations are stored by the NC machine. In machine learning, it is well known that there is a training phase followed by a testing / operation phase. As described above, the machine learning applied by the NC machine achieves a reduction in positioning delivery time, i.e., a reduction in the delay between the start and completion of the automated drilling tool's positioning to execute drilling.

[0057] The following are the prerequisites required to ensure proper process optimization:

[0058] - At the level of automated units (where NC machines operate);

[0059] - The ability to record all relevant locations for each borehole during the execution phase.

[0060] - The ability to adapt part programs executed by NC machines.

[0061] -At the programming level:

[0062] - The ability to identify each borehole using a unique ID.

[0063] - This ID remains unchanged during each execution until the design / product of the automatic drilling system changes; if it changes, the NC machine must issue a new ID.

[0064] -At the execution level:

[0065] - The ability to retrieve relevant data from the execution process.

[0066] - The ability to avoid significant differences between executions.

[0067] - The ability to adapt part programs to learned positions calculated by machine learning.

[0068] Note that in this document, the term “comprises” and its derivatives (such as “comprising”, etc.) should not be construed as meaning exclusion; that is, these terms should not be interpreted as excluding the possibility that the content described and defined may include further elements, steps, etc.

Claims

1. A method for optimizing the execution of automatic drilling and / or countersunk hole drilling tools controlled by a CNC machine, characterized in that, The method includes the following steps: i) Identify each hole drilled by the automatic drilling tool at time t using a unique identifier defined by the CNC machine; ii) Store the theoretical position of each drilled hole at time t, the theoretical position being a pre-configured static position; iii) Use a machine learning model and calculate the learned position based on the position stored at said time t; iv) Estimate the median position by applying a trend statistical function; and If the difference between the intermediate position and the learned position is below a pre-configured threshold, the intermediate position is used to position the drilling tool so as to perform the next drilling at the next time t + 1. Otherwise, store the intermediate position at the next time t+1, and repeat steps iii) and iv). in, Step iii) of calculating the learning position at time t includes: - Calculate the normal vector orthogonal to the surface of the borehole at time t, the normal vector consisting of three components I t J t and K t definition; - The correction vector at time t is computed using a machine learning model, and the correction vector is corrected by three components _I t Correction_J t and correction_K t Defined, wherein the machine learning model will be used for component I t The first machine learning model for the component J t The second machine learning model and the component K t The third machine learning model is combined, and in which, The machine learning model will include at least three component theories _X t Theory_Y t Theory Z t The defined theoretical vector is used as input data, and the three components are spatial coordinates in three-dimensional space, which defines the theoretical location of the borehole to be drilled at the time t. The first machine learning model corrects the first component of the correction vector. t Return as output; The second machine learning model corrects the second component of the correction vector. t Returned as output; and The first machine learning model corrects the third component of the correction vector, K. t Return as output; - via the spatial coordinates X t Y t and Z t The learning position is defined in the three-dimensional space.

2. The method according to claim 1, wherein, Steps i) to iv) are performed by the CNC machine.

3. The method according to claim 1 or 2, wherein, The trend statistics function is a regression algorithm.

4. The method according to claim 1, wherein, The input data acquired by the machine learning further includes a correction vector calculated at a previous time t - 1, the correction vector having component correction _I t-1 Correction_J t-1 and correction_K t-1 .

5. The method according to any one of claims 1, 2, and 4, wherein, The components of the correction vector are calculated in degrees of the angular motion of the automatic drilling tool.

6. The method according to any one of claims 1, 2, and 4, further comprising: The error at time t is calculated by the CNC machine to determine whether the difference between the intermediate position and the learning position is below a pre-configured threshold. For the first machine learning model, the error is calculated as |correction_I|. t -Correction_I t-1 For the second machine learning model, the error is calculated as |correction_J t -Correction_J t-1 Furthermore, for the third machine learning model, the error is calculated as |correction_K|. t -Correction_K t-1 |

Images