Machine learning estimation with normalized tightening categories
Through normalization of torque and angle values and machine learning model training, the problem of difficult to accurately judge the tightening operation results in the prior art is solved, and more accurate and efficient tightening operation identification and alarm provision are achieved.
Patent Information
- Application Number
- CN202410511656.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-04
- Filing Date
- 2024-04-26
- Publication Date
- 2024-11-05
AI Technical Summary
The prior art is difficult to accurately determine the results of the tightening operation performed by the tightening tool, especially due to the complexity of torque and angle data and the time-consuming and inaccurateness of manual judgments.
By obtaining torque and angle values, identifying the screw-in and ending tightening phases, normalizing, and using machine learning models to determine the tightening category, providing alerts to guide the operator.
Improves the accuracy and efficiency of tightening operations, reduces undesired tightening results, and enhances the ability to identify tightening operation types.
Smart Images

Figure CN118893594A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for being able to determine the tightening category of a tightening operation performed by a tightening tool and a device for carrying out the method.
[0002] Furthermore, a computer program comprising computer executable instructions is provided for causing the apparatus to perform the steps of the method when the computer executable instructions are executed on a processing unit comprised in the apparatus.
[0003] Furthermore, a computer program product comprising a computer readable medium having the computer program embodied thereon is provided. Background Art
[0004] In the process of tightening a fastener such as a bolt or a screw using a tightening tool, several undesirable tightening results may occur to the tightened bolt or screw.
[0005] Analyzing sensor data from a tightening tool provides valuable insights into tightening results. Analytical models can be developed to determine tightening results based on sensor data from a tightening tool.
[0006] However, whether the tightening result is correct may be difficult to determine by a human operator or a machine, and is more cumbersome and time-consuming.
[0007] To this end, machine learning (ML) can be used to analyze the tightening results of torque and angle. However, proper training of the ML model is crucial for the ML model to subsequently perform accurate tightening estimation. Summary of the invention
[0008] An object is to solve or at least alleviate this problem in the prior art and therefore to provide an improved method for being able to determine the tightening class of a tightening operation performed by a tightening tool.
[0009] In a first aspect, a method for an apparatus capable of determining a tightening category of a tightening operation performed by a tightening tool is provided. The method comprises: obtaining a set of measured torque and angle values of a fastener tightened by the tightening tool, identifying a screw-in phase and an end tightening phase of the tightening of the fastener from the obtained set of measured torque and angle values, normalizing the torque value of the end tightening phase using a determined torque value range of the end tightening phase and normalizing the angle value of the end tightening phase using a determined angle value range of the end tightening phase, and training a machine learning model using the normalized torque and angle values of the end tightening phase and at least one tightening category associated with the normalized torque and angle values of the end tightening phase, wherein the tightening category identifies the type of tightening operation applied to the fastener.
[0010] In a second aspect, a device configured to determine a tightening category of a tightening operation performed by a tightening tool is provided. The device includes a processing unit, which operates to cause the device to: obtain a set of measured torque and angle values of a fastener tightened by the tightening tool, identify a screw-in phase and an end tightening phase of the fastener tightening from the obtained set of measured torque and angle values, normalize the torque value of the end tightening phase using a determined torque value range of the end tightening phase and normalize the angle value of the end tightening phase using a determined angle value range of the end tightening phase, and train a machine learning model using the normalized torque and angle values of the end tightening phase and at least one tightening category associated with the normalized torque and angle values of the end tightening phase, the tightening category identifying the type of tightening operation applied to the fastener.
[0011] Advantageously, by separating the torque and angle values of the screw-in phase from the torque and angle values of the final tightening phase and normalizing these two data subgroups separately, the accuracy of normalization is greatly improved. Specifically, for the angle values of the final tightening phase, they are typically normalized using tens of angle values (e.g., 30° to 40°) rather than thousands of angle values (e.g., 3000° to 4000°) that constitute the screw-in phase.
[0012] In one embodiment, a further acquired and normalized set of measured torque and angle values of a fastener tightened by a tightening tool at the end of a tightening phase is further provided to the trained machine learning model, wherein the trained machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values at the end of a tightening phase.
[0013] In one embodiment, in the process of identifying the screw-in phase and the end tightening phase of fastener tightening from a set of measured torque values and angle values obtained, it is further determined whether the obtained torque value exceeds a predetermined torque threshold, and if so, the obtained torque value and the corresponding angle value are determined to belong to the end tightening phase, and if not, the obtained torque value and the corresponding angle value are determined to belong to the screw-in phase.
[0014] In one embodiment, the normalization further includes: normalizing the torque value of the screw-in phase using a determined torque value range of the screw-in phase and normalizing the angle value of the screw-in phase using a determined angle value range of the screw-in phase; training the machine learning model further includes: training the machine learning model using the normalized torque and angle values of the screw-in phase and at least one tightening category associated with the normalized torque and angle values of the screw-in phase.
[0015] In one embodiment, providing a further acquired and normalized set of measured torque and angle values for the end of the tightening stage to the trained machine learning model further includes: providing a further acquired and normalized set of measured torque and angle values of the fastener tightened by the tightening tool in the screw-in stage to the trained machine learning model, wherein the trained machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values for the screw-in stage.
[0016] In one embodiment, training the machine learning model further includes: training a first machine learning model using the normalized torque and angle values of the end tightening stage and at least one tightening category associated with the normalized torque and angle values of the end tightening stage, and training a second machine learning model using the normalized torque and angle values of the screw-in stage and at least one tightening category associated with the normalized torque and angle values of the screw-in stage.
[0017] In one embodiment, providing a further acquired and normalized set of measured torque and angle values for the end of the tightening phase to the trained machine learning model further includes: providing a further acquired and normalized set of measured torque and angle values for the end of the tightening phase to the trained first machine learning model, wherein the trained first machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values for the end of the tightening phase, and providing a further acquired and normalized set of measured torque and angle values for the screw-in phase to the trained second machine learning model, wherein the trained second machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values for the screw-in phase.
[0018] In one embodiment, the determined torque value range and / or angle value range is divided into smaller sub-ranges for normalization.
[0019] In one embodiment, the normalization performed includes maximum-minimum normalization.
[0020] In one embodiment, an alert is provided indicating at least one estimated tightening category.
[0021] In one embodiment, the alert is provided to an operator of the tightening tool, the tightening tool itself, a supervisory control room, or a remote cloud function.
[0022] In a third aspect, there is provided a computer program comprising computer executable instructions, which, when executed on a processing unit included in an apparatus, are used to cause the apparatus to perform the steps described in the method of the first aspect.
[0023] In a fourth aspect, there is provided a computer program product comprising a computer readable medium having embodied thereon the computer program according to the third aspect.
[0024] Generally, all terms used in the claims should be interpreted according to their ordinary meaning in the technical field, unless otherwise explicitly defined herein. Unless otherwise explicitly stated, all references to "element, device, component, means, step, etc." should be openly interpreted as referring to at least one example of the element, device, component, means, step, etc. Unless explicitly stated otherwise, the steps of any method disclosed herein do not have to be performed in the exact order disclosed. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Various aspects and embodiments will now be described by way of example with reference to the accompanying drawings, in which:
[0026] Figure 1 A tightening tool configured to apply torque to a fastener is shown in which tool embodiments may be implemented;
[0027] Figure 2a to Figure 2d shows different types of undesirable tightening operations that may be performed by a tool during a tightening procedure;
[0028] Figure 3 The training of a machine learning (ML) model is shown;
[0029] Figure 4 Shows the use of Figure 3 A trained ML model to estimate the tightening category;
[0030] Figure 5A flow chart illustrating a method for determining a tightening category of a tightening operation performed by a tightening tool is shown;
[0031] Figure 6a and Figure 6b The screw-in phase and the final tightening phase of the trace are shown;
[0032] Figure 7 A flow chart illustrating a method of estimating a tightening category according to one embodiment is shown;
[0033] Figure 8 shows the training of an ML model according to one embodiment;
[0034] Fig. 9 shows the use of Figure 8 A trained ML model to estimate the tightening category;
[0035] Fig.10 shows a flow chart illustrating a method of estimating a tightening category according to another embodiment;
[0036] Fig.11 shows training of an ML model according to another embodiment; and
[0037] Fig.12 shows the use of Fig.11 A trained ML model to estimate the tightening category. DETAILED DESCRIPTION
[0038] Various aspects of the invention will be described more fully hereinafter with reference to the accompanying drawings, in which specific embodiments of the invention are shown.
[0039] However, these aspects can be embodied in many different forms and should not be interpreted as limiting; rather, these embodiments are provided by way of example so that the present invention will be detailed and complete and the scope of all aspects of the present invention will be fully conveyed to those skilled in the art. Throughout the specification, the same reference numerals represent the same elements.
[0040] Figure 1 An industrial tool in which tool embodiments may be implemented is shown in the form of a tightening tool 10 configured to apply torque to a fastener (eg, a bolt 25) for tightening a joint.
[0041] The tightening tool 10 may be cordless or powered via a wire and has a body 11 and a tool head 12. The tool head 12 has an output shaft 13 having a sleeve (not shown) configured to be rotatably driven by a motor disposed inside the body 11 to apply torque to the bolt 25.
[0042] The tightening tool 10 may be arranged with a display 14 , via which information relating to the operation of the tool 10 may be presented to an operator of the tool 10 , and an interface 15 , via which the operator may input data into the tool 10 .
[0043] The tightening tool 10 may further be provided with communication capabilities in the form of a radio transmitter / radio receiver 16 for wirelessly transmitting operating data (e.g., applied torque) to a remotely located controller (e.g., a cloud server 30) or a device such as a server executing on-site. Alternatively, communication between the tool 10 and the controller may be performed via a wired connection.
[0044] Thus, for example, the tool 10 can transmit measured operational data to the controller for further evaluation, while, for example, the controller can send operational settings applied by the tool 10 or instructions displayed to the operator via the display 14, or even automatically configure the tool 10. As will be appreciated, the method of determining the configuration of the tool 10 according to the embodiment can be performed in the tool 10 or the cloud server 30 (or even in a combination, with some steps performed in one device and other steps performed in another device). Thus, as will be described below, the tool 10 is typically equipped with a control device 20, and the cloud server 30 includes a similar control device 35, which houses the same or similar data processing components.
[0045] The steps of the method performed by the tool 10 and / or the cloud server 30 to be described below are actually performed by the control device 20 and / or the control device 35, respectively, which includes a processing unit 17, 32 implemented in the form of one or more microprocessors, and the processing unit 17, 32 is arranged to execute a computer program 18, 33 downloaded to a storage medium 19, 34 associated with the microprocessor (for example, a random access memory (RAM), flash memory or hard drive). When a suitable computer program 18, 33 containing computer executable instructions is downloaded to the storage medium 19, 34 and executed by the processing unit 17, 32, the processing unit 17, 32 is arranged to cause the tool 10 and / or the cloud server 30 to perform a method according to the embodiment. The storage medium 19, 34 can also be a computer program product including the computer program 18, 33. Alternatively, the computer program 18, 33 may be transferred to the storage medium 19, 34 by means of a suitable computer program product (e.g., a Digital Versatile Disc (DVD) or a memory stick). As another option, the computer program 18, 33 may be downloaded to the storage medium 19, 34 via a network. Alternatively, the processing unit 17, 32 may be implemented in the form of a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), a Complex Programmable Logic Device (CPLD), etc. The control device 20, 35 is communicatively connected to an interface for external communication (e.g., from the tool 10 to the cloud server 30, and vice versa).
[0046] The control device 20 may be arranged inside the tightening tool 10 or connected to the tool 10 , for example as a control device 20 attached to the outside of the body 11 of the tool 10 .
[0047] The processing unit 17 may be communicatively coupled to one or more sensors (not shown) for measuring the torque applied to the bolt 25 and the angle of rotation of the output shaft 13 of the tightening tool 10 when the torque is applied.
[0048] Now, when the operator uses the tightening tool 10 to tighten a fastener such as a bolt 25, it is important to correctly perform the tightening operation to maintain the tightening durability of the tightened bolt 25. If not, there is a risk of tightening deterioration, which may cause the bolt 25 to unscrew in the worst case. Therefore, it is crucial to correctly tighten the bolt 25, and if not, accordingly, it is desirable to obtain an indication so that the operator can correctly retighten the bolt 25 using the tool 10. In general, incorrect tightening is referred to as not ok (NOK) tightening, and conversely, correct tightening is referred to as OK.
[0049] One such tightening operation performed by the tool 10 (and its human operator) is called a breakaway. During a breakaway, the torque value suddenly drops to zero before the tightening procedure being performed is completed.
[0050] For example, if the torque value suddenly drops to zero before tightening is complete, disengagement between the output shaft 13 (and sleeve) and the bolt 25 may occur. This may be caused by operator behavior, worn tool parts and / or fasteners, incorrect tool programming, etc.
[0051] Figure 2a to Figure 2d Various undesirable tightening operations are shown as being performed by the tool 10. The combination of torque and angle values of a performed tightening procedure is sometimes referred to as a trace.
[0052] like Figure 2a to Figure 2d As indicated by the trace of , the X-axis (i.e., the angle scale) is simplified for the following reasons: in the so-called screw-in phase, the torque may be relatively constant for hundreds or even thousands of degrees (e.g., 3000° to 4000°) of the output shaft 13 rotation angle, while in the so-called end tightening phase, the torque increases rapidly, which may only last for an angle interval of tens of degrees (e.g., 20° to 30°). This will be discussed in further detail below.
[0053] Figure 2a One specific type of tightening operation, commonly referred to as socket slip, is shown, which may occur when torque values increase and decrease at periodic angular intervals.
[0054] Figure 2b Another specific type of tightening operation, commonly referred to as stick slip, is shown, which may occur when torque values fluctuate at a high frequency.
[0055] Figure 2cAnother specific type of tightening operation is shown, called thread lock, which may occur when the applied torque reaches the target torque immediately after the screw-in phase. This is usually caused by a sudden locking of the threads of the tightened bolt 25, for example due to the bolt 25 entering the thread at an angle, or the incorrect size.
[0056] at last, Figure 2d Another specific type of tightening operation, called high screw-in torque, is shown, which may occur when the applied torque reaches the target torque during the screw-in phase. This is usually caused by higher than normal friction in the threads of the bolt 25, for example due to the bolt 25 having an incorrect size, the wrong thread coating or the bolt 25 tilting when entering the thread.
[0057] As understood, these are examples of incorrect tightening operations performed by the tool 10 and should preferably be avoided. However, if they still occur, it is desirable to notify the operator or some supervisory function of them.
[0058] Therefore, in one embodiment, a trained machine learning (ML) model is used to determine whether an incorrect tightening operation is performed based on the acquired torque and angle values, and if so, which specific type of tightening operation is performed, that is, whether it is one of the above-mentioned "breakaway", "sleeve slip", "stick slip", "thread locking" and "high screw-in torque" (or any other identified tightening operation). Figure 2a to Figure 2d There are many different types besides those shown.
[0059] Figure 3 The training of the ML model according to the example is shown, while Figure 4 Utilizing a trained ML model to estimate a tightening class for a provided set of torque and angle values during a tightening operation is shown. Figure 5 , Figure 5 A flow chart illustrating a method for determining a tightening category of a tightening operation performed by the tightening tool 10 is shown.
[0060] Therefore, in the training phase, one or more sets of measured torque values T and angle values A of the bolts 25 tightened by the tightening tool 10 are obtained in step S101. Therefore, these are previously referenced Figure 2a to Figure 2d The graph shows the torque and angle values.
[0061] For example, each group may include the one hundred torque values and corresponding angle values obtained by the processing unit 17 when the tightening tool 10 performs the tightening procedure. In addition, hundreds or even thousands of groups of torque and angle values can be provided to the ML model for thorough training.
[0062] Any acquired torque and angle values may be stored locally in memory 19 and / or wirelessly transmitted via radio transmitter / radio receiver 16 to a remotely located cloud server 30 or other suitable device separate from tool 10 , such as a server located on-site.
[0063] In step S102, a tightening category TC is associated with each set of torque and angle values obtained. Thus, the tightening category identifies the type of tightening operation applied to the fastener. As previously illustrated, the tightening category may include "breakaway", "sleeve slip", "stick-slip", "thread locking", "high screw-in torque", etc. or any other identified type of tightening operation.
[0064] In practice, the operator of the tool 10 will typically identify a particular tightening category associated with a set of measured values based on the acquired set of torque and angle values. For example, if the measured torque value T and angle value A have the same value when presented to the operator via the display 14, Figure 2b , the operator concludes that the associated tightening category should be “stick-slip” and inputs information identifying the tightening category to the tool 10 via the interface 15 .
[0065] In step S103 , the identified tightening category is provided to the ML model for training together with a set of acquired torque and angle values.
[0066] This is typically repeated for a large number of sets of torque and angle values measured, and thus, at step S103 , the ML model will effectively be trained to associate a tightening class TC with each set of torque and angle values T, A provided.
[0067] Although it is possible to utilize a single-label tightening ML model in which each set of torque and angle values T, A measured is associated with only one tightening category TC (such as "stick-slip", "sleeve slip", "thread locking", etc. discussed above), Figure 3 As shown, it is also conceivable to use a multi-label tightening ML model, in which each set of torque and angle values T, A measured is associated with a plurality of tightening categories TC1, TC2, ..., TCn.
[0068] As an example, it is conceivable to define the above-mentioned tightening category TC1 using the first tightening category TC1. Figure 2a to Figure 2d The invention also provides a method for indicating each of the tightening operation types (i.e., “sleeve slip”, “stick slip”, “thread locking” and “high screw-in torque”) of the present invention, while utilizing a supplementary second tightening category TC2 to indicate an OK / NOK tightening operation.
[0069] As understood, the tightening tool 10 can generally indicate OK / NOK tightening based on a predetermined threshold, while the ML model can also indicate OK / NOK tightening based on training data. Therefore, the OK / NOK estimates of the tightening tool 10 and the ML model can be evaluated separately, wherein the ML model provides additional opinions on the tightening operation performed.
[0070] For example, if the tightening tool 10 itself (ie, the processing unit 17 ) indicates NOK operation, while the trained ML model indicates OK operation—or vice versa—there may be a problem, and the operator of the tool may be alerted accordingly.
[0071] In another example, it may also be that for the same trace, the trace indicates two different types of tightening operations performed. For example, a trace may constitute both a "sleeve sliding" and a "stick sliding", in which case (at least) two categories are assigned to the trace; TC1 = "sleeve sliding" and TC2 = "stick sliding" to indicate that the trace represents both a sleeve sliding operation and a stick sliding operation, and if appropriate, even more categories such as TC3 = "thread locking". Thus, in a multi-label setup, multiple exclusive tightening categories can be correctly assigned to the same trace.
[0072] In another example, there are usually two stages during the tightening operation, the first stage is the screw-in stage when the resistance of the joint tightened by the bolt 25 is low, and the second stage is the end tightening stage when the tool 10 slows down as the torque increases. Alternatively, the end tightening stage can be called a torque building stage. Therefore, it is desirable to assign a first tightening category TC1 (e.g., "OK / NOK") to the torque and angle values T, A measured in the initial screw-in stage, and to assign a second tightening category TC2 (e.g., "OK / NOK") to the torque and angle values T, A at the final end tightening stage.
[0073] In another example, it is conceivable that multiple tightening categories (such as TC1 and TC2) are associated with one trace, for example, TC1 classifies the screw-in phase of the trace, while TC2 classifies the end tightening phase. For example, for a particular trace, it is conceivable that TC1 = "sleeve slide screw-in" and TC2 = "sleeve slide end tightening".
[0074] Furthermore, a tightening category TC3 may be introduced which indicates whether the bolt 25 is damaged for some reason during the tightening operation, and even a tightening category TC4 may be introduced which indicates whether there is an indication of damage to the joint to be tightened by the bolt 25. As will be appreciated, a number of tightening categories TC1, TC2, ..., TCn may be associated with a tightening operation.
[0075] Therefore, if Figure 4 As shown, when a tightening procedure is subsequently performed by the tool 10 , a further acquired set of measured torque and angle values T, A of the performed tightening procedure is provided to the trained ML model at step S104 .
[0076] As a result, for another set of measured torque and angle values T, A provided to the trained ML model during the tightening operation, the trained ML model will advantageously output (e.g., via the display 14) one or more estimated tightening categories TC1, TC2, ..., TCn so that the operator can determine whether to re-execute the tightening procedure. Thus, the trained ML model will help to detect the tightening category TC based on the torque and angle values T, A recorded during the tightening operation provided.
[0077] Alternatively, instead of having the control device 20 of the tool 10 perform steps S101 to S104, it is conceivable that the torque and angle values T, A are transmitted by the tool 10 via the radio transmitter / radio receiver 16 to the cloud server 30, which is equipped with a corresponding control device 35 for performing steps S101 to S104, which will reduce the computational burden of the processing unit 17 of the tool 10. The cloud server 30 can send information indicating the estimated tightening category to the tool 10 via the radio transmitter / radio receiver 16, and the tool 10 presents an alarm of the tightening category to the operator via the display 14, or via a small speaker included in the tool 10 in an audible manner.
[0078] In another alternative, the training of the ML model as performed in steps S101 to S103 is performed by the cloud server 30, which then provides the trained ML model to the tool 10. The tool 10 may then store the trained ML model in the memory 19 and thus perform step S104 locally.
[0079] It is conceivable that the cloud server 30 has access to a large number of torque and angle values for tightening tools of the same type as the tool 10, thus facilitating the training of the ML model by utilizing a large amount of training data.
[0080] As previously combined Figure 2a to Figure 2d As mentioned, for the rotation angle of the output shaft 13 of the tool 10 , the screw-in phase may last for several thousand degrees, whereas the final tightening phase lasts only for tens of degrees.
[0081] Figure 6a The screw-in phase lasts for more than 4000°, while Figure 6b The end tightening phase is shown to last only about 35°. In other words, the range of the raw data values varies greatly.
[0082] Now, since ML models cannot work properly with such a wide range of data, ML models are usually performed with data normalized to the range of [0, 1]. That is, the data used to train the ML model should actually be in the range of 0 to 1. This is usually called feature scaling. Usually, max-min normalization is applied, in which the data is normalized using the total data range.
[0083] Usually, the maximum and minimum normalization of [0, 1] is defined as:
[0084] Where x is the original value, and x' is the normalized value, and max(x) is the maximum value in the range to be normalized, and min(x) is the minimum value in the range to be normalized. In the above-mentioned example of the angle range of the tightening tool 10, the minimum value is usually zero. When using the ML model, the maximum and minimum values in the range to be normalized are calculated based on the training set of data. Since the maximum and minimum normalization depends largely on the highest and lowest values of the data set, the training data from the tightening operation (which has the highest and lowest values) can be used to determine max(x) and min(x).
[0085] Reference again Figure 6a and Figure 6b , assuming that the rotation angle applied during the screw-in phase and the final tightening phase totals about 4200°, the complete final tightening phase will be normalized to 35° / 4200°=0.0083, thus equivalent to 0.83% of the total angular range. In other words, the final tightening phase only occupies 0.83% of the total normalized range, which makes the ML model training inefficient. Although Figure 6a The applied rotation angle is shown to be up to 4200°, but in practice the angle may be larger, for example up to about 30000° (in this case the angle of 35° would be normalized to 35° / 30000°=0.12%).
[0086] To solve this problem, in one embodiment, a set of torque and angle values measured (which form a track) is separated into a subgroup of screw-in data and a subgroup of end-tightening data. Afterwards, the two subgroups, or at least the end-tightening subgroup, are normalized separately. As will be appreciated, the end-tightening stage may be more interesting to analyze than the screw-in stage, in which case the screw-in data subgroup may be discarded.
[0087] use Figure 6a and Figure 6bIn the example shown, the data of the screw-in stage and the end-tightening stage are separated into a screw-in data subgroup and an end-tightening data subgroup, and then the two separated data subgroups are normalized using the range of torque and angle values of each subgroup. The maximum angle value of the end-tightening stage is normalized to 35 / 35 by the maximum and minimum values, instead of 35 / 4200 discussed previously.
[0088] In addition, if Figure 6a As shown, the torque in the screw-in stage is about 0.2 Nm, which is normalized to 0.2 Nm / 0.2 Nm, rather than 0.2 Nm / 45 Nm (the 45 Nm is the torque range of the total tightening stage, ie, the maximum torque value applied at the end of the tightening stage).
[0089] Figure 7 A flow chart illustrating a method for estimating a tightening class of a tightening operation performed by a tightening tool 10 according to one embodiment is shown.
[0090] Further references Figure 8 , Figure 8 shows the training of an ML model according to one embodiment, and Fig. 9 Estimation of a tightening class using a trained ML model for a provided normalized set of torque and angle values during a tightening operation is shown according to one embodiment.
[0091] according to Figure 7 As further illustrated in the embodiments, normalization is often performed as a preprocessing procedure before training the ML model.
[0092] Therefore, with reference Figure 5 Similar to step S101 described in the flowchart of FIG. 1 , in step S201 , during the training phase, one or more sets of measured torque values T and angle values A of the bolts 25 tightened by the tightening tool 10 are obtained.
[0093] For example, each group may include one hundred torque values and corresponding angle values acquired by the processing unit 17 when the tightening tool 10 performs a tightening procedure. In addition, hundreds or even thousands of groups of torque and angle values may ultimately be provided to the ML model for comprehensive training.
[0094] However, in step S202 , the method according to the embodiment identifies the screw-in phase and the end-tightening phase of the tightening performed, so that the torque and angle values of each of the two phases can be separated into screw-in subgroups TR, AR and end-tightening subgroups TE, AE.
[0095] Then, in step S203:
[0096] - Use the torque value range and angle value range of the screw-in phase respectively (in Figure 6a and Figure 6b In the example, 0.2 Nm and 4200° are used to normalize the maximum and minimum values of the torque and angle values TR and AR in the screw-in stage, thereby obtaining the normalized torque and angle values TR_N and AR_N in the screw-in stage.
[0097] - Use the torque value range and angle value range of the end tightening phase respectively (in Figure 6a and Figure 6b The torque and angle values TE and AE at the end of the tightening stage are normalized to their maximum and minimum values (40 Nm and 35° in the example), thereby obtaining the normalized torque and angle values TE_N and AE_N at the end of the tightening stage.
[0098] As understood, although in the description of step S202 , both TR, AR and TE, AE are output, it is conceivable that only TE, AE are output in step S202 and normalized in step S203 .
[0099] In step S204, the normalized set of torque and angle values TR_N, AR_N and TE_N, AE_N is provided to the ML model together with one or more tightening categories TC1, TC2, ..., TCn (the one or more tightening categories TC1, TC2, ..., TCn are associated with each normalized set of torque and angle values TR_N, AR_N and TE_N, AE_N). Again, it is conceivable that in step S204, only TE, AE are normalized and provided to the ML model.
[0100] Thus, the tightening category (at least) identifies the type of tightening operation applied to the fastener. As previously illustrated, the tightening categories may include "breakaway", "sleeve slip", "stick-slip", "thread locking", and "high screw-in torque", etc. or any other identified type of tightening operation. In this particular example, TC1 = "sleeve slip screw-in" may be associated with the normalized screw-in group TR_N, AR_N, while TC2 = "sleeve slip end tightening" may be associated with the normalized end tightening group TE_N, AE_N, and possibly, there may be more tightening categories, such as a third category TC3 indicating whether the bolt is damaged, etc.
[0101] Therefore, in step S204, the identified multiple tightening categories are provided to the ML model for training together with the normalized multiple sets of torque and angle values TR_N, AR_N and TE_N, AE_N.
[0102] This is typically repeated for a large number of sets of measured torque and angle values, and thus, at step S204, the ML model will effectively be trained to associate at least one tightening category TC with each provided normalized set of torque and angle values TR_N, AR_N and TE_N, AE_N.
[0103] As will be discussed in more detail below, the trained ML model may then be used to estimate the tightening class at step S205 .
[0104] Advantageously, by separating the torque and angle values TR, AR of the screw-in stage from the torque and angle values TE, AE of the end tightening stage, and normalizing these two data subgroups respectively (obtaining two normalized data subgroups TR_N, AR_N and TE_N, AE_N), the normalization accuracy is greatly improved. Specifically, the angle value of the end tightening stage is normalized using 35° instead of +4000°, and the torque value of the screw-in stage is normalized using 0.2Nm instead of 45Nm.
[0105] Although Figure 7 and Figure 8 It is shown that the torque and angle values TR, AR and TE, AE of the screw-in phase and the end tightening phase are normalized in step S203 and provided to the ML model in step S204, but in another embodiment, it is conceivable that only the torque and angle values TE, AE of the end tightening phase are normalized in step S203 and provided to the ML model in step S204, in which case the torque and angle values TR, AR of the screw-in phase can be discarded; a given trace is usually characterized by its end tightening pause, because in many cases, the screw-in phase has a similar appearance even for traces of different classifications. For example, return to Figure 2a and Figure 2b The screw-in phase of the sleeve sliding operation is roughly the same as the screw-in phase of the stick-sliding operation, but the end tightening phase is quite different. Therefore, in step S204, only the torque and angle values TE and AE of the end tightening phase can be used to train the ML model.
[0106] Fig. 9 It is shown that the tightening category TC is estimated by using the trained ML model for a further acquired set of measured torque and angle values T, A, before which, as described above with reference to Figure 7 and Figure 8 As described in steps S202 and S203, the screw-in phase and the end tightening phase are identified and the torque and angle values of each phase are normalized, so that the normalized torque and angle values TR_N, AR_N of the screw-in phase and the normalized torque and angle values TE_N, AE_N of the end tightening phase are provided to the trained ML model.
[0107] Therefore, if Figure 7 As shown, in step S205, a further acquired set of measured torque and angle values, including normalized torque and angle values TR_N, AR_N of the screw-in phase and normalized torque and angle values TE_N, AE_N of the end tightening phase, are provided to the trained ML model for estimating one or more tightening categories of the bolt tightened by the tool 10.
[0108] So, for example, assuming:
[0109] - A set of torque and angle values TR_N, AR_N of the screw-in phase after further acquisition and normalization indicates that a sleeve sliding tightening operation is performed, and the trained ML model will indeed output TC1 = "sleeve sliding screw-in", and
[0110] - A set of torque and angle values TE_N, AE_N at the end of the tightening stage after further acquisition and normalization indicates that a sleeve sliding tightening operation is performed, and the trained ML model will indeed output TC2 = "sleeve sliding ends tightening" as the tightening category estimate.
[0111] As previously described, at step S205 , the trained ML model may evaluate only the end-tightening phase data, in which case the screw-in phase data may be discarded prior to normalization at step S203 .
[0112] Fig.10 It is shown that in the implementation scheme, based on a set of measured torque and angle values T, A obtained in step S201, by determining in step S202a whether the obtained torque value exceeds a predetermined torque threshold Thres, the screw-in phase and the end tightening phase of the bolt 25 are identified in step S202.
[0113] As before Figure 6a As shown, in the screw-in stage, the torque is about 0.2 Nm, and once the end tightening stage is reached, the torque increases rapidly. Therefore, in one example, the threshold Thres can be set to Thres=0.5 Nm, and for the torque value T (and the corresponding angle value A) exceeding the threshold Thres, the measured torque and angle data T, A are determined to belong to the end tightening stage. If not exceeded, the measured data T, A are determined to belong to the screw-in stage. Thereafter, the process proceeds to steps S203 to S205 as described above.
[0114] As will be appreciated, it is possible to determine by training an ML model whether a particular portion of a trace belongs to a screw-in phase or an end tightening phase. For example, the screw-in phase and / or the end tightening phase may be manually indicated by an operator, respectively, and based on this, one or more ML models are trained to distinguish between the two. Advantageously, this may provide a more sophisticated and potentially more accurate separation between the two phases.
[0115] Fig.11 Another embodiment is shown, which is Figure 8 The difference between the implementation schemes is that in S204, the normalized torque and angle values TR_N, AR_N of the screw-in phase and the normalized torque and angle values TE_N, AE_N of the end tightening phase are provided to different ML models (i.e., ML model 2 and ML model 1) together with the associated tightening categories TC1, TC2, ..., TCn for training.
[0116] Advantageously, each of the two ML Models 2 and ML Model 1 can be respectively applied to the end tightening phase and the screw-in phase, which enables more accurate training of the ML models in the case of input data for training.
[0117] Accordingly, Fig.12 An implementation scheme is shown, which in S205 provides the further acquired and normalized torque and angle values TR_N, AR_N of the screw-in phase and the further acquired and normalized torque and angle values TE_N, AE_N of the end tightening phase to different trained ML models, namely ML model 2 and ML model 1, for estimating tightening categories TC1, TC2, ..., TCn.
[0118] In another embodiment, instead of using the complete angle and torque range of each stage for normalization, that is, in the example given above, for the screw-in stage, the determined torque range is 0.2 Nm, the determined angle range is approximately 4200°, and for the end tightening stage, the torque range is 40 Nm, and the angle range is 35°, normalization can be performed by dividing the data of the screw-in stage and / or the data of the end tightening stage into smaller sub-ranges.
[0119] For example, instead of using the entire angular range of 4200° of the screw-in phase for normalization, the angular range is divided into 12 sub-ranges, each of which reaches a range of 4200° / 12=350°. This provides a maximum-minimum normalization with a normalization accuracy that is 12 times higher.
[0120] Similarly, the angle range for the end of the tightening phase can be divided into smaller sub-ranges, such as 35° / 7=5°, which results in a higher normalized accuracy.
[0121] In an embodiment, methods for training ML models include neural networks, random forest-based classification, regression analysis, etc. One advantage of neural networks is that they can learn to extract both temporal and spatial patterns from a data set.
[0122] The training of the ML model and the subsequent determination of the tightening class for the torque and angle values measured during the tightening procedure may be performed locally by the processing unit 17, or by the cloud server 30. For example, the determined tightening class may be immediately communicated to the operator of the tool 10 via the display 14. As a result, the operator may unscrew the bolt 25 and perform a new tightening procedure in order to obtain a proper tightening.
[0123] It is conceivable that steps S201 to S205 of the method according to the embodiment may be performed by a control device 20 arranged in the tool 10 itself. However, it is also conceivable that the torque and angle values and the background data are measured by the tool 10 and then transmitted via the radio transmitter / radio receiver 16 to a cloud server 30, which is equipped with a corresponding control device 35 for performing the TC estimation and all subsequent steps S201 to S205, which will reduce the computational burden of the processing unit 17.
[0124] It is further conceivable that the tool 10 performs all steps S201 to S205 and then alerts the cloud server 30 of the estimate of the tightening category (possibly simultaneously alerting the operator), which in turn can save a database or provide an alarm, for example, to a supervisory control room.
[0125] Aspects of the invention have been described above mainly with reference to some embodiments and examples thereof. However, as readily appreciated by a person skilled in the art, other embodiments than the ones disclosed above are equally possible within the scope of the invention as defined by the appended patent claims.
[0126] Therefore, although various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for illustrative purposes only and are not intended to be limiting, with the true scope and spirit being indicated by the appended claims.
Claims
1. A method for a device (20, 35) capable of determining a tightening category of a tightening operation performed by a tightening tool (10), the method comprising: Acquiring (S201) a set of measured torque and angle values of a fastener (25) tightened by a tightening tool (10); Identifying (S202) a screw-in phase and an end-tightening phase of tightening a fastener (25) from a set of torque and angle values measured and acquired; The torque value at the end of the tightening stage is normalized using the determined torque value range at the end of the tightening stage (S203), and the angle value at the end of the tightening stage is normalized using the determined angle value range at the end of the tightening stage (S203); as well as A machine learning model is trained (S204) using the normalized end-tightening phase torque and angle values and at least one tightening category associated with the normalized end-tightening phase torque and angle values, wherein the tightening category identifies the type of tightening operation applied to the fastener (25).
2. The method according to claim 1, further comprising: A further acquired and normalized set of measured torque and angle values of a fastener tightened by a tightening tool (10) at the end of a tightening phase is provided (S205) to the trained machine learning model, wherein the trained machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values at the end of a tightening phase.
3. The method according to claim 1 or 2, further comprising: In the process of identifying (S202) the screw-in phase and the end tightening phase of the tightening of the fastener (25) from the acquired set of measured torque and angle values, determining (S202a) whether the acquired torque value exceeds a predetermined torque threshold; and if so: The torque value and the corresponding angle value obtained are determined to belong to the end tightening stage; and if not: The acquired torque values and the corresponding angle values are determined to belong to the screw-in phase.
4. A method according to any one of the preceding claims, wherein: Normalization (S203) further includes: The torque value of the screw-in stage is normalized using the determined torque value range of the screw-in stage (S203) and the angle value of the screw-in stage is normalized using the determined angle value range of the screw-in stage (S203); and training (S204) the machine learning model further includes: The machine learning model is trained ( S204 ) using the normalized torque and angle values of the screw-in phase and at least one tightening category associated with the normalized torque and angle values of the screw-in phase.
5. The method according to claim 4, wherein: Providing (S205) a further acquired and normalized set of measured torque and angle values for the end of the tightening phase to the trained machine learning model further comprises: A further acquired and normalized set of measured torque and angle values of a fastener tightened by a tightening tool (10) at a screw-in stage is provided to the trained machine learning model, wherein the trained machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values at a screw-in stage.
6. The method according to claim 4 or 5, wherein: Training (S204) the machine learning model further includes: training a first machine learning model using the normalized end-tightening phase torque and angle values and at least one tightening category associated with the normalized end-tightening phase torque and angle values; and The second machine learning model is trained using the normalized torque and angle values of the screw-in phase and at least one tightening category associated with the normalized torque and angle values of the screw-in phase.
7. The method according to claims 5 and 6, wherein: Providing (S205) a further acquired and normalized set of measured torque and angle values for the end of the tightening phase to the trained machine learning model further comprises: providing a further acquired and normalized set of measured end-tightening-phase torque and angle values to the trained first machine learning model, wherein the trained first machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured end-tightening-phase torque and angle values; and A further acquired and normalized set of measured torque and angle values of the screw-in phase is provided to the trained second machine learning model, wherein the trained second machine learning model outputs at least one estimated tightening category for the further provided normalized set of measured torque and angle values of the screw-in phase.
8. The method according to any of the preceding claims, the determined torque value range and / or angle value range being divided into smaller sub-ranges for normalization.
9. The method according to any one of the preceding claims, wherein the normalization performed comprises maximum-minimum normalization.
10. The method according to any one of the preceding claims, further comprising: An alert is provided indicating at least one estimated tightening category.
11. The method according to claim 10, wherein: The alert is provided to an operator of the tightening tool (10), the tightening tool (10) itself, a supervisory control room or a remote cloud function (30).
12. A computer program (18, 33), comprising computer executable instructions, which, when executed on a processing unit (17, 32) included in the device (20, 35), are used to cause the device (20, 35) to perform the steps described in any one of claims 1 to 11.
13. A computer program product comprising a computer readable medium (19, 34) having embodied thereon a computer program (18, 33) according to claim 12.
14. A device (20, 35) configured to be able to determine a tightening category of a tightening operation performed by a tightening tool (10), the device (20, 35) comprising a processing unit (17, 32) operative to cause the device (20, 35): Acquiring (S201) a set of measured torque and angle values of a fastener (25) tightened by a tightening tool (10); Identifying (S202) a screw-in phase and an end-tightening phase of tightening a fastener (25) from a set of torque and angle values measured and acquired; The torque value at the end of the tightening stage is normalized using the determined torque value range at the end of the tightening stage (S203), and the angle value at the end of the tightening stage is normalized using the determined angle value range at the end of the tightening stage (S203); as well as A machine learning model is trained (S204) using the normalized end-tightening phase torque and angle values and at least one tightening category associated with the normalized end-tightening phase torque and angle values, wherein the tightening category identifies the type of tightening operation applied to the fastener (25).
15. The device (20, 35) configured to determine a tightening category of a tightening operation performed by a tightening tool (10) according to claim 14, further being operable to: A further acquired and normalized set of measured torque and angle values of the fastener tightened by the tightening tool (10) at the end of the tightening phase is provided (S205) to the trained machine learning model, wherein: The trained machine learning model outputs at least one estimated tightening category for a further provided normalized set of measured end-of-tightening phase torque and angle values.
