A mechanical arm force-sensitive self-adaptive clamping system and method for fruit bag packaging

Abstract

This invention relates to the field of automated packaging and robot control technology, specifically to a force-sensitive adaptive clamping system and method for fruit bag packaging. The system acquires a preset acceleration timing sequence during the clamping and conveying stage. It obtains the current normal force value and the finger-surface tangential acceleration signal through a clamping surface normal force sensing module and a tangential acceleration sensing module. Simultaneously, it acquires stress wave signals through a disturbance spring and a stress wave sensor. The two signals are compared in the time domain. If the waveform similarity exceeds a threshold, it is determined that the fruit inside the bag is disturbed, and a disturbance suppression flag is generated. If the tangential acceleration signal fluctuates while the stress wave signal is steady, it is determined that the interface is pre-slipping, and a slippage warning flag is generated. Only when a slippage warning flag is received and no disturbance suppression flag is received, the peak value of the tangential inertial force is predicted based on the preset acceleration timing sequence. The clamping safety margin is calculated in conjunction with the current normal force. If it is below the threshold, a clamping force pre-increase command is issued before the peak inertial force is reached.

Description

Technical Field

[0001] This invention relates to the field of automated packaging and robot control technology, and more specifically, to a force-sensitive adaptive clamping system and method for fruit bag packaging. Background Technology

[0002] In automated fruit packaging production lines, when robotic arms grip fruit bags for high-speed transport, tangential force fluctuations frequently occur at the gripping interface due to the rolling or shaking of the fruit inside the bag and the inertial forces generated by the acceleration and deceleration of the robotic arm. Existing robot gripping control largely relies on normal force sensors or simple acceleration threshold judgments, typically focusing only on the instantaneous value of the current gripping force. This makes it difficult to effectively distinguish whether the true source of the tangential force fluctuation is the disturbance of the fruit itself inside the bag or pre-slippage between the gripping fingers and the bag. If the fruit disturbance is misjudged as slippage and the clamping force is increased, it may lead to pressure damage to the fruit. If the true source of the tangential force fluctuation is not properly identified... If pre-slippage is not identified and responded to in a timely manner, the fruit bag may slip during high-speed movement, affecting the production cycle and packaging quality. Therefore, how to design a composite sensing scheme that can simultaneously detect the deformation of the fruit bag sidewall and the tangential acceleration of the clamping fingers, and use motion pre-reading data to predict inertial loads in advance, so as to accurately identify interface pre-slippage and implement pre-clamping force compensation under the premise of eliminating internal disturbances, has become an urgent problem to be solved in the flexible clamping control of robotic arms for fruit bag packaging. To solve this problem, we provide a force-sensitive adaptive clamping system and method for robotic arms for fruit bag packaging. Summary of the Invention

[0003] The purpose of this invention is to provide a force-sensitive adaptive clamping system and method for fruit bag packaging, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, a force-sensitive adaptive clamping system for fruit bag packaging is provided, comprising: The clamping execution unit has at least one clamping finger, and the clamping surface of the clamping finger is provided with a normal force sensing module and a tangential acceleration sensing module; The motion pre-reading unit is used to obtain the preset acceleration timing of the gripping and conveying phase from the robotic arm controller; The bag disturbance identification unit includes a disturbance spring fixed to the side of the clamping finger and a stress wave sensor attached to the root of the disturbance spring; The free end of the disturbance spring is attached to the side wall of the fruit bag in the clamping state, which is used to transmit the deformation of the bag wall caused by the shaking of the fruit inside the bag into a stress wave signal that can be detected by the stress wave sensor. The bag disturbance identification unit compares the stress wave signal with the finger surface tangential acceleration signal detected by the tangential acceleration sensing module in the time domain. If the same frequency fluctuation occurs, it is determined that the current tangential force fluctuation is caused by the disturbance of the fruit inside the bag, and a disturbance suppression mark is generated. If only the finger surface tangential acceleration signal fluctuates while the stress wave signal remains steady, it is determined that the current tangential force fluctuation is caused by the pre-slippage of the interface between the fruit bag and the clamping finger, and a slippage warning mark is generated. The compensation control unit calculates the clamping safety margin based on the peak value of the tangential inertial force predicted by the preset acceleration timing and the current normal force value only when it receives the slip warning flag and does not receive the disturbance suppression flag. When the clamping safety margin is lower than the preset threshold, it issues a clamping force pre-increase command to the clamping execution unit before the peak value of the tangential inertial force is reached.

[0005] The second objective of this invention is to provide a method for implementing a force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging, comprising the steps described above: S1. Obtain the preset acceleration timing for the clamping and conveying stage; S2. In the clamping state, the current normal force value and finger surface tangential acceleration signal are obtained through the normal force sensing module and the tangential acceleration sensing module, respectively, and the stress wave signal is obtained through the disturbance spring and stress wave sensor. S3. After synchronously acquiring and bandpass filtering the stress wave signal and the finger surface tangential acceleration signal, perform time-domain comparison and calculate the waveform similarity value between the two signal waveforms. S4. If the waveform similarity value exceeds the preset feature similarity threshold, it is determined to be a same-frequency fluctuation, and a disturbance suppression mark is generated. S5. If the tangential acceleration signal of the finger surface fluctuates and the waveform similarity value is lower than the feature similarity threshold, while the time domain feature parameters of the stress wave signal remain in a steady state, it is determined to be interface pre-slip and a slip warning mark is generated. S6. When the slip warning flag is received but the disturbance suppression flag is not received, the peak value of the tangential inertial force is predicted according to the preset acceleration timing, and the clamping safety margin is calculated in combination with the current normal force value. S7. When the clamping safety margin is lower than a preset threshold, generate and issue a clamping force pre-increase command before the peak value of the tangential inertial force is reached.

[0006] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention incorporates a normal force sensing module and a tangential acceleration sensing module on the clamping surface of the clamping fingers, along with a disturbance spring and a stress wave sensor on the side of the clamping fingers. This transmits the deformation of the bag wall caused by fruit shaking inside the bag as a stress wave signal, which is then compared in the time domain with the tangential acceleration signal on the finger surface. This solves the problem of traditional clamping control's difficulty in distinguishing between internal fruit disturbances and interface pre-slippage, achieving accurate identification of disturbance sources and avoiding misadjustment of clamping force. Furthermore, the motion pre-reading unit acquires a preset acceleration timing sequence to predict the peak value of the tangential inertial force. Combined with the current normal force, it calculates the clamping safety margin. When only a slippage warning marker is detected and the safety margin is below the threshold, a clamping force pre-increase command is issued before the peak inertial force arrives. This achieves proactive force-sensitive compensation ahead of slippage, effectively preventing the fruit bag from slipping during high-speed transport and avoiding excessive clamping damage to the fruit caused by internal disturbances, thus improving the reliability of the fruit bag packaging robotic arm. Attached Figure Description

[0007] Figure 1 This is an overall block diagram of the present invention; Figure 2 This is the overall flowchart of the present invention. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0009] This invention provides a force-sensitive adaptive clamping system for robotic arms used in fruit bag packaging. Please refer to [link / reference]. Figure 1 As shown, the clamping execution unit has at least one clamping finger, and the clamping surface of the clamping finger is provided with a normal force sensing module and a tangential acceleration sensing module. The motion pre-reading unit is used to obtain the preset acceleration timing of the gripping and conveying phase from the robotic arm controller; The bag disturbance identification unit includes a disturbance spring fixed to the side of the gripping finger and a stress wave sensor attached to the root of the disturbance spring. The free end of the disturbance spring is attached to the side wall of the fruit bag in the clamping state. It is used to transmit the deformation of the bag wall caused by the shaking of the fruit inside the bag into a stress wave signal that can be detected by the stress wave sensor. The bag disturbance identification unit compares the stress wave signal with the finger surface tangential acceleration signal detected by the tangential acceleration sensing module in the time domain. If the same frequency fluctuation occurs, it is determined that the current tangential force fluctuation is caused by the disturbance of the fruit inside the bag, and a disturbance suppression mark is generated. If only the finger surface tangential acceleration signal fluctuates while the stress wave signal remains steady, it is determined that the current tangential force fluctuation is caused by the pre-slippage of the interface between the fruit bag and the clamping finger, and a slippage warning mark is generated. The compensation control unit calculates the clamping safety margin based on the peak value of the tangential inertial force predicted by the preset acceleration timing and the current normal force value when it receives a slip warning flag but does not receive a disturbance suppression flag. When the clamping safety margin is lower than the preset threshold, it issues a clamping force pre-increase command to the clamping execution unit before the peak value of the tangential inertial force is reached.

[0010] The motion pre-reading unit connects to the robotic arm controller based on an industrial-grade real-time communication architecture, and obtains the motion parameters of the robotic arm's gripping and conveying process in advance and calculates the acceleration timing. This method can realize the pre-judgment of motion state, providing accurate motion data support for subsequent dynamic adjustment of gripping force and prevention of interface slippage, and effectively reducing the probability of fruit bags slipping off during the dynamic movement of the robotic arm.

[0011] The motion pre-reading unit and the robotic arm controller first establish a two-way real-time data communication link following the common real-time communication protocol used in industrial settings. This link integrates basic operating mechanisms such as data integrity verification, automatic reconnection after communication interruption, and retransmission of abnormal data packets to ensure stable and reliable data transmission. Before officially starting the fruit bag clamping and conveying task, the robotic arm controller generates a complete motion planning data packet according to a preset data format and actively sends it to the motion pre-reading unit via the communication link. The motion planning data packet contains multiple structured data items. A fixed identifier is set in the header to distinguish the data packet type, and a unique number for the current robotic arm operation is written in it. The main body of the data packet is divided into multiple independent data areas. The first area records segment labels for the entire operation process of the robotic arm. These segment labels are used to divide different work stages such as the robotic arm standby stage, gripping stage, gripping and conveying stage, and fruit bag placement stage. The second area stores the three-dimensional spatial coordinate sequence of the gripping finger end effector along the entire path in the robotic arm's base coordinate system. Each set of coordinate data is bound to a corresponding discrete sampling timestamp. The third area records the linear velocity parameters of the robotic arm end effector and the rotation angle parameters of each joint at each discrete time point. The fourth area labels the attitude angle parameters of the gripping finger during the entire path operation. The fifth area uniformly labels the spatial coordinate system parameters used for the entire set of motion data and the fixed sampling time interval across the entire domain. A cyclic redundancy check code is appended to the end of the data packet for subsequent verification of the data's authenticity and integrity.

[0012] After receiving the motion planning data packet, the motion pre-reading unit first retrieves the cyclic redundancy check code at the end of the data packet using the built-in verification algorithm for comparison. If the comparison results are inconsistent, it is determined that the data transmission is abnormal, and a data packet retransmission request is immediately sent to the robotic arm controller. If the comparison results are consistent, the data packet is determined to be complete and valid, and then the hierarchical parsing process is initiated. The parsing process decomposes each data item according to the preset hierarchy of the data packet. First, it reads the unique task number and the segment label of the entire operation process. Based on the segment label, it filters out the data interval that belongs only to the clamping and conveying stage, and removes all redundant data corresponding to the standby stage, the clamping stage, and the fruit bag placement stage. After completing the data partitioning and filtering, it extracts the planning trajectory corresponding to the clamping and conveying stage separately. The planning trajectory consists of the three-dimensional spatial coordinates of the clamping finger tip arranged in chronological order and the corresponding timestamp. During the parsing process, the robotic arm body base coordinate system is locked as a unified calculation reference to avoid calculation deviations caused by coordinate system switching. The extracted planning trajectory is stored in a temporary calculation cache according to the original time sequence as the basic data source for subsequent acceleration calculation.

[0013] After extracting the planned trajectory, the linear acceleration vector of the gripping finger's end is calculated one by one from the trajectory data at each discrete time point. A three-point difference operation is used to improve the accuracy of the acceleration calculation. First, the global sampling time interval is confirmed to be a fixed value. Then, each discrete time point with an independent timestamp within the planned trajectory is traversed. For a single target discrete time point, the three sets of three-dimensional spatial coordinates corresponding to that time point, the previous adjacent discrete time point, and the next adjacent discrete time point on the time axis are retrieved from the cache. The three-dimensional displacement from the previous discrete time point to the target discrete time point is calculated. The three-dimensional displacement is then decomposed into unidirectional displacement values ​​along the X, Y, and Z axes in the base coordinate system. Each unidirectional displacement value is divided by the fixed sampling time interval to obtain the velocity components of the three axes corresponding to the previous motion period. Finally, the target discrete time is calculated. The three-dimensional displacement of the intermediate point points to the next adjacent discrete time point is calculated using the same operation method to obtain the three-axis velocity components corresponding to the next motion period. Then, the linear velocity of the single axis of the next motion period is subtracted from the linear velocity of the axis corresponding to the previous motion period. The result of the subtraction is then divided by twice the fixed sampling time interval. Twice the sampling time interval represents the total motion duration corresponding to the three sets of coordinates. The acceleration calculation of the three axes X, Y, and Z is completed in sequence. The acceleration components of the three axes are integrated into a whole, which is the three-dimensional linear acceleration vector of the end of the gripping finger at the current discrete time point. All discrete time points within the planned trajectory of the gripping and conveying stage are traversed according to the same three-point difference operation rule. The linear acceleration vector corresponding to all time points is solved one by one. All linear acceleration vectors are associated with corresponding timestamps and temporarily arranged and stored in chronological order.

[0014] After all online acceleration vectors are calculated, vector component decomposition and a preset acceleration time sequence are constructed. Based on the mechanical structure and conventional installation posture of the gripping fingers, a fixed spatial vector direction corresponding to the normal to the gripping surface is calibrated in the robot arm's body coordinate system. This delineates the normal direction and the tangential plane perpendicular to the normal. The tangential plane corresponds to the motion plane where the fruit bag and gripping fingers are prone to relative slippage. For the three-dimensional linear acceleration vector at each discrete time point, spatial vector projection calculations are performed, decomposing the linear acceleration vector into a normal component along the gripping surface normal direction and a tangential component perpendicular to the gripping surface normal direction. Only the tangential component is retained. As valid data, the tangential component values ​​corresponding to all discrete time points are extracted one by one. According to the chronological order of the timestamps of each discrete time point, the tangential component values ​​and their corresponding timestamps are arranged and combined sequentially. The ordered data sequence formed by the continuous arrangement is the preset acceleration timing sequence. The generated preset acceleration timing sequence is stored in the dedicated storage area of ​​the motion pre-reading unit. The subsequent compensation control unit will retrieve the timing data in real time to predict the peak value of the tangential inertial force in the future time window, thereby completing the calculation of the clamping safety margin and triggering the clamping force pre-increase command. This allows the clamping control action of the robotic arm to match the changes in motion state in advance, realizing the pre-control of force-sensitive adaptive clamping.

[0015] The disturbance spring and its matching sensing structure rely on the overall architecture of precise mechanical parameter design, curved surface adaptation processing, graded calibration of pre-pressure, and step-by-step deformation transmission to achieve the complete conversion of the fruit shaking behavior inside the fruit bag into a standardized electrical signal. It can stably extract the characteristic signal corresponding to the disturbance inside the bag, providing real and effective raw data for subsequent time-domain comparison of the two types of signals and determination of the disturbance type.

[0016] The disturbance spring is integrally formed from a homogeneous flexible material with elastic deformation capability. Bending stiffness is an inherent mechanical parameter characterizing the spring's ability to resist bending deformation under external forces. This parameter needs to be preset based on the physical characteristics of the fruit bag and the sensing requirements. First, the maximum deformation displacement value of the sidewall generated by various fruit bags in the working scene during the shaking of the fruit inside is collected. At the same time, the minimum trigger strain value required for the stress wave sensor to normally recognize the signal is measured. Then, the calculation rules of cantilever beam bending deformation in mechanics of materials are used to derive the value. The preset cross-sectional size of the spring, the elastic modulus of the selected material, the maximum deformation, and other parameters are substituted in sequence to calculate the bending stiffness value that can take into account both deformation transmission efficiency and structural stability. This value is set as the preset bending stiffness of the disturbance spring. This stiffness can ensure that the small deformation generated by the sidewall of the fruit bag can be completely transmitted to the root of the spring, while avoiding excessive stiffness of the spring itself to hinder deformation transmission or excessive stiffness to cause irregular autonomous vibration, thus preventing clutter from interfering with the effective signal.

[0017] Integrated machining is performed at the free end of the disturbance spring to form a micro-convex arc surface. The curvature parameters of the micro-convex arc surface are obtained from on-site measurement data. Staff collect the side wall surface coordinates of fruit bags of different specifications in natural and compressed states in batches. The standard curvature of the side wall of the fruit bag is obtained through curve fitting calculation. The micro-convex arc surface is milled according to the standard curvature to make the surface shape of the micro-convex arc surface perfectly match the side wall surface of the fruit bag. This achieves large-scale surface contact rather than local point contact, ensuring uniform and lossless deformation transmission.

[0018] Pre-pressure is the static contact force applied to the sidewall of the fruit bag by the slightly convex arc surface after the clamping action. This parameter is set through three complete processes: basic range calculation, working condition correction, and on-site physical calibration. First, the basic range calculation is carried out by measuring the effective contact area between the slightly convex arc surface and the sidewall of the fruit bag. Then, the compressive yield strength and the upper limit of safe contact pressure corresponding to the fruit bag material are retrieved. The upper limit of safe contact pressure is multiplied by the effective contact area to obtain the maximum allowable value of pre-pressure. At the same time, the minimum contact tightness required for normal acquisition of stress wave signals is combined to calculate the minimum effective value of pre-pressure. The two values ​​constitute the basic value range of pre-pressure. If the pre-pressure is lower than the minimum effective value, it will cause intermittent separation between the arc surface and the bag wall, and the deformation cannot be stably transmitted. If the pre-pressure is higher than the maximum allowable value, it will excessively compress the fruit bag, causing passive deformation distortion of the bag body and thus interfering with the test results. After the basic interval is defined, the working condition correction stage begins. Two types of working conditions are distinguished: single-fruit-loaded bags and multi-fruit-loaded bags. The fruit inside multi-fruit-loaded bags shakes more, so the pre-pressure value is increased within the basic interval. The shaking amplitude of single-fruit-loaded bags is smaller, so the middle value of the basic interval is selected as the corrected value. At the same time, combined with the conventional gripping posture of the robotic arm, the pre-pressure is slightly compensated for the tilted gripping operation to counteract the contact position displacement caused by gravity. Finally, on-site calibration is carried out. With the robotic arm completing the standard gripping action, the current contact force is read in real time using the auxiliary force detection element installed at the root of the disturbance spring. The installation extension length and fixed angle of the disturbance spring are gradually fine-tuned until the detected value is consistent with the corrected target pre-pressure. After calibration, the installation structure of the spring is locked to ensure that the slightly convex arc surface can adhere to the side wall of the fruit bag with a constant pre-pressure in every gripping operation.

[0019] Once the clamping execution unit drives the clamping fingers to complete the fruit bag clamping action and enter a stable clamping state, the micro-convex arc surface maintains continuous surface contact with the side wall of the fruit bag under the action of the preset pre-pressure, forming a stable mechanical transmission channel. During the process of the robotic arm performing the clamping and conveying action, the fruit inside the fruit bag will be affected by factors such as motion acceleration and turning inertia, causing it to sway back and forth. The fruit repeatedly hits and squeezes the inner wall of the fruit bag, which in turn causes the side wall of the fruit bag to produce periodic reciprocating deformation. The reciprocating deformation of the side wall of the fruit bag will directly act on the micro-convex arc surface at the contact position. Relying on the tight surface contact structure, the deformation of the bag wall will be completely transmitted to the entire disturbance spring. The disturbance spring with preset bending stiffness will then produce synchronous bending deformation. The root of the spring is the fixed constraint end, and the mechanical effect brought about by the deformation will be concentrated at this position, causing the root of the disturbance spring to produce periodic alternating strain. The fluctuation frequency and deformation amplitude of the alternating strain are completely synchronized with the reciprocating deformation of the side wall of the fruit bag.

[0020] The stress wave sensor is fixed to the root of the disturbance spring using a high-strength bonding process. The sensor integrates a strain sensing element and a pre-signal processing circuit. The strain sensing element can capture the strain changes at the root of the spring in real time. When the alternating strain continuously acts on the surface of the sensing element, the physical parameters inside the element will change periodically with the strain. The sensor first converts the alternating strain of the mechanical state into a raw analog electrical signal, and then completes signal amplification and impedance matching processing through the internal circuit. Finally, it outputs a continuous stress wave signal. The fluctuation characteristics of this stress wave signal are completely consistent with the deformation characteristics of the fruit bag sidewall, realizing the linear conversion from mechanical deformation to electrical signal. The stress wave signal is transmitted to the bag disturbance identification unit in real time and compared synchronously with the finger-surface tangential acceleration signal output by the tangential acceleration sensing module. This distinguishes between two different working conditions: fruit disturbance inside the bag and pre-slippage of the clamping interface, providing accurate sensing basis for subsequent mark generation and clamping force compensation control.

[0021] As the core device for collecting dynamic motion data of the gripping fingers, the three-axis MEMS accelerometer adopts an embedded installation method to avoid detection errors caused by external collisions and environmental contact. With a complete signal processing flow including orientation calibration, gravity component removal, and multi-level filtering, it can extract a pure finger surface tangential acceleration signal from the raw data mixed with various types of interference. This signal is the core basis for distinguishing between fruit disturbance inside the fruit bag and pre-slippage of the gripping interface. The entire acquisition and processing flow can be adapted to the gripping and conveying conditions of the robotic arm in multiple postures and routes, ensuring the continuity and accuracy of signal detection.

[0022] The triaxial MEMS accelerometer is embedded in a pre-designed, sealed mounting slot within the gripping finger. The mounting slot is machined to match the sensor's dimensions. After insertion, the sensor is secured using structural clips and insulating potting material, limiting displacement or deflection during gripping finger movement and ensuring the stability of the detection reference at the hardware level. Subsequently, the sensor's mounting orientation is calibrated. Calibration is performed with the robotic arm in a horizontal, stationary position without external load. First, the normal direction of the gripping surface is determined. The gripping surface is the working surface where the gripping finger directly contacts the fruit bag; the spatial line perpendicular to this working surface is the normal direction. The first sensitive axis of the triaxial MEMS accelerometer is adjusted to ensure its axis is perfectly parallel to the normal direction of the gripping surface and locked in position. Then, considering the operational scenario, the main motion direction during the gripping and conveying phase is determined, and the second sensitive axis is aligned and installed. The main motion direction refers to the longest displacement distance and the largest proportion of continuous motion at the end of the gripping finger per unit time during the conveying operation from the gripping point to the target placement point after the robotic arm completes the fruit bag gripping action. The high spatial travel direction and main motion direction are uniformly defined by the historical motion planning trajectory stored in the internal memory of the robotic arm controller. All discrete trajectory coordinate points of the clamping and conveying stage under a single conveying route are extracted, and the displacement vector between adjacent coordinate points is calculated in sequence. All displacement vectors within the same route are subjected to vector synthesis operation. The orientation pointed to by the main vector obtained after synthesis is the main motion direction corresponding to that route. If there are multiple fixed conveying routes at the work site, vector synthesis is performed separately for each route and the corresponding main motion direction is calibrated. The orientation parameters of all routes are uniformly stored in the local parameter table. When the robotic arm starts the corresponding route operation, the matching parameters are automatically retrieved. After the main motion direction of the current operation is determined, the second sensitive axis of the three-axis MEMS accelerometer is adjusted so that the axis of the sensitive axis is parallel to the main motion direction. The remaining third sensitive axis of the sensor will naturally form a spatial layout of pairwise orthogonal with the first two sensitive axes. After the orientation calibration is completed, a mapping relationship table between the three sensitive axes and the spatial direction is generated and permanently stored in the local storage unit as the benchmark for subsequent data classification and reading.

[0023] The three-axis MEMS accelerometer operates continuously at a fixed sampling frequency preset by the hardware. During operation, the three sensitive axes will synchronously output three sets of raw acceleration data. Each set of raw data will be equipped with a globally unified timestamp. The three sets of data correspond to the acceleration values ​​in the normal direction of the clamping surface, the main direction of clamping and conveying motion, and the orthogonal auxiliary direction, respectively. The raw acceleration data also includes the dynamic acceleration generated by the movement of the robotic arm, the static gravitational acceleration caused by the Earth's gravity, and high-frequency noise caused by mechanical joint vibration, electromagnetic coupling of circuits, and micro-vibrations of the structure. It cannot be directly used for working condition determination. Therefore, it is necessary to perform the calculation to remove the gravitational acceleration component and the high-frequency noise filtering process in sequence.

[0024] When performing calculations to remove the gravitational acceleration component, the real-time spatial attitude of the gripping finger is first calculated based on the raw acceleration data of the three axes. The horizontal reference attitude from the calibration phase is used as a reference. In this reference attitude, the normal direction of the gripping surface is perpendicular to the horizontal plane, and this axis of the sensor will acquire complete standard gravitational acceleration values. The gravity projection components of the other two sensitive axes are zero. When the robotic arm pitches or tilts during transport, the spatial attitude of the gripping finger changes synchronously, and the gravitational acceleration will form projection components of different magnitudes on the three sensitive axes. The spatial deflection angle is calculated based on the attitude. The standard gravitational acceleration value is multiplied by the cosine of the deflection angle corresponding to each sensitive axis, and the gravitational acceleration projection component corresponding to each sensitive axis at the current moment is calculated one by one. For the sensitive axis used to detect the main motion direction of tangential motion, the original acceleration value of that axis is extracted separately. The original acceleration value at the current moment is subtracted from the gravitational acceleration projection component calculated at the same moment to obtain the intermediate acceleration data after eliminating the influence of static gravity. Since the posture of the robotic arm is in a dynamic state, this set of subtraction operations will be performed point by point with each sampling action, eliminating the numerical offset caused by gravity throughout the process.

[0025] After gravity component removal, the process proceeds to high-frequency noise filtering. This process employs a two-stage composite filtering architecture for layered noise reduction. The first stage is sliding mean filtering. First, considering the effective signal frequency range corresponding to fruit shaking within the bag and pre-slipping of the clamping interface, the number of consecutive sampling points included in the sliding window is set. The acceleration data sequence is traversed window by window along the time axis, extracting the values ​​of all sampling points within a single window and calculating the arithmetic mean. This average value replaces the original data at the center of the window. Sliding mean filtering removes instantaneous spike-like high-frequency interference signals. The data after the first stage of processing enters the second stage, finite-length unit impulse response low-pass filtering. The cutoff frequency of the low-pass filter is pre-set based on the measured effective signal frequency range. The cutoff frequency is set to be higher than the maximum effective signal frequency but lower than the minimum high-frequency noise frequency. The time-series data after sliding mean processing is input into the low-pass filter, which blocks all signal components with frequencies higher than the cutoff frequency, retaining only the effective dynamic acceleration data within the low-frequency range. The two-stage filtering process... After completion, a continuous and stable finger-surface tangential acceleration signal is generated. This signal, equipped with a synchronization timestamp, is transmitted in real time to the bag disturbance identification unit. The unit then compares this signal with the stress wave signal output by the stress wave sensor in the time domain to determine the specific source of the tangential force fluctuation. This approach adapts to complex operation scenarios with multiple conveying routes, avoiding the limitations of single-direction calibration. Based on the reference posture and dynamic deflection angle, the gravity projection component is calculated and canceled point by point, solving the problem of gravity interference that cannot be uniformly eliminated under the multi-posture operation of the robotic arm. The precise partitioning calibration of the three-axis sensitive axes divides the normal direction and tangential motion direction, enabling directional acquisition of acceleration data. A two-stage composite filtering architecture distinguishes between spike noise and continuous high-frequency noise for layered processing. While filtering out interference, it fully preserves the effective waveform characteristics corresponding to fruit disturbance and interface pre-slip. The finger-surface tangential acceleration signal, after the entire process, has high purity and high timeliness, providing reliable raw data support for subsequent disturbance type determination, marker generation, and clamping force compensation control.

[0026] The bag disturbance identification unit performs a complete process of signal synchronous acquisition, timestamp binding, bandpass filtering, waveform truncation, and time-domain feature extraction. It can achieve time alignment, interference filtering, standardization, and feature quantization of two detection signals. It extracts effective waveform information corresponding to fruit shaking and interface pre-slip from the original sensor data, providing standardized comparison samples and feature data for subsequent dynamic time warping algorithms to perform waveform similarity calculation and disturbance source determination. The entire process adopts the design concept of co-source clock, dynamic frequency band calibration, same parameter filtering, and multi-dimensional time-domain feature combination, which can adapt to the frequency changes caused by different fruit loading quantities and different conveying speeds in fruit bag packaging scenarios, ensuring the consistency and effectiveness of data processing.

[0027] The bag-in-the-bag disturbance identification unit is equipped with dual parallel signal acquisition channels. The two channels share the same hardware clock source as a global time reference to achieve synchronous acquisition of stress wave signals and finger surface tangential acceleration signals. The hardware clock source outputs timing signals with microsecond-level precision. The two acquisition channels read data simultaneously according to the same sampling frequency. One channel continuously receives the stress wave signal output from the stress wave sensor at the root of the disturbance spring, while the other channel continuously receives the finger surface tangential acceleration signal after gravity component removal and two-stage noise reduction processing. After each round of synchronous sampling, the same global timestamp is applied to a set of data frames acquired by the two channels at the current moment. The unified timestamp can completely eliminate the timing offset problem between the two signals. All raw signal data carrying timestamps are stored in independent circular buffers. The circular buffers temporarily store timing data of a specified duration according to the first-in-first-out rule, which can avoid short-term data loss and reserve sufficient data space for waveform truncation of subsequent complete fluctuation events.

[0028] After completing the synchronous acquisition and timestamp binding of the signals, bandpass filtering of the two signals was immediately initiated. Before conducting the filtering operation, it was necessary to determine the effective frequency band corresponding to the fruit shaking frequency and the pre-slippage triggering frequency. This frequency band was determined by combining offline calibration and online dynamic verification. During offline calibration, a simulation test platform was built to simulate two working conditions: natural shaking of the fruit inside the fruit bag and pre-slippage of the interface between the fruit bag and the clamping finger. Stress wave signals and finger tangential acceleration signals were acquired for a long time under the condition of a single disturbance variable. Fast Fourier transform was performed on each set of acquired time-domain signals to convert the time-domain signals into frequency-domain spectra. The frequency distribution intervals corresponding to the effective signals within the spectrum were extracted frame by frame. After summarizing multiple sets of test samples, the continuous frequency intervals and interface corresponding to the fruit shaking were determined. The continuous frequency range corresponding to the pre-slip is merged to obtain the total effective frequency band. At the same time, the interference frequency ranges corresponding to the mechanical arm structure vibration, circuit electromagnetic interference, and environmental background noise are statistically analyzed. After confirming that all interference frequency bands are outside the total effective frequency band, the minimum frequency of the total effective frequency band is set as the lower cutoff frequency of the bandpass filter, and the maximum frequency of the total effective frequency band is set as the upper cutoff frequency of the bandpass filter, thus completing the calibration of the basic frequency band parameters. The online dynamic verification will be automatically executed according to a fixed cycle. The reference signal is collected and spectrum analysis is performed when the equipment is unloaded and without any disturbance. If the frequency of the environmental interference is detected to be deviated, the upper and lower cutoff frequencies of the bandpass filter are adjusted synchronously to ensure that the filter only allows the frequency band components corresponding to the fruit shaking frequency and the frequency caused by the pre-slip.

[0029] After determining the filtering parameters, the bag disturbance identification unit configures digital bandpass filters with identical structure, order, and tap coefficients for both signals. A finite-length unit impulse response architecture is used to perform the filtering operation. The unit sequentially reads the timing data of the stress wave signal and the timing data of the finger-plane tangential acceleration signal, both carrying the same timestamp, from the ring buffer. The two data streams are then fed into filtering channels with identical parameters. The filtering process employs a point-by-point convolution operation mode, sequentially extracting values ​​within the local sampling window of the signal along the time axis and combining them with preset tap coefficients to complete the convolution calculation. Signal components with frequencies between the upper and lower cutoff frequencies are fully preserved, while low-frequency baseline drift components below the lower cutoff frequency and high-frequency clutter components above the upper cutoff frequency are attenuated and filtered out. Because the parameters of the two filters are completely identical, only the target effective frequency band components are retained after filtering, and the frequency characteristics of the waveforms will not exhibit artificial differences. The filtered signals still retain the original global timestamp and are stored back in a dedicated signal buffer for subsequent processing.

[0030] After bandpass filtering, the waveform truncation stage begins. First, the current fluctuation event is identified in the filtered time-series signal. The system presets a small fluctuation threshold and iterates through all sampling points of the signal. When the amplitude of multiple consecutive sampling points deviates from the long-term steady-state mean of the signal and the deviation exceeds the small fluctuation threshold, it is determined that a fluctuation event has occurred at the current position, and the start timestamp of the fluctuation event is recorded. The device presets a fixed total waveform truncation duration, which can completely cover the entire fluctuation cycle of a single set of fruit shaking or interface pre-slip. Based on the start timestamp of the fluctuation event, the preceding sampling data is selected forward according to the fixed duration, and the following sampling data is selected backward according to the same duration. The two data segments are spliced ​​together to form a waveform segment containing the complete fluctuation event. Relying on the unified timestamp of the two signals, the corresponding data is extracted from the stress wave signal buffer area and the finger surface tangential acceleration signal buffer area according to the same time range. Finally, the signal waveform with two time segments of completely equal length, precise time axis alignment, and both containing the current fluctuation event is obtained. Even if multiple sets of consecutive fluctuation events occur within the truncation range, the preset truncation duration remains unchanged to ensure that all waveform segments participating in feature extraction have uniform specifications.

[0031] After waveform truncation, the time-domain features of the two signal waveforms are extracted. A consistent feature calculation process is performed for each waveform segment. First, the amplitudes of all sampling points within the waveform segment are iterated, and all amplitudes are summed sequentially and divided by the total number of sampling points to obtain the arithmetic mean of the waveform amplitudes. This parameter characterizes the overall DC offset level of the waveform. Second, the difference between the amplitudes of all sampling points and the arithmetic mean is calculated. The difference is squared, and the average is obtained. Then, the square root of the average is taken to obtain the amplitude standard deviation, which reflects the overall fluctuation dispersion of the waveform. Finally, the maximum and minimum amplitudes within the waveform segment are extracted. The peak-to-peak value is obtained by subtracting the minimum amplitude from the maximum amplitude. The peak-to-peak value directly reflects the overall fluctuation amplitude of a single fluctuation event. The amplitude sign of adjacent sampling points is detected point by point along the time axis. When the amplitude of two adjacent sampling points changes from positive to negative or from negative to positive, a zero-crossing action is recorded. The total number of zero-crossings in the entire waveform segment is accumulated. This parameter is used to characterize the oscillation frequency of the waveform. Then, the amplitude difference between every two adjacent sampling points is calculated. The amplitude difference is divided by the time interval between two samplings to obtain the single-point change rate. The arithmetic mean of all single-point change rates is calculated to obtain the average change rate of the waveform. This parameter reflects the overall rise and fall rate of the waveform amplitude. Finally, the total number of sampling points in the waveform segment whose amplitude falls within the preset steady-state range is counted. This number is divided by the total number of sampling points in the waveform segment to obtain the steady-state percentage. This parameter is used to distinguish between continuous disturbances and short-term sudden disturbances.

[0032] The arithmetic mean, standard deviation, peak-to-peak value, number of zero crossings, average rate of change, and steady-state percentage are combined in a fixed order to generate a set of time-domain feature vectors with identical dimensions for the waveform segments corresponding to stress waves and the waveform segments corresponding to the tangential acceleration of the finger surface. The two sets of time-domain feature vectors, along with the corresponding original waveform data and timestamp information, are transmitted to the subsequent computing module for waveform alignment and similarity calculation in the dynamic time warping algorithm, providing reliable data support for the bag disturbance identification unit to distinguish between fruit disturbance and interface pre-slip.

[0033] After extracting the temporal features of the corresponding waveform segments of the stress wave signal and the finger surface tangential acceleration signal, the bag disturbance identification unit uses a dynamic time warping algorithm to perform adaptive temporal alignment and overall similarity calculation for the two temporal waveforms with local temporal stretching or compression distortion. This eliminates the temporal misalignment problem caused by fruit shaking and micro-slippage of the clamping interface. After calculating the standardized waveform similarity value, it combines it with the pre-set feature similarity threshold to complete the judgment of the same frequency fluctuation. Relying on the optimized dynamic time warping logic and multi-dimensional calibration threshold, it can maintain the stability and accuracy of the judgment result under complex vibration waveforms.

[0034] First, the two sets of waveform data involved in the calculation are identified: one set is the stress wave time-domain waveform after bandpass filtering and equal-length truncation; the other set is the finger-plane tangential acceleration time-domain waveform obtained through the same processing flow. The total number of sampling points in the stress wave time-domain waveform is recorded as the first total number of samples, and the total number of sampling points in the finger-plane tangential acceleration time-domain waveform is recorded as the second total number of samples. Each sampling point in both sets of waveforms corresponds to an independent amplitude and the previously extracted multi-dimensional time-domain features. The first step of the dynamic time warping algorithm is to construct a two-dimensional composite distance matrix. The number of rows in the matrix equals the first total number of samples, and the number of columns in the matrix equals the second total number of samples. Each row and column intersection unit within the matrix corresponds to the matching relationship between a single sampling point of the stress wave and a single sampling point of the tangential acceleration. Calculation... The unit numerical calculation method abandons the traditional single amplitude distance calculation method and adopts a composite distance that integrates amplitude and time domain features. First, the absolute value of the amplitude difference between two sampling points is calculated, and this result is the amplitude distance. Then, the time domain feature vectors corresponding to the two sampling points are retrieved, and the Euclidean distance between the two sets of feature vectors is calculated. This result is the feature distance. Fixed weights are pre-configured for amplitude distance and feature distance respectively, and the sum of the two weights remains constant. The amplitude distance is multiplied by the corresponding weight and the feature distance is multiplied by the corresponding weight to finally obtain the composite single-point distance of the current unit. The numerical filling of the entire matrix unit is completed row by row and column by column according to this calculation rule. The design of composite distance can simultaneously consider the waveform amplitude shape and time domain variation law, and improve the matching effectiveness of distorted waveforms.

[0035] After constructing the distance matrix, a cumulative distance matrix and a path marker matrix with identical dimensions are created simultaneously. The cumulative distance matrix records the shortest cumulative distance from the matrix's starting point to the current cell, while the path marker matrix records the predecessor position of the regularized path. The algorithm sets the cell in the first row and first column of the top left corner of the matrix as the starting point of the calculation, and directly assigns the composite single-point distance of the starting point position to the starting cell of the cumulative distance matrix. At the same time, the corresponding position of the path marker matrix is ​​marked as having no predecessor node. To conform to the physical characteristics of local expansion and contraction of mechanical vibration waveforms, the algorithm adds a dedicated path step size constraint, stipulating that the regularized path can only move one column to the right, one row down, or one row and one column diagonally to the lower right from the current cell. Reverse movement and jumping across multiple cells are prohibited. This constraint can avoid invalid paths that violate the timing logic and reduce the overall computational overhead of the algorithm.

[0036] According to the path constraint rules, traverse all remaining cells of the cumulative distance matrix. For any non-starting cell, sequentially search for the three types of reachable predecessor cells: left cell, upper cell, and upper left diagonal cell. Read the cumulative distance values ​​corresponding to the three types of predecessor cells and filter out the minimum value. Add the minimum value to the composite single-point distance of the current cell. The sum is used as the cumulative distance of the current cell. At the same time, record the predecessor position number corresponding to the current cell in the path marking matrix. Continue to iterate until the endpoint cell in the last row and last column of the bottom right corner of the cumulative distance matrix is ​​assigned a value.

[0037] After the endpoint unit is assigned a value, the optimal regularized path backtracking process is started. Starting from the endpoint unit, the process traces back point by point according to the predecessor position recorded by the path marking matrix, all the way back to the matrix starting unit. The coordinates of all matrix units passed through the backtracking process are arranged in order to form the optimal dynamic regularized path between the two waveforms. This path realizes the adaptive alignment of the local timing of the two waveforms, completely offsetting the timing deviation caused by signal transmission and mechanical vibration.

[0038] Next, the total number of units contained in the optimal regular path is counted. The total cumulative distance is obtained by summing the composite single-point distances corresponding to all units on the path. The larger the total cumulative distance value, the higher the overall difference between the two waveforms. In order to form a unified evaluation index, the total cumulative distance needs to be converted into a waveform similarity value. First, the total cumulative distance is divided by the total number of units in the path to obtain the average single-point distance. Then, the preset normalization mapping function is called to perform the reverse mapping of the average single-point distance input function. Finally, the waveform similarity value with a value between zero and one is output. The smaller the average single-point distance, the closer the waveform similarity value is to the value of one, which means that the synchronous change characteristics of the two waveforms are more significant.

[0039] The feature similarity threshold, serving as a critical value to distinguish between same-frequency and non-same-frequency fluctuations, needs to be fully set through three stages: offline multi-condition calibration, on-site condition correction, and online dynamic verification. During offline calibration, a physical simulation platform is built to separately simulate three basic working conditions: fruit shaking inside the fruit bag, pre-slippage of the interface between the fruit bag and the clamping fingers, and undisturbed steady state. For each working condition, a large number of waveform samples are collected over a long period of time, and the dynamic time warping algorithm is run one by one to calculate the corresponding waveform similarity value. The minimum value of all similarity values ​​under the pure fruit shaking condition is counted, which is the critical lower limit of similarity under the same-frequency fluctuation state. Then, the maximum value of all similarity values ​​under the pure interface pre-slippage condition is counted, which is the critical upper limit of similarity under the non-same-frequency fluctuation state. The intermediate critical value is selected as the basic value of the feature similarity threshold by combining the two sets of critical values.

[0040] After completing the basic numerical settings, on-site working condition corrections were carried out. Based on the fruit bag loading type, two scenarios were divided into single-fruit loading and multi-fruit loading. Under the multi-fruit loading condition, the fruit shaking waveform is more complex, so the feature similarity threshold was slightly lowered based on the basic values. Under the single-fruit loading condition, the waveform regularity is higher, so the feature similarity threshold was slightly raised. At the same time, the speed of the robotic arm's conventional conveying speed was divided into levels. The higher the conveying speed, the greater the probability of waveform distortion, and the threshold was fine-tuned and adapted accordingly.

[0041] Finally, the online dynamic verification mechanism is activated. After a fixed period of cumulative operation, the device automatically retrieves recent steady-state samples and standard disturbance samples to recalculate the similarity distribution. The feature similarity threshold is slightly updated based on the on-site signal quality and interference intensity. All feature similarity thresholds that have undergone multi-layer calibration and correction are stored in the read-only parameter area of ​​the disturbance identification unit in the bag, and can be directly retrieved and used during device operation.

[0042] After obtaining the waveform similarity value and the feature similarity threshold, a numerical comparison judgment is performed. When the waveform similarity value is greater than the feature similarity threshold, the stress wave signal and the finger surface tangential acceleration signal are determined to be fluctuating at the same frequency, indicating that the current tangential force fluctuation is caused by the shaking of the fruit inside the bag. The bag disturbance identification unit then generates a disturbance suppression mark. When the waveform similarity value is less than or equal to the feature similarity threshold, the two signals are determined not to be fluctuating at the same frequency. At this time, a second judgment is made in combination with the steady-state parameters of the previous stress wave signal, thereby identifying the interface pre-slip and generating a slip warning mark. The generated waveform similarity value, the same-frequency fluctuation judgment result and the corresponding mark signal are transmitted outward in real time as the basis for the compensation control unit to perform adaptive adjustment of clamping force, realizing a complete data link from signal acquisition, waveform processing, similarity calculation to control command output.

[0043] After completing the dynamic time warping algorithm and obtaining the waveform similarity value, the bag disturbance identification unit initiates a multi-condition joint verification process. Combining the fluctuation state of the finger surface tangential acceleration signal, the waveform similarity index, and the steady-state condition of the stress wave signal in the time domain, the working condition is determined. This process relies on a comprehensive design of layered noise filtering, precise time interval delineation, multi-feature joint steady-state verification, and physical mechanism matching. It can distinguish between two different disturbance sources, namely fruit shaking inside the bag and pre-slippage of the clamping interface, from the signal performance level. This effectively avoids the misjudgment problem caused by single index judgment. The judgment result will be directly used to generate the corresponding marker signal, providing a trigger basis for the force control adjustment action of the back-end compensation control unit.

[0044] The finger surface tangential acceleration signal is a regularized time-series signal after gravity component removal, two-stage noise reduction, and bandpass filtering. The equipment first continuously collects hundreds of frames of amplitude data of the finger surface tangential acceleration signal under standard steady-state conditions where the robotic arm completes standard clamping, the fruit bag has no internal shaking, and the clamping interface has no relative motion. The arithmetic mean of the amplitude of all sampling points is calculated, and this average value is set as the global steady-state reference amplitude of the signal. At the same time, combined with the circuit noise floor and the small amplitude offset caused by the micro-vibration of the mechanical structure measured on site, the maximum offset of the signal amplitude under steady state is calculated, and a fixed safety margin is added on this basis to finally determine the fluctuation trigger threshold.

[0045] The bag-in-bag disturbance identification unit traverses the time-series sampling points of the finger-plane tangential acceleration signal point by point according to the fixed hardware sampling step size. For each sampling point whose amplitude is read, the amplitude offset is obtained by subtracting the global steady-state reference amplitude from the current amplitude. If the amplitude offset of only a single sampling point is greater than the fluctuation trigger threshold, the phenomenon is directly determined to be instantaneous spike noise interference and is not included in the effective fluctuation category. If the amplitude offsets corresponding to a preset number of consecutive adjacent sampling points are all greater than the fluctuation trigger threshold, the finger-plane tangential acceleration signal is officially determined to have an effective fluctuation. At the same time, the global timestamp corresponding to the first sampling point that exceeds the fluctuation trigger threshold is recorded, and this timestamp is used as the starting time point of this fluctuation event. The number of consecutive sampling points is calibrated through offline simulated noise test and is specifically used to distinguish random noise from real continuous fluctuations.

[0046] After confirming that the tangential acceleration signal has a valid fluctuation, the timing sampling points are traversed point by point, and the amplitude offset of each sampling point is continuously calculated. When the amplitude offset of a consecutive number of adjacent sampling points falls back to within the fluctuation trigger threshold, the global timestamp corresponding to that position is recorded as the fluctuation end time point. The complete timing interval from the fluctuation start time point to the fluctuation end time point is the fluctuation duration of this event. Relying on the shared source clock and unified timestamp of the two signals, the sampling data and time domain characteristic parameters of all stress wave signals within this time interval are synchronously captured to ensure that the verification intervals of the two signals are completely corresponding. Next, the definition and calibration method of the steady-state threshold range are clarified. The steady-state threshold range is the numerical range for judging whether the stress wave signal is in a normal static state. Using the previously extracted six time-domain feature parameters, namely the arithmetic mean of amplitude, standard deviation, peak-to-peak value, number of zero crossings, average rate of change, and steady-state percentage, a dedicated steady-state threshold range is defined for each parameter. During calibration, stress wave signals are collected for a long time under a standard steady-state environment where the fruit bag is completely still and the clamping interface is free from pre-slippage. The value of each time-domain feature is calculated frame by frame. The minimum and maximum values ​​of each parameter in the full steady-state sample are counted. The minimum value is set as the lower limit of the steady-state threshold range of the parameter, and the maximum value is set as the upper limit. This forms the independent steady-state threshold range for each of the six time-domain features. Considering the presence of weak background disturbances in the field environment that cannot be eliminated, the steady-state threshold range allows the parameters to change slightly within the range. Once the range is exceeded, it indicates that the stress wave signal has undergone substantial changes. All steady-state threshold range parameters are uniformly stored in the local read-only storage area. At the same time, it supports the staff to make small on-site corrections according to the fruit bag specifications, fruit types, and robotic arm conveying speed, adapting to diverse operating scenarios.

[0047] After defining the interval, the steady-state verification of the stress wave signal's time-domain characteristics begins. The defined fluctuation duration is divided into multiple interconnected time sub-windows with fixed short durations. The duration of each sub-window is much shorter than the overall fluctuation duration, thus achieving high-density encrypted verification. Within each time sub-window, the complete set of six time-domain characteristic parameters corresponding to the stress wave signal is recalculated. Then, the calculated value of each parameter is compared one by one with the upper and lower limits of the steady-state threshold range corresponding to that parameter. If the parameter value is greater than or equal to the lower limit and less than or equal to the upper limit, the individual time-domain characteristic is determined to be within the steady-state interval. If the parameter value is lower than the lower limit or higher than the upper limit, the individual characteristic is determined to have an anomaly. Only when all six time-domain characteristic parameters of all time sub-windows within the fluctuation duration are stable within their respective steady-state threshold ranges can the stress wave signal be considered to have maintained a steady state within the fluctuation duration. If any parameter in any sub-window exceeds the corresponding interval, the stress wave signal is determined to have an anomaly and does not meet the steady-state condition.

[0048] After sequentially completing signal fluctuation judgment, fluctuation duration determination, and stress wave steady-state verification, the final working condition judgment is carried out based on the previously obtained waveform similarity value. First, three necessary judgment conditions are identified: the first condition is that the tangential acceleration signal on the finger surface has been confirmed to have valid fluctuations; the second condition is that the waveform similarity value output by the dynamic time warping algorithm is lower than the device's preset feature similarity threshold; and the third condition is that all time-domain feature parameters of the stress wave signal are stable within the corresponding steady-state threshold range during the fluctuation duration. When all three conditions are met simultaneously, the final judgment is completed by combining the physical formation mechanisms of the two types of disturbances. The shaking of the fruit inside the fruit bag will impact and squeeze the sidewall of the fruit bag, causing the disturbance spring to undergo periodic deformation and output alternating stress wave signals. At this time, the stress wave signal and the tangential acceleration signal on the finger surface will change at the same frequency, and the waveform similarity value will be higher than the feature similarity threshold. The pre-slippage of the clamping interface only occurs on the contact surface between the clamping finger and the fruit bag, which is a small tangential relative slippage. This movement only causes the clamping finger to generate tangential acceleration. The fluctuations do not cause deformation of the fruit bag sidewalls, so the stress wave signal will always maintain steady-state characteristics. Based on this logic, it can be determined that the current tangential force fluctuation is not caused by fruit disturbance inside the fruit bag, but by pre-slippage of the interface between the fruit bag and the clamping fingers. By continuously sampling points, instantaneous noise and effective fluctuations are distinguished, reducing misjudgments caused by environmental interference from the source. The duration of synchronous fluctuations is defined based on a unified timestamp, ensuring that the timing of the two signals is completely matched and avoiding timing misalignment from affecting the judgment result. Steady-state threshold ranges are calibrated for six time-domain features, breaking through the limitations of single numerical evaluation and comprehensively depicting the operating state of the stress wave signal. A sub-window encrypted verification method is used to segment and check long-term fluctuation intervals to prevent local short-term anomalies from being masked by the overall interval. A multi-condition joint judgment framework is built based on the physical mechanism of disturbance formation, combining signal morphology, similarity, and steady-state state. Compared with single index judgment, it has higher reliability. The threshold parameter system that can be corrected on-site can adapt to different fruit bags, fruits, and transportation conditions.

[0049] After completing the interface pre-slip determination, the bag disturbance identification unit immediately generates a standardized slip warning mark. The mark is sent to the compensation control unit in real time in the form of a data message. Under the premise that only the slip warning mark is received and the disturbance suppression mark is not received, the compensation control unit will retrieve the preset acceleration timing of the motion pre-reading unit to predict the peak value of the tangential inertial force, and calculate the clamping safety margin in combination with the real-time data of the normal force sensing module. Then, it will decide whether to issue a clamping force pre-increase command based on the threshold condition.

[0050] Same-frequency fluctuation is defined as follows: when the waveform similarity value exceeds the feature similarity threshold, the stress wave signal and the finger surface tangential acceleration signal are determined to be same-frequency fluctuations; if the waveform similarity value is lower than the feature similarity threshold, they are determined to be non-same-frequency fluctuations.

[0051] After the bag disturbance identification unit completes the working condition judgment and generates a slippage warning mark and a no-output disturbance suppression mark, the corresponding mark signal will be transmitted to the compensation control unit in real time. The compensation control unit then starts the clamping safety margin calculation process based on motion timing prediction, real-time force detection, and friction characteristic calculation. This process relies on the previous motion data to predict the maximum tangential inertial load in the future, combines the real-time normal clamping force and interface friction parameters to quantify the anti-slip safety margin, and uses a mechanical model to complete standardized calculations. It can identify the slippage risk of the clamping interface in advance, provide accurate quantitative basis for the clamping force pre-increase command, and realize active anti-slip control in the process of the robotic arm's fruit bag clamping and conveying.

[0052] The compensation control unit first retrieves the preset acceleration time sequence previously parsed and stored by the motion pre-reading unit through a dedicated high-speed data interface inside the device. The preset acceleration time sequence is a sequence of tangential acceleration components arranged in chronological order. Each discrete data point in the sequence is equipped with a globally unified timestamp, which can accurately correspond to the tangential motion state of the end of the gripper finger at different times. At the same time, the compensation control unit reads the global timestamp of the current operation of the robotic arm, thereby locking the position of the robotic arm in the entire motion time sequence and establishing a time sequence benchmark for subsequent future interval data screening.

[0053] Next, the preset future time window is defined and activated. The future time window is a continuous duration interval defined by combining hardware response characteristics and on-site working conditions. During calibration, the hardware response delay from the issuance of the robotic arm control command to the actual change in the clamping force of the gripping finger is measured. Then, the shortest evolution time from the appearance of the fruit bag to the complete slippage is calculated. The two types of duration values ​​are added together to obtain the base duration of the future time window. At the same time, dynamic adaptation is performed by distinguishing three typical working conditions: normal smooth conveying, route turning conveying, and rapid acceleration and deceleration conveying. In the turning conveying and rapid acceleration and deceleration conveying working conditions, the tangential motion of the robotic arm changes more drastically, and the window duration is expanded by a fixed proportion on the base duration. In the smooth conveying working condition, the base duration is directly used. The starting boundary of the future time window is set as the current global timestamp of the robotic arm, and the ending boundary is the current timestamp superimposed with the corresponding window duration. The entire window range strictly falls within the future motion interval that has not yet been executed in the preset acceleration timing sequence, ensuring that the selected data all correspond to the subsequent motion state of the robotic arm.

[0054] After determining the range of the future time window, the compensation control unit traverses the tangential acceleration values ​​corresponding to all discrete time points covered by the window, calculates the absolute value of each acceleration value in turn, compares all the absolute values, and selects the result with the largest value. This result is the maximum absolute value of acceleration within the future time window. This value represents the peak value of the tangential motion intensity of the robotic arm in the subsequent conveying stage, and is also the corresponding motion parameter where the tangential inertial force reaches its maximum.

[0055] The equivalent mass parameter is used to characterize the overall motion mass of the fruit bag and the fruit loaded inside. This parameter adopts a dual-mode setting method of offline batch calibration combined with online dynamic correction. In the offline stage, for different specifications of fruit bags and different fruit loading quantities used at the work site, the standard weighing equipment is used to weigh the mass of the empty fruit bag itself, the mass of a single fruit, and the total mass of multiple fruits. The mass of the empty fruit bag is added to the total mass of the fruit to obtain the basic equivalent mass under different working conditions and it is classified and stored. During online operation, at the moment the robotic arm completes the fruit bag gripping action, the actual total mass of the fruit bag and fruit is calculated by combining the initial clamping force change of the gripping fingers and the acceleration data at the moment the equipment starts, through a mechanical back-calculation algorithm. This actual total mass is used to correct the offline basic equivalent mass, offsetting the errors caused by the thickness of the fruit bag material and the individual weight differences of the fruit. The corrected value is the equivalent mass parameter corresponding to the current operation.

[0056] After obtaining the maximum absolute value of acceleration and the equivalent mass parameter, calculations are carried out according to the rules of inertial force calculation. The maximum absolute value of acceleration is multiplied by the equivalent mass parameter, and the result obtained after multiplication is the predicted peak value of tangential inertial force. The peak value of tangential inertial force is the maximum inertial force that the fruit bag will bear along the tangential direction of motion in the future period of time, and it is also the core load that induces the slippage of the clamping interface.

[0057] While completing the prediction of the peak inertial force, the compensation control unit continuously reads the detection data of the normal force sensing module. The normal force sensing module is embedded in the gripping surface of the gripping finger and outputs the clamping force data between the gripping finger and the fruit bag according to a fixed sampling frequency. Each set of current normal force values ​​read is simultaneously equipped with a global timestamp to ensure that the normal force data is completely aligned with the acceleration time sequence and the future time window, avoiding calculation deviations caused by time sequence misalignment of multi-source data.

[0058] The coefficient of kinetic friction is a parameter describing the frictional characteristics between the outer surface of the fruit bag and the clamping surface of the clamping fingers. This parameter is calibrated through multivariate offline testing combined with online environmental correction. In the offline stage, a dedicated friction test platform is built, and the mainstream fruit bag materials and clamping finger contact surface materials are selected. The normal clamping force, relative sliding speed of the interface, ambient temperature and humidity, and other influencing factors are changed in turn to carry out multiple sets of repeated friction tests. For each set of tests, the sliding friction force and the corresponding normal force are measured. The friction ratio of a single set is obtained by dividing the sliding friction force by the normal force. The arithmetic mean of all test results is calculated and used as the basic value of the coefficient of kinetic friction. During the operation of the equipment, the supporting temperature and humidity acquisition element monitors the on-site environmental parameters in real time. A humid environment will change the physical properties of the fruit bag surface and thus reduce the friction capacity. The coefficient of kinetic friction is slightly adjusted down according to the preset correction rules. In a dry environment, the basic value is kept unchanged. The value after environmental correction is the effective coefficient of kinetic friction under the current working conditions.

[0059] After obtaining the current normal force value and the effective dynamic friction coefficient, the current maximum available static friction force is calculated. The calculation method is to multiply the current normal force value by the dynamic friction coefficient. The product result is the current maximum available static friction force. This value represents the limit force that the clamping interface can resist under the existing clamping force.

[0060] Finally, the clamping safety margin is calculated by subtracting the predicted peak tangential inertial force from the current maximum available static friction force. The difference between the two is the clamping safety margin. The clamping safety margin directly reflects the remaining anti-slip capability in the clamping state. A difference greater than zero means that the existing friction force can offset the maximum tangential inertial force, and the clamping state is within the safe range. The closer the difference is to zero, the smaller the anti-slip margin, and the slippage risk gradually increases. A difference less than zero means that the existing friction force cannot resist the inertial load, and interface slippage will quickly develop into complete slippage. The dynamically adjustable future time window is designed in conjunction with hardware latency and operating condition characteristics, overcoming the limitation that fixed time windows cannot adapt to complex conveying scenarios such as turning, acceleration, and deceleration, and significantly improving the prediction accuracy of the peak inertial force. The equivalent mass parameter adopts offline calibration plus online feedback. The push-correction mode solves the problem of inaccurate quality parameters caused by individual differences in fruit bags and fruits, making the inertial force calculation more in line with actual working conditions. The dynamic friction coefficient integrates multi-variable test calibration and dynamic correction of environmental temperature and humidity, fully considering the influence of the on-site environment and contact surface condition on friction characteristics. Compared with the fixed friction coefficient calculation method, it has stronger fault tolerance. The time sequence alignment mechanism with unified timestamps throughout the entire link ensures the time sequence synchronization of acceleration data, normal force data, and window interval data, avoiding calculation errors from the data level. The pre-prediction calculation logic relies on the time sequence data of the motion pre-reading unit to predict future dangerous loads in advance. Unlike the traditional passive control method of adjusting after slippage occurs, it realizes risk prevention and control in advance. The clamping safety margin calculation model based on classical mechanics facilitates the subsequent triggering of control commands by combining thresholds.

[0061] The calculated clamping safety margin is saved to the calculation cache in real time. The compensation control unit will then compare this value with the preset safety margin threshold. When the clamping safety margin is lower than the preset threshold, a clamping force pre-increase command is sent to the clamping execution unit before the motion moment corresponding to the peak of the tangential inertial force is reached. By increasing the normal clamping force in advance, the maximum static friction force of the interface is increased, thereby offsetting the slippage risk caused by the tangential inertial force, forming a force-sensitive adaptive anti-slip closed-loop control for the clamping operation of the fruit bag packaging robot arm.

[0062] Please see Figure 2 As shown, a second objective of this invention is to provide a method for implementing a force-sensitive adaptive clamping system for fruit bag packaging using a robotic arm, comprising any one of the above-mentioned features, including the following steps: S1. Obtain the preset acceleration timing for the clamping and conveying stage; S2. In the clamping state, the current normal force value and finger surface tangential acceleration signal are obtained through the normal force sensing module and the tangential acceleration sensing module, respectively, and the stress wave signal is obtained through the disturbance spring and stress wave sensor. S3. After synchronously acquiring and bandpass filtering the stress wave signal and the finger surface tangential acceleration signal, perform time-domain comparison and calculate the waveform similarity value between the two signal waveforms. S4. If the waveform similarity value exceeds the preset feature similarity threshold, it is determined to be a same-frequency fluctuation, and a disturbance suppression mark is generated. S5. If the tangential acceleration signal of the finger surface fluctuates and the waveform similarity value is lower than the feature similarity threshold, while the time domain feature parameters of the stress wave signal remain in a steady state, it is determined to be interface pre-slip and a slip warning mark is generated. S6. When a slip warning flag is received but a disturbance suppression flag is not received, the peak value of the tangential inertial force is predicted according to the preset acceleration timing, and the clamping safety margin is calculated in combination with the current normal force value. S7. When the clamping safety margin is lower than the preset threshold, generate and issue a clamping force pre-increase command before the peak value of the tangential inertial force is reached.

[0063] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging, characterized in that, include: The clamping execution unit has at least one clamping finger, and the clamping surface of the clamping finger is provided with a normal force sensing module and a tangential acceleration sensing module; The motion pre-reading unit is used to obtain the preset acceleration timing of the gripping and conveying phase from the robotic arm controller; The bag disturbance identification unit includes a disturbance spring fixed to the side of the clamping finger and a stress wave sensor attached to the root of the disturbance spring; The free end of the disturbance spring is attached to the side wall of the fruit bag in the clamping state, which is used to transmit the deformation of the bag wall caused by the shaking of the fruit inside the bag into a stress wave signal that can be detected by the stress wave sensor. The bag disturbance identification unit compares the stress wave signal with the finger surface tangential acceleration signal detected by the tangential acceleration sensing module in the time domain. If the same frequency fluctuation occurs, it is determined that the current tangential force fluctuation is caused by the disturbance of the fruit inside the bag, and a disturbance suppression mark is generated. If only the finger surface tangential acceleration signal fluctuates while the stress wave signal remains steady, it is determined that the current tangential force fluctuation is caused by the pre-slippage of the interface between the fruit bag and the clamping finger, and a slippage warning mark is generated. The compensation control unit calculates the clamping safety margin based on the peak value of the tangential inertial force predicted by the preset acceleration timing and the current normal force value only when it receives the slip warning flag and does not receive the disturbance suppression flag. When the clamping safety margin is lower than the preset threshold, it issues a clamping force pre-increase command to the clamping execution unit before the peak value of the tangential inertial force is reached.

2. The robotic arm force-sensitive adaptive clamping system for fruit bag packaging according to claim 1, characterized in that: The motion pre-reading unit is used to obtain the preset acceleration timing sequence of the gripping and conveying phase from the robotic arm controller, specifically including: A real-time data communication link is established between the motion pre-reading unit and the robotic arm controller. The real-time data communication link is used to receive motion planning data packets sent by the robotic arm controller before performing the clamping and conveying task. The motion pre-reading unit parses the motion planning data packets, extracts the planning trajectory corresponding to the clamping and conveying stage, and calculates the linear acceleration vector of the clamping finger tip at each discrete time point on the conveying path from the planning trajectory in chronological order. The component values ​​of the linear acceleration vector in the direction perpendicular to the normal of the clamping surface are arranged in chronological order to form the preset acceleration time sequence.

3. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 1, characterized in that: The free end of the disturbance spring plate, when clamped, adheres to the side wall of the fruit bag, and is used to transmit the deformation of the bag wall caused by the shaking of the fruit inside the bag into a stress wave signal detectable by the stress wave sensor, specifically including: The disturbance spring is a flexible spring with a preset bending stiffness. The free end of the disturbance spring is machined with a slightly convex arc surface that matches the curvature of the side wall of the fruit bag. In the clamping state, the slightly convex arc surface maintains surface contact with the side wall of the fruit bag through a preset pre-pressure. When the fruit inside the fruit bag shakes, the side wall of the fruit bag undergoes reciprocating deformation. The reciprocating deformation is transmitted to the disturbance spring through the slightly convex arc surface, causing alternating strain at the root of the disturbance spring. The stress wave sensor attached to the root of the disturbance spring detects the alternating strain and converts it into a stress wave signal with the same frequency as the deformation of the side wall of the fruit bag.

4. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 1, characterized in that: The tangential acceleration sensing module is a three-axis MEMS accelerometer embedded inside the clamping finger. Its installation orientation is calibrated so that the sensitive axis is parallel to the normal direction of the clamping surface of the clamping finger, and the other sensitive axis is parallel to the main motion direction during the clamping and conveying stage, so as to detect tangential acceleration. During operation, the three-axis MEMS accelerometer outputs raw acceleration data in three axes in real time. By reading the output data of the sensitive axis parallel to the main motion direction, and after filtering to remove the gravitational acceleration component and high-frequency noise, the tangential acceleration signal of the finger surface is obtained.

5. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 4, characterized in that: The stress wave signal is compared in the time domain with the finger-surface tangential acceleration signal detected by the tangential acceleration sensing module, specifically including: The bag disturbance identification unit synchronously acquires stress wave signals and finger surface tangential acceleration signals, and applies the same timestamp to the stress wave signals and finger surface tangential acceleration signals. The stress wave signals and finger surface tangential acceleration signals carrying timestamps are subjected to the same bandpass filtering process to retain frequency band components related to the fruit shaking frequency and the pre-slip initiation frequency. From the filtered stress wave signals and finger surface tangential acceleration signals, two signal waveforms with the same time length, including the current fluctuation event, are extracted, and the time domain features of these two signal waveforms are extracted.

6. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 5, characterized in that: After extracting the time-domain features of the two signal waveforms, the bag-in-the-bag disturbance identification unit uses a dynamic time warping algorithm to align and calculate the similarity between the time-domain waveform of the stress wave signal and the time-domain waveform of the finger-surface tangential acceleration signal to obtain a waveform similarity value. The bag-in-the-bag disturbance identification unit has a preset feature similarity threshold to determine whether they belong to the same frequency fluctuation.

7. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 6, characterized in that: The same-frequency fluctuation is defined as follows: when the waveform similarity value exceeds the feature similarity threshold, the stress wave signal and the finger surface tangential acceleration signal are determined to be same-frequency fluctuations; if the waveform similarity value is lower than the feature similarity threshold, they are determined to be non-same-frequency fluctuations.

8. The force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 6, characterized in that: When the finger surface tangential acceleration signal fluctuates and the waveform similarity value is lower than the feature similarity threshold, while the time domain feature parameters of the stress wave signal stabilize within a preset steady-state threshold range during the duration of the fluctuation, it is determined that only the finger surface tangential acceleration signal fluctuates while the stress wave signal remains steady, and it is further determined that the current tangential force fluctuation originates from interface pre-slip.

9. A force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging according to claim 6, characterized in that: The compensation control unit reads the preset acceleration time sequence, finds the maximum absolute value of acceleration within a preset future time window from the preset acceleration time sequence, multiplies the maximum absolute value of acceleration by the equivalent mass parameter obtained from the total mass of the fruit bag and the fruit inside, and calculates the predicted peak value of the tangential inertial force. Simultaneously, the compensation control unit reads the current normal force value detected by the normal force sensing module in real time. The calculation process for the clamping safety margin is as follows: The product of the current normal force value and the dynamic friction coefficient is taken as the current maximum available static friction force. The difference between the current maximum available static friction force and the predicted peak value of the tangential inertial force is taken as the clamping safety margin.

10. A method for implementing a force-sensitive adaptive clamping system for a robotic arm used in fruit bag packaging, comprising any one of claims 1-9, characterized in that: Includes the following steps: S1. Obtain the preset acceleration timing for the clamping and conveying stage; S2. In the clamping state, the current normal force value and finger surface tangential acceleration signal are obtained through the normal force sensing module and the tangential acceleration sensing module, respectively, and the stress wave signal is obtained through the disturbance spring and stress wave sensor. S3. After synchronously acquiring and bandpass filtering the stress wave signal and the finger surface tangential acceleration signal, perform time-domain comparison and calculate the waveform similarity value between the two signal waveforms. S4. If the waveform similarity value exceeds the preset feature similarity threshold, it is determined to be a same-frequency fluctuation, and a disturbance suppression mark is generated. S5. If the tangential acceleration signal of the finger surface fluctuates and the waveform similarity value is lower than the feature similarity threshold, while the time domain feature parameters of the stress wave signal remain in a steady state, it is determined to be interface pre-slip and a slip warning mark is generated. S6. When the slip warning flag is received but the disturbance suppression flag is not received, the peak value of the tangential inertial force is predicted according to the preset acceleration timing, and the clamping safety margin is calculated in combination with the current normal force value. S7. When the clamping safety margin is lower than a preset threshold, generate and issue a clamping force pre-increase command before the peak value of the tangential inertial force is reached.

Images