Contact establishment overload suppression method based on time convolutional network and force-position hybrid switching control

Abstract

The invention provides a contact establishment overload suppression method based on a time convolutional network and force-position hybrid switching control. The method comprises the following steps: synchronously acquiring displacement, speed and driving side load in an idle stroke approaching stage, performing baseline correction, and outputting'to-be-contacted 'and'contact confirmation' indication sequences by adopting a causal expansion time convolutional network; a switching preparation section and an amplitude limiting position channel are set according to a light touch approaching speed and a controlled freezing force channel cumulant, a driving side load is converted into a candidate pipe connection reference force according to calibration, and a pipe connection reference force is formed under the constraint of an overload upper limit; a force control takeover permission is generated in combination with a load increase criterion, contact is generated according to speed limit and amplitude limit to establish transition force, and a force channel tracks target loading force in a closed loop mode; the scheme does not depend on fixed position or load single-point triggering, contact impact and instantaneous overload can be inhibited, and the smoothness of contact establishment and the stability of loading force control are improved.

Description

Technical Field

[0001] This invention relates to the field of force control technology for contact loading, and in particular to a contact overload suppression method based on temporal convolutional networks and force-position hybrid switching control. Background Technology

[0002] In assembly pressing, component contact loading, and clamping processes, the actuator typically first completes the idle approach via the position channel, and then switches to the force channel closed loop to track the target loading force after contact is established. The contact moment at the end of the approach is uncertain due to the influence of part tolerances, assembly posture, friction, and inherent bias on the drive side. The drive side load is often indirectly characterized by current, torque, or hydraulic pressure. If the sudden load change and impact at the moment of contact are not properly controlled, it can easily lead to overload and vibration, affecting workpiece quality and equipment safety. Therefore, contact recognition and smooth switching during the approach phase are important background issues in this field.

[0003] Existing technologies mainly employ threshold triggering and rule-based switching: one type triggers a switch from position control to force control when the displacement reaches a preset position or stroke margin; another type considers contact as the drive-side load (converted from current, torque, or hydraulic pressure) reaching a fixed threshold; there are also contact determinations based on fixed time windows or simplified filtering, followed by direct integration of the capacity channel loop. To reduce impact, measures such as approaching the speed limit, limiting the rate of change of output, and overload limiting are often used, followed by PI closed-loop control based on the target loading force after the switch is connected; there are also impedance / admittance control or soft switching strategies, which achieve the transition from speed to force in the contact range by adjusting stiffness and damping.

[0004] However, single-point triggering relying on fixed positions or load thresholds is insufficient to handle uncertain contact times and offset variations, easily leading to premature or delayed switching and causing shocks. The lack of a force reference consistent with the overload limit at the moment of takeover, coupled with a mismatch between the switching start point and the limit, can result in instantaneous overload. Furthermore, the absence of a unified reference locked at the switching edge to determine load growth makes takeover permission susceptible to noise and short-term fluctuations, resulting in unstable timing and uneven contact establishment.

[0005] Therefore, a contact-based overload suppression method that can overcome the shortcomings of the prior art is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a contact establishment overload suppression method based on temporal convolutional networks and force-position hybrid switching control. The core technical problem to be solved by this application is: in assembly pressing scenarios where the contact establishment time is uncertain, based on the real-time timing of displacement, velocity and drive-side load characterization, to achieve smooth connection from position control to force control, while effectively suppressing instantaneous contact overload during the switching and connection process.

[0007] The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to embodiments of the present invention includes: S1. During the assembly pressing, component contact loading, or clamping and holding process, obtain the displacement, velocity, and drive-side load characterization quantities of the loading mechanism during the empty stroke approach stage, and arrange them according to the control cycle to form the empty stroke approach time sequence quantity; S2. Input the empty travel approach time series into the temporal convolutional network. The temporal convolutional network includes multi-level causal temporal convolutional layers connected in a temporal causal order. Dilated convolutions are set in at least two levels of causal temporal convolutional layers to cover the time span from the end of the empty travel to just before contact. The temporal convolutional network calculates the empty travel approach time series to obtain the contact indication sequence, and further calculates the contact confirmation indication sequence based on the contact indication sequence. S3. Determine the switching preparation segment flag based on the upcoming contact indication sequence and the preset upcoming contact threshold. Under the constraint of the switching preparation segment flag, adjust the position channel speed command to the light touch proximity speed command, and update the cumulative amount of the force control channel to the force control takeover starting reference force. Generate force control allowable conditions based on the contact confirmation indication sequence and the preset contact confirmation threshold, and limit the force control takeover starting reference force under the preset overload upper limit constraint to form a takeover reference force. S4. Calculate the load increase based on the load characterization on the drive side. When the force control allowable conditions are met and the load increase meets the preset load increase threshold, generate a force control takeover permit. If not, generate a continue approach permit. S5. Generate a contact establishment transition force command based on the force control takeover permission, the takeover reference force, and the light touch approach speed command, and form a force control takeover control command based on the contact establishment transition force command. S6. Generate target loading control commands based on the force control pipe control commands and the target loading force, which are used to drive the loading mechanism to complete pressing or clamping and holding.

[0008] Optionally, S1 is as follows: When the loading mechanism performs the no-stroke approach motion, the displacement and velocity are synchronously collected in the control cycle. The displacement is taken from the position sensor, and the velocity is calculated by displacement differential or obtained from the velocity sensor. The drive-side load characteristics are collected in the same control cycle. The drive-side load characteristics are taken from one of the drive current, drive torque or hydraulic pressure, and converted into a uniform load characteristics according to a preset ratio. The baseline correction load characterization is obtained by biasing the drive-side load characterization with the preset idle travel baseline load, and the displacement, velocity, and baseline correction load characterization are combined into a three-dimensional sampling vector in a time-causal order. A preset window length is used to slide and frame the three-dimensional sampling vector of the continuous control cycle to form an empty travel approach timing quantity. The preset window length covers the critical time period from the end of the empty travel to just before contact.

[0009] Optionally, S2 is as follows: The idle travel time sequence is used as the input sequence of the temporal convolutional network. The input sequence contains three feature components in the order of control cycle: displacement, velocity, and driving side load characterization. The input sequence is convolved step by step by a multi-level causal temporal convolutional layer connected in a time-causal order. The output of each level of causal temporal convolutional layer is determined only by the displacement, velocity, and driving-side load characterization of the current control cycle and the historical control cycle. In the step-by-step convolution calculation process, the contact trigger point is not generated based on the displacement reaching a fixed position or the driving-side load characterization reaching a fixed threshold. Dilated convolutions are set in at least two levels of causal temporal convolutional layers in the multi-level causal temporal convolutional layers. The sampling interval of the dilated convolutions covers the time span from the end of the empty journey to the moment of contact, forming the backbone features of the changes in different historical spans. The computation order of ordinary causal temporal convolution and dilated convolution is configured in a hierarchical manner, so that the backbone features of the earlier layers correspond to the trend changes at the end of the empty journey, and the backbone features of the later layers correspond to the changes just before contact. The backbone features obtained by dilated convolution covering a long time span are input into the first output layer, the contact indication values ​​for each control cycle are calculated, and they are arranged according to the control cycle to form a contact indication sequence. The main features near the end layer and the contact indication sequence are used as inputs to the second output layer to calculate the contact confirmation indication value for each control cycle, and then arranged according to the control cycle to form a contact confirmation indication sequence.

[0010] Optionally, the controller performs constraint calculations on the historical coverage length of the temporal convolutional network based on a preset window length, and uses the historical coverage length as a direct basis for selecting the dilated convolution parameters, so that the dilated convolution covers the time span from the end of the empty stroke to the moment of contact. The historical coverage length is determined by a coverage function, wherein the coverage function is specifically: ; in, For temporal convolutional networks The number of historical control cycles covered in a single output control cycle. The number of layers in a multi-level causal temporal convolutional layer. This refers to the layer number of the causal temporal convolutional layer. For the first Kernel length of a first-order causal temporal convolutional layer For the first The dilation coefficient of a causal temporal convolutional layer. The sequence number is the hierarchical number of the causal temporal convolutional layer, determined by the controller based on a preset window length. Select All levels With all levels This extends the historical coverage length. Covering the preset window length This ensures that the calculation of the indicator value to be touched can utilize the trend information from the end of the empty journey to the point of contact, while maintaining causal constraints and not introducing future sampling points.

[0011] Optionally, S3 specifically refers to: Real-time threshold determination is performed on the upcoming contact indication sequence. When the upcoming contact indication value for a consecutive preset number of cycles meets the upcoming contact threshold, the switching preparation segment flag is set. When the upcoming contact indication value does not meet the upcoming contact threshold, the switching preparation segment flag is reset. Under the constraint of setting the switching preparation segment flag, the position channel speed command is adjusted from the empty travel approach speed command to the light touch approach speed command. The light touch approach speed command is obtained by limiting the position channel speed command by a preset light touch speed upper limit. Under the constraint of setting the switching preparation segment flag, the cumulative amount of the force control channel is updated and limited to the starting reference force of the force control maneuver, the current value of the cumulative amount of the force control channel is reset to the starting reference force of the force control maneuver, and the error based on the target loading force and feedback load is prohibited from continuing to accumulate. Under the constraint of setting the switching preparation segment flag, the contact confirmation indication sequence is judged in real time. When the contact confirmation indication value of a consecutive preset number of cycles meets the contact confirmation threshold, the force control allow condition is generated. When the contact confirmation indication value does not meet the contact confirmation threshold, the force control allow condition is canceled. When the force control allowable conditions are met, the candidate control reference force is generated based on the drive-side load characterization quantity of the current control cycle according to the preset force calibration coefficient; Under the preset overload upper limit constraint, the reference force of the candidate nozzle is limited, and the limiting result is written into the force control nozzle starting reference force to form the nozzleable reference force; The output touch proximity speed command, force control allowance conditions, and manageable reference force are used as inputs for subsequent generation of contact establishment transition force commands.

[0012] Optionally, a nozzle control function is used to limit the candidate nozzle reference force under a preset overload upper limit constraint, and the limiting result is written into the force-controlled nozzle starting reference force to form a remandable reference force. The nozzle control function is specifically as follows: ; in, For the reference force that can be taken over, The preset overload limit, The reference force for the start of the force control nozzle in the previous control cycle. To confirm the continuous counter, To preset the number of duration periods, As a benchmark force for candidate takeover, To perform the calculation that yields the smaller value, the controller will Write the starting reference force of the force control unit This ensures that the starting point of the contact establishment transition force command is consistent with the overload upper limit constraint.

[0013] Optionally, S4 specifically refers to: In each control cycle, the current drive-side load characterization value is read, and in the control cycle when the switching preparation segment flag changes from reset to set, the drive-side load characterization value is locked as the load growth reference value. The load increase is calculated based on the current driving-side load characterization and the load increase benchmark. When the force control allowable conditions are met, a threshold judgment is made on the load increase amount. When the load increase amount for a consecutive preset number of durations meets the load increase threshold, a force control takeover permit is generated. When the force control permission condition is not met or the load increase amount does not meet the load increase threshold, a continued approach permission is generated. The force control takeover permission and the continued approach permission are mutually exclusive output within the same control cycle.

[0014] Optional, S5 specifically includes: The force control takeover permission is determined. When the force control takeover permission is established, the takeover reference force is set as the starting force value of the contact establishment transition force command, and the rise rate of the contact establishment transition force command is determined according to the light touch approach speed command. During subsequent control cycles, the contact establishment transition force command is incrementally updated at the aforementioned rate of increase, and a preset rate of change limit and overload upper limit limit are applied to the contact establishment transition force command. When force control permission is granted, a contact establishment transition force command is used as the force target of the force control channel and the capability control channel outputs, forming a force control control command. When force control permission is not granted, the force control channel output is prohibited and the position channel maintains a light touch approach speed command as the control output.

[0015] Optionally, step S6 specifically includes: In each control cycle, it receives force control takeover control commands and, based on these commands, causes the force control channel to enter a closed-loop output state. In closed-loop output mode, the load characterization value on the drive side is read and converted according to the preset force calibration coefficient to obtain the current loading force; The loading force error is generated based on the target loading force and the current loading force, and the output of the force control channel is calculated under the condition that the cumulative amount of the force control channel is updated and the control command of the force control tube is constrained. The overload upper limit and rate of change limit are applied to the output of the force control channel to obtain the target loading control command and output it to the loading mechanism to complete the pressing or clamping retention.

[0016] The beneficial effects of this invention are: (1) This proposal proposes an improved contact establishment judgment and force-position hybrid switching method. It adopts a causal connection time convolutional network and sets dilated convolution in multiple levels. Based on the preset window and coverage function to constrain the historical coverage length, it inputs the displacement, velocity and baseline correction drive-side load timing sequence, and outputs the contact imminent and contact confirmation indication sequences respectively. The switching preparation segment flag and force control allowable conditions are generated by threshold and duration period count, which replaces fixed position or single-point load triggering, so that the contact identification and switching timing remain stable under the condition of uncertain contact time, and reduce misjudgment and timing deviation.

[0017] (2) This proposal puts forward a novel technique for forming the reference force and generating the transition force. After the switching preparation section is set, the accumulated amount of the force channel is frozen in a controlled manner, and the driving side load is converted into a candidate reference force according to the force calibration coefficient. The reference force is then written into the controllable reference force through the control function under the overload upper limit constraint. The rise rate of the transition force is determined based on the light touch approach speed. The change rate limit and overload limit are applied to the contact-established transition force, so that the starting point of the control is consistent with the limit and the slope is controlled, effectively suppressing the sudden force change and overload at the moment of contact, while maintaining the controllability of the target loading force closed loop.

[0018] (3) This proposal presents an overall method for load growth permission and force-position channel mutual exclusion control. During the switching preparation phase, the load growth reference amount is locked from reset to set. Force control takeover permission is generated based on the load growth amount and the duration period threshold, and is mutually exclusive with the continue approach permission output. After the force channel is enabled, a constrained proportional integral closed loop is applied to the output, and the rate of change and overload limit are applied. At the same time, the load characterization quantities of electric drive and hydraulic system are mapped to be consistent through a unified force calibration coefficient. The overall strategy maintains consistency of takeover permission criteria in the presence of noise and bias, and force-position mutual exclusion avoids channel conflicts, improves the smoothness of the contact establishment phase and equipment safety, which is in line with the technical goal of this application to achieve smooth takeover and suppress instantaneous overload in scenarios where the contact establishment time is uncertain. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1The flowchart shows a contact overload suppression method based on temporal convolutional network and force-position hybrid switching control proposed in this invention. Figure 2 The flowchart shows the empty stroke approach time series formation process of a contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control proposed in this invention. Figure 3 The flowchart shows the time convolutional network calculation process for the contact overload suppression method based on time convolutional network and force-position hybrid switching control proposed in this invention. Detailed Implementation

[0020] In Example 1, reference Figures 1 to 3 A contact overload suppression method based on temporal convolutional networks and force-position hybrid switching control includes: S1. During the assembly pressing, component contact loading, or clamping and holding process, obtain the displacement, velocity, and drive-side load characterization quantities of the loading mechanism during the empty stroke approach stage, and arrange them according to the control cycle to form the empty stroke approach time sequence quantity; S2. Input the empty travel approach time series into the temporal convolutional network. The temporal convolutional network includes multi-level causal temporal convolutional layers connected in a temporal causal order. Dilated convolutions are set in at least two levels of causal temporal convolutional layers to cover the time span from the end of the empty travel to just before contact. The temporal convolutional network calculates the empty travel approach time series to obtain the contact indication sequence, and further calculates the contact confirmation indication sequence based on the contact indication sequence. S3. Determine the switching preparation segment flag based on the upcoming contact indication sequence and the preset upcoming contact threshold. Under the constraint of the switching preparation segment flag, adjust the position channel speed command to the light touch proximity speed command, and update the cumulative amount of the force control channel to the force control takeover starting reference force. Generate force control allowable conditions based on the contact confirmation indication sequence and the preset contact confirmation threshold, and limit the force control takeover starting reference force under the preset overload upper limit constraint to form a takeover reference force. S4. Calculate the load increase based on the load characterization on the drive side. When the force control allowable conditions are met and the load increase meets the preset load increase threshold, generate a force control takeover permit. If not, generate a continue approach permit. S5. Generate a contact establishment transition force command based on the force control takeover permission, the takeover reference force, and the light touch approach speed command, and form a force control takeover control command based on the contact establishment transition force command. S6. Generate target loading control commands based on the force control pipe control commands and the target loading force, which are used to drive the loading mechanism to complete pressing or clamping and holding.

[0021] In this embodiment, step S1 specifically includes: In assembly pressing, component contact loading, or clamping and holding processes, the loading mechanism is in a position channel-dominated approach motion state during the no-travel approach phase. In this embodiment, a fixed control cycle is set in the controller. and by control cycle number The current sampling time is identified, and a synchronous sampling is triggered at the beginning of each control cycle. This ensures that the empty travel received by the subsequent temporal convolutional network is arranged in a temporal causal order, thus avoiding the introduction of future sampling points into the contact trend judgment of the current control cycle. The controller in the control cycle Displacement is obtained by reading the output of the position sensor. The position sensor uses an encoder, optical scale, or magnetic scale, and the speed... One of two acquisition methods is used: the first method is based on displacement. displacement from the previous control cycle The difference is calculated and then divided by the control cycle. Get speed The second method is to directly read the speed sensor output to obtain the speed. When the first method is selected, the controller retains the displacement of the previous control cycle in each control cycle. As a differential input, speed is guaranteed. Determined solely by current and historical information; Controller in the same control cycle Acquire drive-side load characterization When the loading mechanism is an electrically driven structure, The load is taken from the driver current or driver torque when the loading mechanism is a hydraulic drive structure. Taken from hydraulic pressure, the controller will Multiply by a preset scaling factor to obtain a unified driving-side load characterization quantity. The preset scaling factor is used to map the measurements of different driving sources to the same load characterization space, so that the temporal convolutional network maintains a consistent physical meaning under three-dimensional input features. The controller selects a preset baseline duration within the initial interval of the no-travel approach phase and characterizes the drive-side load within this interval. The arithmetic mean is taken to obtain the preset idle travel baseline load. This is used to characterize the inherent bias and frictional preload on the drive side under non-contact conditions, and is a characteristic quantity of the drive-side load for the current control cycle by the controller. After biasing, the baseline correction load characterization is obtained. ,in Depend on deduct Once obtained, the controller will then displace the target. ,speed Baseline correction load characterization The control cycle sequence is used to combine the three-dimensional sampling vectors. ,in The three components correspond to displacement, velocity, and baseline correction load characterization quantities, respectively, and serve as the basic input units of the temporal convolutional network. Controller settings preset window length And in each control cycle, the current three-dimensional sampling vector Write to the sliding buffer to form a time-ordered sequence. This sequence constitutes an empty run close to the time series. Preset window length The selection makes the idle stroke close to the timing quantity. It covers the critical time period from the end of the idle journey to the moment of contact, enabling the multi-level causal temporal convolutional layers of the temporal convolutional network to be based solely on the current control cycle and historical control cycles. , , The system performs stepwise calculations and enables dilated convolutional layers to cover a longer historical span to extract trend changes that will occur during contact, thereby providing a continuous and real-time updated temporal input for subsequent outputs of contact indication sequences and contact confirmation indication sequences.

[0022] In this embodiment, step S2 specifically includes: During the approach phase of the no-travel period, the controller operates within the control cycle. Obtaining the time series of the empty journey ,in To control the cycle number, From the recent The three-dimensional sampling vectors of each control cycle are composed in chronological order, and the three-dimensional sampling vectors are denoted as follows: ,in To control the cycle number, Displacement ,speed Baseline correction load characterization constitute, The value of the drive-side load characterization quantity is corrected for the preset no-travel baseline load. The input sequence is used as the input sequence for the temporal convolutional network, so that the network receives sequences arranged in a temporally causal order at each control cycle. , , and with the control period corresponding to the end of the sequence. As the current moment of reasoning; Temporal convolutional networks are denoted as , It contains multi-level causal temporal convolutional layers connected in a temporal causal order, with the layer numbers as follows: to ,in The number of causal temporal convolutional layers is denoted as . , No. The first-order causal temporal convolutional layer uses a one-dimensional convolutional kernel group, and the number of channels in the convolutional kernel group is denoted as . The kernel length is denoted as The controller will enter The convolution calculation is performed hierarchically: each causal temporal convolutional layer takes only sampling points from the current control cycle and its historical control cycles in the temporal direction. The sampling points are weighted and summed according to the convolution kernel weights, and then a bias is added. A nonlinear transformation is then performed to obtain the backbone feature output of this layer. The backbone feature output consists only of data from control cycles no later than the current control cycle. , , It was decided that, during the progressive convolution calculation process, no triggering logic based on the displacement reaching a fixed position or the driving-side load characterization reaching a fixed threshold would be set, so that the determination of contact establishment would be provided by the timing indicators continuously output by the network. To cover the critical time span from the end of the empty journey to the point of contact, an expansion coefficient is configured for each causal temporal convolutional layer. And set the layer that uses ordinary causal temporal convolution to Pick Set the layer that uses dilated convolution to Take greater than The integers enable at least two levels of causal temporal convolutional layers in a multi-level causal temporal convolutional layer to use dilated convolution. Dilated convolution introduces a fixed time interval at the sampling position of the convolution kernel, enabling the backbone features of a single control cycle to aggregate input changes over a longer historical span, forming backbone features for changes over different historical spans. The calculation order of ordinary causal temporal convolution and dilated convolution is configured in a hierarchical manner, so that the backbone features of the earlier layers correspond to the trend changes at the end of the empty journey, and the backbone features of the later layers correspond to the changes just before contact, thereby obtaining long-span trend representations and short-span change representations simultaneously within the same network backbone. During the above hierarchical configuration process, the controller determines the window length according to the preset window length. The historical coverage length of the temporal convolutional network is constrained and calculated, and this historical coverage length is used as the direct basis for selecting the dilated convolution parameters, ensuring that the dilated convolution covers the time span from the end of the empty travel to the point of contact. Determined by the coverage function: ; in, For temporal convolutional networks The number of historical control cycles covered in a single output control cycle. The number of layers in a multi-level causal temporal convolutional layer. This refers to the layer number of the causal temporal convolutional layer. For the first Kernel length of a first-order causal temporal convolutional layer For the first The dilation coefficient of a causal temporal convolutional layer. The sequence number is the hierarchical number of the causal temporal convolutional layer, determined by the controller based on a preset window length. Select All levels With all levels This extends the historical coverage length. Covering the preset window length This ensures that the calculation of the indicator value to be contacted can utilize the trend information from the end of the empty journey to the point of contact, while maintaining causal constraints and not introducing future sampling points; The backbone features are selected from layers containing dilated convolutions and covering a long time span as the input to the first output layer. These backbone features are then used within a controlled period. The corresponding channel vector is denoted as The first output layer uses fully connected weights and bias pairs. Channel compression and nonlinear mapping are performed to obtain the contact indication value. and the continuous control cycle obtained Arranged chronologically to form a sequence of indicators to be accessed. The instruction sequence will be accessed. To characterize the change in proximity during the transition from the no-travel proximity phase to the contact establishment phase, the controller updates the latest [data / feedback] in each control cycle. The decision link for outputting the subsequent switch preparation segment flag; From the main features near the terminal layer, the main features that are more sensitive to changes immediately before contact are selected as the input to the second output layer. The terminal main features are then used in the control cycle. The corresponding channel vector is denoted as and the upcoming contact indication value of the current control cycle. As auxiliary input and The input vectors are concatenated to form the input vector of the second output layer. The second output layer maps this input vector using fully connected weights and biases and performs a nonlinear transformation to obtain the contact confirmation indicator value. and the continuous control cycle obtained Arranged chronologically to form a contact confirmation instruction sequence Contact confirmation instruction sequence Used to characterize changes in established contact trends, the controller updates the latest data in each control cycle. The output is sent to the generation link of the subsequent force control allowable conditions, so that the switching preparation segment flag and the force control allowable conditions are driven by two timing outputs respectively, which can adapt to the working conditions where the contact establishment time is not fixed.

[0023] In this embodiment, step S3 specifically includes: In the control cycle sequence number is During online control, the controller obtains two timing outputs from the temporal convolutional network, which are the contact indication sequences. Contact confirmation instruction sequence And take the contact indication value of the current control cycle. Contact confirmation indicator value As the input for the switching determination in this control cycle, the preset contact threshold is denoted as... The preset contact confirmation threshold is denoted as The preset duration period is denoted as follows: and The switch preparation segment flag is denoted as Force control allowable conditions are denoted as The current position channel speed command is recorded as ,in During the no-travel approach phase, corresponding to the no-travel approach speed command, the drive-side load characterization value for the current control cycle is denoted as: The preset force calibration coefficient is denoted as This is used to convert the driving-side load characteristic into a feedback load, and the preset overload upper limit is denoted as... The target loading force is denoted as , The process parameters are given in advance and written into the controller before the loading cycle begins; controller to Perform real-time threshold determination and set the continuous counter to be triggered. The continuous counter will be contacted in the previous control cycle. ,when Meet the threshold for upcoming contact At that time, the controller is Add one to the base to get ,when The threshold for contact is not met. At that time, the controller will Clear to zero and switch to the preparation segment flag. After resetting, the controller then determines... Has the preset duration been reached? ,when achieve Time-set switching preparation segment flag And in subsequent control cycles only when Dissatisfied again Reset only when Enable the switch preparation segment flag From the sequence of instructions to be contacted The continuous trend drive avoids the use of single-point triggering logic that requires displacement to reach a fixed position or the load characterization quantity on the drive side to reach a fixed threshold. The controller is switching the preparation phase flag. Update position channel speed command under position constraint, preset tap speed upper limit is denoted as ,when When set, the controller remains The sign remains unchanged, for The amplitude is limited so that it does not exceed [the specified value]. Receive a light touch approach speed command and with As the current control cycle speed output command of the position channel, when During reset, the controller outputs To maintain the original control strategy for the no-travel approach speed command; The controller is switching the preparation phase flag. The cumulative amount of the constraint force control channel is updated under the set constraint. The cumulative amount of the force control channel is denoted as... The initial reference force for force control is recorded as... The starting reference force of the force control unit in the previous control cycle is recorded as... The feedback load is denoted as ,in Characterized by driving-side load Multiply by the preset force calibration coefficient The conversion yields that when When the controller changes from reset to set, it will The current value is reset to and in Freeze during the holding period The error accumulation update makes No longer based on target loading force With feedback load The error between them is cumulatively expanded when Upon reset, the controller is unfrozen and allows the force control channel to update according to the normal closed-loop structure. ; The controller is switching the preparation phase flag. under set constraint Perform real-time threshold determination and set a contact confirmation duration counter. Contact confirmation of the continuous counter from the previous control cycle ,when Set and Meets the contact confirmation threshold At that time, the controller is Add one to the base to get ,when Reset or The contact confirmation threshold is not met. At that time, the controller will Clear and cancel force control allow conditions Subsequently, the controller determines Has the preset duration been reached? ,when achieve Time-based force control allowable conditions And in subsequent control cycles only when Dissatisfied again or Cancel only during reset Force control allows conditions Strictly dependent on the switching preparation segment flag Contact confirmation instruction sequence The trend of continuous confirmation; Controller under force control allowable conditions Within the established control cycle, a candidate takeover reference force is generated, and the candidate takeover reference force is denoted as... The controller's representation of the drive-side load in the current control cycle. According to the preset force calibration coefficient After conversion, we can obtain and make The direction is consistent with the loading direction, thus directly mapping the instantaneous load state after contact confirmation to a candidate input for the takeover reference; The controller is at the preset overload limit. Constraint on candidate takeover benchmark force Perform a limit update and establish a takeover reference force, denoted as . and confirm the continuous counter by contact The update amplitude of the nozzle reference force is modulated so that the update is more complete when the contact confirmation trend is more stable. Specifically, it is calculated according to the nozzle function: ; in, For the reference force that can be taken over, The preset overload limit, The reference force for the start of the force control nozzle in the previous control cycle. To confirm the continuous counter, To preset the number of duration periods, As a benchmark force for candidate takeover, To perform the calculation that yields the smaller value, the controller will Write the starting reference force of the force control unit This ensures that the starting point of the contact establishment transition force command is consistent with the overload upper limit constraint; The controller outputs a touch proximity speed command. Force control allowable conditions With the reference force that can be taken over and will The position channel is used for the touch proximity phase execution. and As the input for the subsequent generation of contact establishment transition force command, the "switching preparation segment flag determination - light touch approach speed command generation - force control allowable condition generation - takeover reference force formation" are completed in a fixed order within the same control cycle sequence.

[0024] In this embodiment, step S4 specifically includes: In the control cycle sequence number is During online control, the controller switches to the preparation phase flag. Force control allowable conditions Under the constraints of force control takeover permission and continued approach permission, the driving-side load characterization quantity of the current control cycle is denoted as... The value of the previous control cycle of the switch preparation segment flag is recorded as... The load growth benchmark is denoted as The increase in load is denoted as The preset load growth threshold is denoted as The preset duration period is denoted as The load growth duration counter is denoted as The load growth duration counter for the previous control cycle is denoted as Force control takeover permit recorded as Continuing to approach the license is recorded as ; The controller reads the current drive-side load characterization value in each control cycle. The controller performs edge detection on the switching preparation segment flag, and when the condition is met... To reset and When set, the drive-side load characterization value of the current control cycle is set. Locked as the baseline for load growth and load growth duration counter Reset to zero, when When resetting, the controller holds Zero and keep To cancel the state, increase the load by the reference amount. The switch preparation segment flag is relocked only during the control cycle in which the switch preparation segment flag changes from reset to set. The controller characterizes the current drive-side load. With load growth reference amount Calculate the load increase The calculation method is based on minus The difference is obtained and used as the load increase for the current control cycle. The controller will As the input for subsequent threshold determination, the load increase is continuously updated under the same reference base after the switching preparation segment flag is set; Controller under force control allowable conditions When the load increase is satisfied Perform real-time threshold determination when Satisfy and Meets the load growth threshold At that time, the controller continues to use the load growth counter in the previous control cycle. Add one to get ,when Satisfied but The load growth threshold is not met. At that time, the controller will After resetting to zero, the controller then makes a judgment. Has the preset duration been reached? ,when achieve Time-based force control takeover license And cancel the continued approach permission within the same control cycle. ; When force control allows conditions Not met, or load increase The load growth threshold is not met. Or load growth duration counter The preset duration period was not reached. At that time, the controller generates a continued approach to the permission. And cancel the Force Control takeover permit. The controller outputs according to a mutual exclusion rule in each control cycle. and The mutual exclusion rule is When it was established Cancel, When it was established Cancel.

[0025] In this embodiment, step S5 specifically includes: In the control cycle sequence number is During the online control process, the controller receives the touch proximity speed command output in step S3. Force control allowable conditions With the reference force that can be taken over and receive the force control takeover permission output in step S4. With continued approach to permission The controller is set to a fixed control cycle. Set contact to establish transition force command As the force target input for the force control channel, a preset rate of change limit is set. Overload limit It also sets a preset rise rate coefficient for determining the rise rate based on the touch proximity speed command. With preset upper limit of ascent rate The contact establishment transition force command of the previous control cycle is denoted as The feedback load of the force control channel is denoted as ,in The driving-side load characterization value of the current control cycle is determined by a preset force calibration coefficient. The conversion yields, Used to map the driving-side load characterization to load values; Controller to force control takeover permission Make a judgment when When a control cycle changes from cancellation to establishment, the controller will change the current manageable reference force. Set as contact to establish transition force command The initial force value is determined, and the ascent rate is initialized, denoted as . The controller receives a light touch proximity speed command. The amplitude is used as the speed input, and this amplitude is compared with the preset rate of ascent coefficient. Multiply to obtain candidate ascent rates, and use a preset ascent rate upper limit. Limiting the rise rate of the candidates yields ,when The controller remains valid while subsequent control cycles continue to be valid. Based on the result of the previous control cycle, or based on the current cycle. Repeat the above initialization calculations to update ; exist During the subsequent control cycle after its establishment, the controller operates at an increasing rate. Establish transition force command for contact To perform incremental updates, the controller first calculates the candidate increment for a single period. The calculation method is to... With control cycle Multiply to obtain the increment value, and then apply a preset rate of change limit to the candidate increment. The restriction method is to limit the actual increment in a single period. The amplitude is limited to no more than and keep To ensure non-negativity and maintain incremental updates, then... The transition force command is superimposed on the previous control cycle to establish contact. After the update After completing the incremental update, the controller... Apply overload limit Limiting, when Exceed When Limited to ,when If the condition is not met, the controller will prohibit pressing. right Incremental updates, and maintenance This is the value from the previous control cycle; When Force Control Takeover Permit Upon establishment, the controller uses a contact-based transition force command. As the force target of the force control channel, it triggers the output of the force control channel, forming a force control takeover control command. The controller, in the current control cycle, uses... With feedback load The difference is used as the force control deviation input. The drive execution quantity is calculated according to the control law of the force control channel and output to the driver. Simultaneously, writing to the driver's output via the position channel is prohibited. When force control takeover is permitted... If the condition is not met, the controller disables the force control channel output and sets the force control channel output to zero, so that the position channel maintains the light touch approach speed command. As a control output, continue approaching the permitt in this state. This ensures that the force control takeover command and the position channel speed control output are mutually exclusive within the same control cycle.

[0026] In this embodiment, step S6 specifically includes: In the control cycle sequence number is During the online control process, the controller receives the force control take-off control command output in step S5, and uses this command to control the enable state of the force control channel, recording the force control take-off control command as... ,in It is a binary quantity. Taking the value of the closed-loop output of the capability control channel indicates entering the closed-loop output state. Taking the value of the prohibited force control channel output indicates exiting the closed-loop output state. The closed-loop output state of the force control channel is recorded as follows: and make and To maintain consistency, the target loading force is denoted as The preset force calibration coefficient is denoted as The characteristic quantity of the drive-side load in the current control cycle is denoted as... The fixed control cycle of the controller is denoted as The proportional gain of the force control channel is denoted as The integral gain of the force control channel is denoted as ; The controller reads in each control cycle And update ,when When entering the closed-loop output state, the controller performs enable initialization on the internal variables of the force control channel and records the accumulated value of the force control channel as... and will Set as the cumulative amount of the previous control cycle Alternatively, step S3 can be set to write the initial cumulative value corresponding to the starting reference force of the force control nozzle, so that the closed-loop output state has a definite starting point for the cumulative amount. When exiting the closed-loop output state, the controller remains And set the force control channel output to zero so that the force control channel does not output driving force to the loading mechanism; In closed-loop output state The controller reads the load characterization parameters on the drive side. And according to the preset force calibration coefficient The current loading force is calculated and recorded as follows: The controller will and Multiply to get and will As the direct input value for subsequent error calculation; The controller is based on the target loading force With current loading force Generate the loading force error, and denote the loading force error as... Controller calculation ,exist The output state is closed-loop and the cumulative update is affected. Under constraints, the controller... Perform controlled updates: when When the enable value is obtained, the controller updates it based on the integral. ,when When the value is prohibited, the controller remains Subsequently, the controller will proportional term and The output of the force control channel is obtained by adding them together, and the output of the force control channel is denoted as... and will As the target force-type driving input of the loading mechanism; The controller outputs the force control channel. Apply overload upper limit and rate of change limits to obtain the target loading control command and output it to the loading mechanism. The preset overload upper limit is denoted as... The target loading control command of the previous control cycle is denoted as The preset rate of change limit is denoted as The target loading control command for the current control cycle is denoted as... The controller first Apply an overload upper limit to obtain a midpoint limit value, ensuring that this midpoint limit value does not exceed [the specified value]. And keeping the loading direction unchanged, a rate of change limit is applied to the intermediate value of the amplitude limit, so that... relatively The range of change does not exceed The controller will The output is sent to the loading mechanism to complete the pressing or clamping retention in a closed-loop output state, and in When exiting the closed-loop output state, Set to zero to terminate the force control channel output.

[0027] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A contact overload suppression method based on temporal convolutional networks and force-position hybrid switching control, characterized in that, include: S1. During the assembly pressing, component contact loading, or clamping and holding process, obtain the displacement, velocity, and drive-side load characterization quantities of the loading mechanism during the empty stroke approach stage, and arrange them according to the control cycle to form the empty stroke approach time sequence quantity; S2. Input the empty travel approach time series into the temporal convolutional network. The temporal convolutional network includes multi-level causal temporal convolutional layers connected in a temporal causal order. Dilated convolutions are set in at least two levels of causal temporal convolutional layers to cover the time span from the end of the empty travel to just before contact. The temporal convolutional network calculates the empty travel approach time series to obtain the contact indication sequence, and further calculates the contact confirmation indication sequence based on the contact indication sequence. S3. Determine the switching preparation segment flag based on the upcoming contact indication sequence and the preset upcoming contact threshold. Under the constraint of the switching preparation segment flag, adjust the position channel speed command to the light touch proximity speed command, and update the cumulative amount of the force control channel to the force control takeover starting reference force. Generate force control allowable conditions based on the contact confirmation indication sequence and the preset contact confirmation threshold, and limit the force control takeover starting reference force under the preset overload upper limit constraint to form a takeover reference force. S4. Calculate the load increase based on the load characterization on the drive side. When the force control allowable conditions are met and the load increase meets the preset load increase threshold, generate a force control takeover permit. If not, generate a continue approach permit. S5. Generate a contact establishment transition force command based on the force control takeover permission, the takeover reference force, and the light touch approach speed command, and form a force control takeover control command based on the contact establishment transition force command. S6. Generate target loading control commands based on the force control pipe control commands and the target loading force, which are used to drive the loading mechanism to complete pressing or clamping and holding.

2. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, S1 specifically refers to: When the loading mechanism performs the no-stroke approach motion, the displacement and velocity are synchronously collected in the control cycle. The displacement is taken from the position sensor, and the velocity is calculated by displacement differential or obtained from the velocity sensor. The drive-side load characteristics are collected in the same control cycle. The drive-side load characteristics are taken from one of the drive current, drive torque or hydraulic pressure, and converted into a uniform load characteristics according to a preset ratio. The baseline correction load characterization is obtained by biasing the drive-side load characterization with the preset idle travel baseline load, and the displacement, velocity, and baseline correction load characterization are combined into a three-dimensional sampling vector in a time-causal order. A preset window length is used to slide and frame the three-dimensional sampling vector of the continuous control cycle to form an empty travel approach timing quantity. The preset window length covers the critical time period from the end of the empty travel to just before contact.

3. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, S2 specifically refers to: The idle travel time sequence is used as the input sequence of the temporal convolutional network. The input sequence contains three feature components in the order of control cycle: displacement, velocity, and driving side load characterization. The input sequence is convolved step by step by a multi-level causal temporal convolutional layer connected in a time-causal order. The output of each level of causal temporal convolutional layer is determined only by the displacement, velocity, and driving-side load characterization of the current control cycle and the historical control cycle. In the step-by-step convolution calculation process, the contact trigger point is not generated based on the displacement reaching a fixed position or the driving-side load characterization reaching a fixed threshold. Dilated convolutions are set in at least two levels of causal temporal convolutional layers in the multi-level causal temporal convolutional layers. The sampling interval of the dilated convolutions covers the time span from the end of the empty journey to the moment of contact, forming the backbone features of the changes in different historical spans. The computation order of ordinary causal temporal convolution and dilated convolution is configured in a hierarchical manner, so that the backbone features of the earlier layers correspond to the trend changes at the end of the empty journey, and the backbone features of the later layers correspond to the changes just before contact. The backbone features obtained by dilated convolution covering a long time span are input into the first output layer, the upcoming contact indication values ​​for each control cycle are calculated, and they are arranged according to the control cycle to form an upcoming contact indication sequence. The backbone features near the end layer and the contact indication sequence are used as inputs to the second output layer to calculate the contact confirmation indication value for each control cycle, and then arranged according to the control cycle to form a contact confirmation indication sequence.

4. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 3, characterized in that, The controller performs constraint calculations on the historical coverage length of the temporal convolutional network based on a preset window length, and uses this historical coverage length as the direct basis for selecting the dilated convolution parameters, ensuring that the dilated convolution covers the time span from the end of the empty stroke to the point of contact. The historical coverage length is determined by a coverage function, specifically: ； in, For temporal convolutional networks The number of historical control cycles covered in a single output control cycle. The number of layers in a multi-level causal temporal convolutional layer. This refers to the layer number of the causal temporal convolutional layer. For the first Kernel length of a first-order causal temporal convolutional layer For the first The dilation coefficient of a causal temporal convolutional layer. The sequence number is the hierarchical number of the causal temporal convolutional layer, determined by the controller based on a preset window length. Select All levels With all levels This extends the historical coverage length. Covering the preset window length This ensures that the calculation of the indicator value to be touched can utilize the trend information from the end of the empty journey to the point of contact, while maintaining causal constraints and not introducing future sampling points.

5. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, S3 specifically refers to: Real-time threshold determination is performed on the upcoming contact indication sequence. When the upcoming contact indication value for a consecutive preset number of cycles meets the upcoming contact threshold, the switching preparation segment flag is set. When the upcoming contact indication value does not meet the upcoming contact threshold, the switching preparation segment flag is reset. Under the constraint of setting the switching preparation segment flag, the position channel speed command is adjusted from the empty travel approach speed command to the light touch approach speed command. The light touch approach speed command is obtained by limiting the position channel speed command by a preset light touch speed upper limit. Under the constraint of setting the switching preparation segment flag, the cumulative amount of the force control channel is updated and limited to the starting reference force of the force control maneuver, the current value of the cumulative amount of the force control channel is reset to the starting reference force of the force control maneuver, and the error between the target loading force and the feedback load is prohibited from continuing to accumulate. Under the constraint of setting the switching preparation segment flag, the contact confirmation indication sequence is judged in real time. When the contact confirmation indication value of a continuous preset number of cycles meets the contact confirmation threshold, the force control allow condition is generated. When the contact confirmation indication value does not meet the contact confirmation threshold, the force control allow condition is canceled. When the force control allowable conditions are met, the candidate control reference force is generated based on the drive-side load characterization quantity of the current control cycle according to the preset force calibration coefficient; Under the preset overload upper limit constraint, the reference force of the candidate nozzle is limited, and the limiting result is written into the force control nozzle starting reference force to form the nozzleable reference force; The output light touch approach speed command, force control allowable conditions, and manageable reference force are used as inputs for subsequent generation of contact establishment transition force commands.

6. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 5, characterized in that, The candidate nozzle reference force is limited under a preset overload upper limit constraint using a nozzle control function, and the limiting result is written into the force-controlled nozzle initial reference force to form a remandable reference force. The nozzle control function is as follows: ； in, For the reference force that can be taken over, The preset overload limit, The reference force for the start of the force control nozzle in the previous control cycle. To confirm the continuous counter, To preset the number of duration periods, As a benchmark force for candidate takeover, To perform the operation that yields the smaller value, the controller will Write the starting reference force of the force control unit This ensures that the starting point of the contact establishment transition force command is consistent with the overload upper limit constraint.

7. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, S4 specifically refers to: In each control cycle, the current drive-side load characterization value is read, and in the control cycle in which the switching preparation segment flag changes from reset to set, the drive-side load characterization value is locked as the load growth reference value. The load increase is calculated based on the current driving-side load characterization and the load increase benchmark. When the force control allowable conditions are met, a threshold judgment is made on the load increase amount. When the load increase amount for a consecutive preset number of durations meets the load increase threshold, a force control takeover permit is generated. When the force control permission condition is not met or the load increase amount does not meet the load increase threshold, a continued approach permission is generated. The force control takeover permission and the continued approach permission are mutually exclusive output within the same control cycle.

8. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, S5 specifically refers to: The force control takeover permission is determined. When the force control takeover permission is established, the takeover reference force is set as the starting force value of the contact establishment transition force command, and the rise rate of the contact establishment transition force command is determined according to the light touch approach speed command. During subsequent control cycles, the contact establishment transition force command is incrementally updated at the aforementioned rate of increase, and a preset rate of change limit and overload upper limit limit are applied to the contact establishment transition force command. When force control permission is granted, a contact establishment transition force command is used as the force target of the force control channel and the capability control channel outputs, forming a force control control command. When force control permission is not granted, the force control channel output is prohibited and the position channel maintains a light touch approach speed command as the control output.

9. The contact establishment overload suppression method based on temporal convolutional network and force-position hybrid switching control according to claim 1, characterized in that, Step S6 is as follows: In each control cycle, it receives force control takeover control commands and, based on these commands, puts the force control channel into a closed-loop output state. In closed-loop output mode, the load characterization value on the drive side is read and converted according to the preset force calibration coefficient to obtain the current loading force; The loading force error is generated based on the target loading force and the current loading force, and the output of the force control channel is calculated under the condition that the cumulative amount of the force control channel is updated and the control command of the force control tube is constrained. The overload upper limit and rate of change limit are applied to the output of the force control channel to obtain the target loading control command and output it to the loading mechanism to complete the pressing or clamping retention.

Images