Forklift active anti-rollover control system and method with multi-source sensing and dynamic detection functions

Abstract

The invention discloses a forklift active anti-rollover control system and method based on multi-source sensing and dynamic detection. The defects that in the prior art, sensing is single, and control is extensive are overcome. The system comprises a sensing module (a forward binocular camera and an inertia / acceleration / steering angle sensor), a data processing module (a YOLOv8 algorithm and a dynamic model), a hierarchical control module and an execution module. According to the method, cargo images and forklift operation data are collected, cargo inclination parameters are fused and recognized, the gravity center and the roll moment of the forklift are calculated, five risk grades are divided, and progressive control over sound-light alarm, reverse compensation moment, forced deceleration, emergency braking and power cut-off is correspondingly achieved. The system realizes synchronous monitoring of cargo and forklift states, is accurate in control, adapts to various forklifts, effectively improves the anti-rollover early warning accuracy and operation safety, and is suitable for warehouse logistics and other complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of forklift safety control technology, specifically to a multi-source sensing and dynamic detection active anti-rollover control system and method for forklifts. Background Technology

[0002] Forklifts, as special equipment, are commonly used in the field of material handling and are widely used in warehousing, logistics, factories and other scenarios. However, rollover accidents are always the primary safety hazard in their operation, which can not only cause equipment damage and cargo loss, but also potentially lead to personal injury or death.

[0003] While existing technologies have researched solutions for preventing forklift rollovers, significant technical shortcomings remain, making it difficult to meet safety requirements under complex working conditions. For example, Chinese patent application CN120208137A discloses a rollover prevention control method and system for a counterbalanced heavy-duty forklift. This method involves collecting forklift operating data using a sensor array, transmitting it to a central controller via a CAN bus, calculating the balance between the static stability triangle boundary and dynamic lateral torque using a rollover risk assessment algorithm, outputting decision commands, and adjusting the load and torque vector distribution to control wheel braking via hydraulic servo control. However, this technology only focuses on the forklift's operating parameters and does not consider the impact of the cargo's tilt on rollover risk. When the cargo tilts, the forklift's actual center of gravity shifts, but the system cannot identify this shift, easily leading to delayed or misjudged risk assessments.

[0004] In addition, Chinese patent application CN114890345B, a forklift stability control method and control system, discloses a method that collects forklift acceleration, tilt angle, and vertical wheel load to calculate the lateral load transfer rate and determine the forklift state, thereby adjusting the forklift posture. This technology also has limitations: First, the control method is singular, relying solely on posture adjustment to address the risk of tipping over, lacking a tiered control strategy—aggressive control (such as direct braking) when the risk is low can easily lead to increased cargo swaying, while insufficient control when the risk is high cannot effectively prevent tipping over; second, it does not incorporate visual detection methods, making it impossible to capture dynamic risk sources such as cargo tilting and deviation in real time, resulting in a "passive" anti-tip-over control approach.

[0005] In summary, existing forklift anti-tipping technologies generally suffer from two major problems: 1) The perception dimension is singular, only monitoring the state of the forklift itself and ignoring the impact of cargo tilting on the center of gravity, resulting in incomplete risk identification; 2) The control strategy is crude, lacking adaptive hierarchical control based on risk level, resulting in insufficient control accuracy and safety. Summary of the Invention

[0006] The present invention aims to provide a forklift active anti-rollover control system and method based on multi-source sensing and dynamic detection, thereby solving the above-mentioned technical problems existing in the prior art.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following solution: A multi-source sensing and dynamic detection active anti-rollover control system for forklifts includes: The sensing module is used to collect image data of the goods carried by the forklift and the operating status data of the forklift. The sensing module includes a forward-facing binocular camera, an inertial sensor, an acceleration sensor, and a steering angle sensor installed on the top of the forklift mast. The forward-facing binocular camera is used to collect real-time image data of the loaded goods. The inertial sensor is used to collect the forklift's tilt angle and angular velocity data. The acceleration sensor is used to collect the forklift's lateral acceleration data. The steering angle sensor is used to collect the forklift's steering angle and steering speed data.

[0008] The data processing module, electrically connected to the sensing module, is used to fuse and process the image data and forklift operating status data collected by the sensing module to determine whether the forklift is at risk of tipping over. The data processing module has a built-in preset YOLO algorithm and dynamic analysis model. It processes the image data using the YOLO algorithm to identify the tilt parameters of the goods, and calculates the forklift's center of gravity position, tilting moment, and lateral load transfer rate using the dynamic analysis model. The hierarchical control module, electrically connected to the data processing module, is used to output hierarchical control commands based on the forklift rollover risk assessment results. The hierarchical control module includes sequentially progressive control levels: instrument panel audible and visual alarm, application of reverse compensation torque of electronic steering system, forced deceleration of forklift, emergency braking of forklift, and cutoff of drive power.

[0009] The execution module is electrically connected to the hierarchical control module and is used to execute hierarchical control commands; the execution module includes an instrument panel alarm unit, an electronic steering controller, a brake controller, and a power controller.

[0010] Further optimization includes an embedded processor for running the YOLO algorithm and dynamic analysis model in parallel, and for interacting with the sensing module and hierarchical control module via a CAN bus.

[0011] Further optimization involves using the YOLOv8 algorithm, pre-trained on a forklift loading cargo dataset. The cargo tilting parameters include the cargo tilting angle, tilting direction, and tilting rate.

[0012] Further optimization is achieved by defining the control logic of the hierarchical control module as follows: When only a cargo tilt angle of 5°~10° is detected, an audible and visual alarm command is output to the instrument panel. When the tilt angle of the goods exceeds 10° or the side tilt angle of the forklift is 8°~15°, a reverse compensation torque command is output. When the forklift tilt angle exceeds 15° or the lateral acceleration is 1.2g~1.5g, a forced deceleration command is output; An emergency braking command is issued when the forklift tilt angle exceeds 20° or the lateral load transfer rate exceeds 0.8. When the forklift tilt angle still exceeds 25° after emergency braking, a drive power cut-off command is output.

[0013] Further optimizations include a data storage module, electrically connected to the data processing module, used to store the raw data collected by the sensing module, data processing results, and hierarchical control command records, with a storage time of no less than 30 days.

[0014] A multi-source sensing and dynamic detection method for active anti-tipping control of forklifts, applied to the above system, includes the following steps: Step S1: Start the perception module, collect real-time image data of the goods loaded on the forklift through the forward-facing binocular camera, and collect real-time operating status data of the forklift through the inertial sensor, acceleration sensor and steering angle sensor. The data is transmitted to the data processing module via the CAN bus.

[0015] Step S2: The data processing module receives the image data and the real-time operating status data of the forklift, processes the image data using the YOLOv8 algorithm to identify the tilt angle, tilt direction and tilt rate of the goods; processes the real-time operating status data of the forklift using the dynamic analysis model to calculate the forklift's center of gravity position, tilt moment, lateral acceleration and lateral load transfer rate. Step S3: The data processing module determines the rollover risk level based on the cargo tilt parameters and forklift operating parameters. Risk Level 1: The cargo tilt angle is 5°~10°, with no risk of the forklift tipping over; Risk Level 2: Cargo tilt angle >10° or forklift side tilt angle 8°~15°; Risk Level 3: Forklift tilt angle >15° or lateral acceleration 1.2g~1.5g; Risk Level 4: Forklift tilt angle > 20° or lateral load transfer rate > 0.8; Risk Level 5: Forklift roll angle >25° after emergency braking; Step S4: The graded control module outputs corresponding control instructions based on the risk level, and the execution module executes the instructions: Corresponding to risk level 1: The dashboard alarm unit activates an audible and visual alarm; Corresponding risk level 2: The electronic steering controller applies a reverse compensation torque; Corresponding risk level 3: The brake controller performs forced deceleration, reducing the forklift speed to below 30% of the rated speed; Corresponding to risk level 4: The brake controller performs emergency braking, bringing the forklift to a stop within 1.5 seconds; Corresponding risk level 5: The power controller cuts off the power output of the drive engine or electric motor; Step S5: After the instruction is executed, the sensing module continues to collect data, and the data processing module re-evaluates the risk level until the risk level drops below 1, at which point the forklift returns to normal operation.

[0016] Further optimization, in step S2, the process by which the data processing module processes the image data using the YOLOv8 algorithm to identify the cargo tilt angle and tilt direction includes: S211: Perform distortion correction on the image data acquired by the forward-facing binocular camera. The distortion correction is based on a preset camera intrinsic parameter matrix to eliminate radial and tangential distortion of the image and obtain a corrected binocular image. S212: Perform stereo matching on the corrected binocular images. By calculating the disparity map of the left and right eye images and combining the baseline distance and focal length parameters of the binocular cameras, obtain the three-dimensional coordinate data of the feature points on the surface of the cargo. S213: Call the target detection branch of the YOLOv8 algorithm to identify the contour features of the cargo in the 3D coordinate data and determine the boundary contours of the bottom and top surfaces of the cargo. S214: Establish a three-dimensional coordinate system with the center of the bottom surface of the goods as the origin. The x-axis is along the forklift's travel direction, the y-axis is perpendicular to the travel direction, and the z-axis is vertical. Calculate the angle between the lines connecting the feature points on the top and bottom edges of the goods in the yz plane. This angle is the tilt angle of the goods. At the same time, determine the tilt direction of the goods—whether it is tilting to the left, right, forward, or backward—based on the offset direction of the top surface profile relative to the bottom surface profile. S215: The forward-facing binocular camera continuously captures cargo images at a frame rate of not less than 30fps, and the data processing module extracts the cargo tilt angles θ1 and θ2 corresponding to two adjacent frames of images, respectively. Calculate the tilt rate v θ =|θ2-θ1| / Δt, the unit is ° / s; Δt is the time interval between two adjacent frames, Δt=1 / frame rate.

[0017] Further optimization involves step S2, where the real-time operating status data of the forklift is processed using a dynamic analysis model to calculate the center of gravity position, tilting moment, lateral acceleration, and lateral load transfer rate of the forklift and cargo. Specifically, this includes: S221: Based on the three-dimensional coordinate data of the cargo acquired by the forward-facing binocular camera, the coordinates of the geometric center of the cargo are taken as the position of the load center of gravity. x l , yl , z l And estimate the load mass based on the cargo's three-dimensional dimensions and material density. m l Forklift's own center of gravity position ( x v , y v , z v The value is a preset fixed value; the composite centroid is calculated using the following formula ( x cg , y cg , z cg ) position, where: x cg =( m v × x v + m l × x l ) / ( m v + m l ); y cg =( m v × y v + m l × y l ) / ( m v + m l ); z cg =( m v × z cgv + m l × x l ) / ( m v + m l ); in, m v + m l Total mass of forklift m total; S222: Based on the forklift's total mass, load mass, wheelbase, and track width parameters, establish the forklift roll dynamics equations: , in, M lat This is the lateral tilting moment. m v For the sake of the forklift's own quality, a y For lateral acceleration, h cgv The height of the forklift's center of gravity. m l For load mass, h cgl This refers to the height of the load's center of gravity. S223: Calculate the lateral load transfer rate by combining the roll angle θ collected by the inertial sensor: , Where g is the acceleration due to gravity and B is the wheelbase.

[0018] In a further optimization, in step S222, the data processing module performs Kalman filtering on the raw lateral acceleration data to obtain the effective lateral acceleration. a y Specifically: x k =A×x k-1 +K×u k-1 +w k-1 z k =H×x k +v k in, x k for k The time-lapse filtering result, i.e., the effective lateral acceleration a y , x k-1 for k The filtering result at time -1, where A is the state transition matrix and K is the control matrix. u k-1 for k The control quantity at time -1 w k-1 The noise is the process noise, which follows a mean of 0 and a variance of . Gaussian distribution; z k fork (Raw data of lateral acceleration at time step H, where H is the observation matrix). v k To observe the noise, it follows a mean of 0 and a variance of . The Gaussian distribution.

[0019] Further optimization is achieved in step S4, where the magnitude of the reverse compensation torque is dynamically adjusted based on the steering angle and tilt direction: when the forklift turns to the left and the cargo tilts to the left, the electronic steering controller applies a compensation torque to the right, and the magnitude of the torque is positively correlated with the steering angle and the cargo tilt angle.

[0020] The magnitude of the reverse compensation torque applied by the electronic steering controller is dynamically adjusted according to the forklift steering angle and the cargo tilt direction, specifically including: Define the steering angle α, set left turn as positive, range -180°~180°; define the cargo tilt angle θ, set left tilt as positive, range -30°~30°, and the reverse compensation torque T; When α>0 and θ>0, the electronic steering controller applies a rightward compensation torque, satisfying the formula: T= k 1| α |+ k 2| i |, among which k 1 represents the steering angle proportionality coefficient, with a value ranging from 0.005 to 0.02 N·m / °. k 2 is the tilt angle proportionality coefficient, with a value of 0.1~0.3 N・m / °; When α < 0 and θ < 0, the electronic steering controller applies a compensating torque to the left, satisfying the formula: T = -( k 1| α |+ k 2| i |); when α and i When the signs are opposite, the absolute value of the compensation torque T is calculated according to the above formula, and the direction is taken in the opposite direction to the steering angle; The maximum value of the reverse compensation torque is no more than 2 N·m and the minimum value is no less than 0.5 N·m, and it is adjusted in real time through the PID closed-loop control of the electronic steering controller.

[0021] Further optimization is achieved in step S4 by adjusting the braking pressure of the brake controller: the braking pressure is linearly increased from 0.8MPa to 1.5MPa, and the deceleration time is not less than 0.8s, so as to avoid the load from tilting twice due to rapid deceleration.

[0022] Further optimization: In step S5, when the risk level drops below 1, the graded control module outputs a "risk released" signal, the instrument panel displays a normal operation indicator, and the electronic steering controller and brake controller return to their default states.

[0023] Further optimization includes step S6: The data storage module stores the raw data, processing results and control commands from steps S1-S5 in real time, and supports data export via Ethernet interface for fault tracing and algorithm optimization.

[0024] Compared with the prior art, the beneficial effects of the present invention are: 1. More comprehensive perception dimensions: For the first time, a forward-facing binocular camera and YOLO deep learning algorithm are introduced into the forklift anti-tipping system to achieve synchronous monitoring of the cargo tilt state and the forklift body state, which solves the defect of the existing system that "emphasizes the body and neglects the cargo" - by recognizing the cargo tilt angle in real time, the risk of center of gravity shift can be predicted in advance, improving the accuracy of the tipping warning.

[0025] 2. More precise control strategy: The design features a 5-level progressive control logic, with the control intensity strictly matched to the risk level. Low risk only triggers an alarm, medium risk is addressed through torque compensation, and high risk is addressed by braking / power cut-off. This avoids secondary risks (such as cargo falling) caused by the "one-size-fits-all" control of existing technologies, thus improving the safety of forklift operations.

[0026] 3. Strong compatibility and scalability: The system adopts a modular design. The sensing module supports the replacement of cameras with higher resolution ones, the data processing module supports version upgrades of the YOLO algorithm, and the hierarchical control logic can adjust the threshold through software configuration. It is suitable for various types of forklifts such as counterbalance forklifts, reach trucks, and stackers. Detailed Implementation

[0027] The technical solution of the present invention will be described in detail below with reference to the embodiments, but the scope of protection of the present invention is not limited to the embodiments described.

[0028] This embodiment addresses a routine cargo handling scenario in a large warehousing and logistics center. A CPD30 counterbalance electric forklift of a certain brand was selected, with a rated load capacity of 3 tons and a rated speed of 10 km / h. The operating environment was an indoor cement floor (friction coefficient μ=0.7) with no significant slope. The goods being handled were cardboard boxes loaded on standardized wooden pallets, with dimensions of 1.2m × 1.0m × 0.8m and a material density of 1800 kg / m³. 3 The weight of a single pallet is approximately 1728 kg. The operation process includes: pallet lifting → straight travel at a speed of 6 km / h → turning to avoid the passage pillars at a turning angle of 45° → unloading at a fixed point.

[0029] The core verification system demonstrates its effectiveness in preventing rollovers and controlling cargo during a typical risk scenario of "sudden tilting of cargo during turning".

[0030] I. System hardware configuration and installation, core hardware selection is shown in Table 1.

[0031] Table 1 Core Hardware Selection Results

[0032] Hardware installation location and calibration: 1. Forward-facing binocular camera: Fixed to the center of the top of the forklift mast using a custom stainless steel bracket, with the lens horizontally facing the cargo and at a vertical height of 1.8m above the top surface of the cargo, ensuring that the shooting range completely covers the entire cargo (without obstruction). After installation, the intrinsic parameter matrix is ​​calibrated using a Matlab calibration tool.

[0033] 2. Inertial sensor: It is fixed to the center of the forklift frame with bolts (0.8m above the ground) and rigidly connected to the frame. A shock-absorbing pad is installed at the bottom to reduce vibration interference.

[0034] 3. Accelerometer: Installed on the chassis at the bottom of the cab, it is kept on the same horizontal plane as the inertial sensor to ensure that the measurement reference is consistent.

[0035] 4. Steering angle sensor: Integrated at the end of the forklift steering shaft, it rotates synchronously with the steering shaft through a coupling to ensure that the steering angle is collected without delay.

[0036] 5. Data processing module and data storage module: fixed in a waterproof and explosion-proof box at the rear of the cab, it is connected to various sensors and controllers through a CAN2.0B bus interface with a bus transmission rate of 1Mbps.

[0037] 6. Instrument panel alarm unit: installed on the right side of the instrument panel in the driver's cab, ensuring that the operator can see it clearly and hear it clearly.

[0038] II. System Software Configuration 1) Core Algorithm and Parameter Settings: 1. YOLOv8 Algorithm: Trained using the PyTorch 2.0 framework, the training dataset contains over 15,000 images of forklift operations with different goods (cardboard boxes, wooden crates, metal parts) and different tilt angles (-30°~30°), covering common warehouse lighting conditions (natural light, LED lighting). The training iterations are 1000 rounds, the model weight file size is 28MB, and in actual operations, the goods tilt angle recognition latency is ≤25ms, with a recognition error ≤±0.3°.

[0039] 2. Kalman filter parameters: process noise variance s w2 =0.005, Observation noise variance s v 2 =0.02\), state transition matrix A=1, observation matrix H=1, used for filtering and smoothing of raw lateral acceleration data.

[0040] 3. Graded control parameters: Steering angle proportional coefficient k1=0.015N·m / °, tilt angle proportional coefficient k2=0.25N·m / °; forced deceleration time 0.9s, braking pressure adjustment range 0.8~1.5MPa; emergency braking response time ≤0.1s, stopping time ≤1.5s.

[0041] 4. Communication Protocol: The CAN2.0B bus protocol is adopted, with a data frame period of 10ms, to ensure real-time interaction of sensing data, processing results and control commands.

[0042] 2) Software operating environment: Operating System: Ubuntu 22.04LTS, 64-bit; Development Language: Python 3.9 (YOLOv8 algorithm), C++17 (dynamics calculation and bus communication); Dependencies: OpenCV 4.8 (image processing), PyTorch 2.0 (deep learning), Eigen 3.4 (numerical calculation), CANlib 5.4 (CAN bus communication).

[0043] III. The specific implementation process is as follows: 1. The system has completed hardware installation and online debugging. All modules communicate normally with no data packet loss (packet loss rate ≤ 0.1%).

[0044] 2. The goods are loaded smoothly onto the forklift, with the bottom of the pallet completely in contact with the forks, the initial tilt angle is 0°, and the forklift is in a horizontal, stationary position.

[0045] 3. The system completes initialization, the YOLOv8 algorithm model and dynamic analysis model are loaded, and the data storage module starts real-time recording.

[0046] Step S1: Dynamic operation monitoring and data collection: After the forklift starts, the sensing module continuously collects data at a preset frequency. When the forklift turns to avoid the column, the timestamp is t=5.2s, as follows: 1. Forward-facing binocular camera: Captures cargo images at a frame rate of 30fps, with a single frame image size of 2592×1944 and an interval of Δt≈0.033s between adjacent frames. At this time, the top surface of the cargo is significantly offset relative to the bottom surface.

[0047] 2. Inertial sensor: tilt angle θ = 10.5°, angular velocity = 8° / s.

[0048] 3. Accelerometer: The raw data for lateral acceleration is 1.32g (g=9.8m / s²).

[0049] 4. Steering angle sensor: Steering angle α = 45° (left turn, α > 0), steering speed = 30° / s.

[0050] All data is transmitted to the data processing module in real time via the CAN bus.

[0051] Step S2: Data fusion processing. After receiving the data, the data processing module performs image processing and dynamics calculations in parallel. The specific process is as follows: Step S21: Identify cargo tilt parameters based on the YOLOv8 algorithm. Step S211: Image distortion correction, eliminating radial and tangential distortion to obtain a corrected binocular image.

[0052] Step S212: Stereo matching calculation. The disparity map of the left and right eyes is obtained through the SGBM algorithm. Combined with the baseline distance (150mm) and focal length (12mm), the three-dimensional coordinates of 200+ feature points on the surface of the cargo are obtained. For example: the lower left corner of the bottom surface of the cargo (1.4m, -0.5m, 0.3m), and the upper right corner of the top surface (2.6m, 0.6m, 1.1m).

[0053] Step S213: YOLOv8 target detection accurately identifies a rectangle with a bottom boundary of 1.2m×1.0m and a rectangle with a top boundary of 1.2m×1.0m.

[0054] Step S214: Calculate the tilt parameters. Establish a three-dimensional coordinate system with the center of the bottom surface of the cargo as the origin (x-axis along the driving direction, y-axis perpendicular to the driving direction, z-axis vertical direction). Calculate the angle between the line connecting the top surface and the bottom surface edge in the yz plane. The cargo tilt angle θ = 12.3° is obtained, and the tilt direction is to the left.

[0055] Step S215: Calculate the tilt rate. Extract two adjacent frames of images, with the corresponding time t=5.2s and tilt angle θ1=11.8°, and the time t=5.233s and tilt angle θ2=12.3°. Calculate vθ≈15.15° / s using the formula vθ=|θ2-θ1| / Δt.

[0056] Step S22: Calculate forklift operating parameters based on the dynamic model. Step S221: Calculate the composite center of gravity. Given the forklift's own mass mv = 3000 kg, the position of the center of gravity is ( x v =1.6m, y v =0m, zv =1.3m); cargo mass m l =1728kg, center of gravity position ( x l =1.7m, y l =0.3m, z l =1.6m), substitute into the formula: xcg=(3000×1.6+1728×1.7) / (3000+1728)≈1.637m ycg=(3000×0+1728×0.3) / (3000+1728)≈0.109m zcg=(3000×1.3+1728×1.6) / (3000+1728)≈1.412m Step S222: Calculate the roll moment. After Kalman filtering, the raw lateral acceleration data yields ay = 1.30g. Substitute this into the roll dynamics equation: Mlat=mv×ay×hcgv+ml×ay×hcgl=3000×1.30×9.8×1.3+1728×1.30×9.8×1.6≈84580.56N·m Step S223: Calculate the lateral load transfer rate. Forklift wheel track B = 1.9m. Substitute into the formula: LTR=(2×Mlat) / [(mv+ml)×g×B]=(2×84580.56) / [(3000+1728)×9.8×1.9]≈1.92 Step S3: Risk Level Determination The data processing module integrates cargo tilt parameters (tilt angle 12.3°>10°) and forklift operating parameters (side tilt angle 10.5°∈8°~15°), and determines the current risk level as Level 2, medium to low risk, based on preset judgment rules.

[0057] Step S4: Hierarchical control execution The graded control module outputs the command "Apply reverse compensation torque to electronic steering system", and the execution module's response process is as follows: 1. The electronic steering controller receives commands and, based on the steering angle α = 45° (left turn) and the cargo tilt angle θ = 12.3° (tilt to the left), calculates the compensation torque using the formula T = k1|α| + k2|θ|: T = 0.015 × 45 + 0.25 × 12.3 = 3.75 N·m. Since the controller's maximum output torque is 2 N·m, the final output compensation torque to the right is T = 2 N·m, which is applied to the steering system in real time through PID closed-loop control to counteract part of the cargo tilting tendency caused by the centrifugal force of the left turn.

[0058] 2. The data storage module synchronously records the control command (timestamp t=5.24s, command type: reverse compensation torque, torque value 2N·m).

[0059] Step S5: Risk Removal and Status Restoration After applying the reverse compensation torque, the sensing module continuously collects data (sampling period 0.033s), and the changes are as follows: 1. At t=5.5s: the tilt angle of the goods decreases to 9.8°, the tilt rate decreases to 4.2° / s; the forklift side tilt angle decreases to 7.3°, and the lateral acceleration decreases to 0.9g.

[0060] 2. The data processing module re-determines the risk level as Level 1 (low risk), the graded control module outputs a "risk cleared" signal, the electronic steering controller returns to its default state (compensation torque returns to zero), and the instrument panel alarm unit activates an audible and visual alarm to remind the operator to adjust the cargo's posture.

[0061] 3. At t=6.8s: The operator manually adjusts the position of the goods, the tilt angle of the goods is reduced to 3.2°, the risk level is reduced to below 1, the instrument panel displays the normal operation indicator, and the forklift resumes normal operation.

[0062] Step S6: Data Storage and Export The data storage module completely records all data from this operation (total 12.8MB), including: ① Raw data: camera image frames (48 frames from 5.2s to 6.8s), sensor time-series data (tilt angle, acceleration, etc.); ② Processing results: cargo tilt parameters, composite center of gravity coordinates, tilt moment, etc.; ③ Control command records: reverse compensation moment commands, risk clearance signals. Data can be exported via Ethernet interface (format: CSV+JPG) for subsequent operation review and algorithm optimization.

[0063] Implementation effect verification: as shown in Table 2.

[0064] Table 2. Energy Index Test Results

[0065] The system described in this invention identifies the tilt state of the cargo (12.3°) using the YOLOv8 algorithm, predicting the risk of center of gravity shift 0.8 seconds in advance. This avoids the risk lag problem caused by ignoring the cargo status in existing technologies, significantly improving the accuracy of early warnings. Compared with Chinese patent CN114890345B: This system adopts hierarchical control (level 2 → level 1), using reverse compensation torque instead of aggressive braking intervention, preventing secondary tilting of the cargo, improving operational continuity by 40%, and effectively avoiding the secondary risks caused by the "one-size-fits-all" control of existing technologies.

[0066] If the tilt angle of the goods continues to increase during the above operations (reaching 18.7° at t=5.4s), the forklift tilt angle rises to 16.2°, and the lateral acceleration rises to 1.4g, the data processing module will determine the risk level as Level 3 (medium-high risk), and the brake controller will execute forced deceleration: the braking pressure will increase linearly from 0.8MPa to 1.5MPa, the deceleration time will be 0.9s, and the forklift speed will decrease from 6km / h to 2.8km / h (28% of the rated speed); if the tilt angle still exceeds 20° after deceleration, emergency braking will be triggered to ensure that the forklift stops within 1.5s until the risk is eliminated.

[0067] This embodiment, through complete hardware installation, software configuration, and dynamic operation process verification, fully demonstrates the feasibility and effectiveness of the multi-source sensing and dynamic detection active anti-tipping control system and method for forklifts. The system can accurately identify cargo tilt and forklift status, effectively suppressing the risk of tipping over through a hierarchical control strategy while ensuring operational continuity. It is suitable for complex operational scenarios such as warehousing and logistics centers and has significant engineering application value.

[0068] As described above, although the invention has been illustrated and described with reference to specific embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A forklift active roll-over prevention control system with multi-source perception and dynamic detection, characterized in that, include: The sensing module is used to collect image data of the goods carried by the forklift and the operating status data of the forklift. The sensing module includes a forward-facing binocular camera, an inertial sensor, an acceleration sensor, and a steering angle sensor installed on the top of the forklift mast. The forward-facing binocular camera is used to collect real-time image data of the loaded goods. The inertial sensor is used to collect the forklift's tilt angle and angular velocity data. The acceleration sensor is used to collect the forklift's lateral acceleration data. The steering angle sensor is used to collect the forklift's steering angle and steering speed data. The data processing module, electrically connected to the sensing module, is used to fuse and process the image data and forklift operating status data collected by the sensing module to determine whether the forklift is at risk of tipping over. The data processing module has a built-in preset YOLO algorithm and dynamic analysis model. It processes the image data using the YOLO algorithm to identify the tilt parameters of the goods, and calculates the forklift operating status data using the dynamic analysis model to obtain the forklift's center of gravity position, tilting moment, and lateral load transfer rate. The hierarchical control module, electrically connected to the data processing module, is used to output hierarchical control commands based on the forklift rollover risk assessment results. The hierarchical control module includes sequentially progressive control levels: instrument panel audible and visual alarm, application of reverse compensation torque of electronic steering system, forced deceleration of forklift, emergency braking of forklift, and cut-off of drive power. The execution module is electrically connected to the hierarchical control module and is used to execute hierarchical control commands; the execution module includes an instrument panel alarm unit, an electronic steering controller, a brake controller, and a power controller.

2. The multi-source perception and dynamic detection based active roll-over prevention control system for a forklift truck of claim 1, wherein, The data processing module includes an embedded processor for running the YOLO algorithm and dynamic analysis model in parallel, and for interacting with the sensing module and hierarchical control module via a CAN bus.

3. The multi-source perception and dynamic detection based active roll-over prevention control system for a forklift truck of claim 1, wherein, The YOLO algorithm is the YOLOv8 algorithm, and it is pre-trained using a forklift loading cargo dataset. The cargo tilting state parameters include cargo tilting angle, tilting direction, and tilting rate.

4. The multi-source perception and dynamic detection based active roll-over prevention control system for a forklift truck of claim 1, wherein, The control logic of the hierarchical control module is as follows: When only a cargo tilt angle of 5°~10° is detected, an audible and visual alarm command is output to the instrument panel. When the tilt angle of the goods exceeds 10° or the side tilt angle of the forklift is 8°~15°, a reverse compensation torque command is output. When the forklift tilt angle exceeds 15° or the lateral acceleration is 1.2g~1.5g, a forced deceleration command is output; An emergency braking command is issued when the forklift tilt angle exceeds 20° or the lateral load transfer rate exceeds 0.

8. When the forklift tilt angle still exceeds 25° after emergency braking, a drive power cut-off command is output.

5. The multi-source perception and dynamic detection based active roll-over prevention control system for a forklift truck of claim 1, wherein, It also includes a data storage module, which is electrically connected to the data processing module, and is used to store the raw data collected by the sensing module, the data processing results, and the hierarchical control instruction records, with a storage time of not less than 30 days.

6. A forklift active roll-over prevention control method based on multi-source perception and dynamic detection, characterized in that, Applied to the system according to any one of claims 1-5, the method includes the following steps: Step S1: Activate the perception module to collect real-time image data of the goods loaded on the forklift through the forward-facing binocular camera, and collect real-time operating status data of the forklift through the inertial sensor, acceleration sensor, and steering angle sensor. Step S2: The data processing module receives the image data and the real-time running state data of the forklift, processes the image data through the YOLOv8 algorithm to identify the cargo tilt angle, tilt direction and tilt rate, and processes the real-time running state data of the forklift through a dynamics analysis model to calculate the center of gravity position, roll moment, lateral acceleration and lateral load transfer rate of the forklift; Step S3: The data processing module determines the rollover risk level according to the cargo tilt parameters and forklift running parameters: Risk level 1: cargo tilt angle 5°~10°, no forklift body rollover risk; Risk level 2: cargo tilt angle >10° or forklift roll angle 8°~15°; Risk level 3: forklift roll angle >15° or lateral acceleration 1.2g~1.5g; Risk level 4: forklift roll angle >20° or lateral load transfer rate >0.8; Risk level 5: forklift roll angle >25° after emergency braking; Step S4: The hierarchical control module outputs corresponding control instructions according to the risk level, and the execution module executes the instructions: Corresponding risk level 1: the instrument panel alarm unit starts the audible and visual alarm; Corresponding risk level 2: the electronic steering controller applies a reverse compensation moment; Corresponding risk level 3: the brake controller executes forced deceleration to reduce the forklift speed to below 30% of the rated speed; Corresponding risk level 4: the brake controller executes emergency braking to stop the forklift within 1.5s; Corresponding risk level 5: the power controller cuts off the power output of the driving engine or motor; Step S5: After executing the instructions, the sensing module continuously collects data, the data processing module re-determines the risk level until the risk level is reduced to below 1, and the forklift returns to normal operating state.

7. The method of claim 6, wherein, In step S2, the data processing module processes the image data through the YOLOv8 algorithm to identify the cargo tilt angle and tilt direction, which includes: S211: Correcting the image data collected by the forward binocular camera, the distortion correction is based on the pre-set camera intrinsic matrix, eliminating the radial distortion and tangential distortion of the image, and obtaining the corrected binocular image; S212: Stereoscopic matching of the corrected binocular image, by calculating the disparity map of the left and right eye images, combining the baseline distance and focal length parameters of the binocular camera, and obtaining the three-dimensional coordinate data of the cargo surface feature points; S213: Calling the target detection branch of the YOLOv8 algorithm, identifying the contour features of the cargo in the three-dimensional coordinate data, and determining the boundary contour of the cargo bottom surface and the boundary contour of the top surface; S214: Establish a three-dimensional coordinate system O-XYZ with the center of the bottom surface of the cargo as the origin, wherein, x the axis is along the driving direction of the forklift, y the axis is perpendicular to the driving direction, z the axis is along the vertical direction; calculate the included angle between the connecting line of the top edge feature point and the bottom edge feature point of the cargo in the y-z plane, which is the cargo tilt angle; at the same time, according to the offset direction of the top surface profile relative to the bottom surface profile, determine the cargo tilt direction as left tilt, right tilt, forward tilt or backward tilt; S215: The forward binocular camera continuously collects cargo images at a frame rate of not less than 30fps, and the data processing module extracts the cargo tilt angles θ1 and θ2 corresponding to the adjacent two frames of images respectively; Computing the tilt rate v θ = |θ2-θ1| / Δt, unit: ° / s; Δt is the time interval between adjacent two frames, Δt = 1 / frame rate.

8. The method of claim 7, wherein, In step S2, the dynamics analysis model is used to process the real-time running state data of the forklift to calculate the center of gravity position, roll moment, lateral acceleration and lateral load transfer rate of the forklift and cargo, which specifically includes: S221: Based on the three-dimensional coordinate data of the cargo acquired by the forward-facing binocular camera, the coordinates of the geometric center of the cargo are taken as the position of the load center of gravity. x l , y l , z l And estimate the load mass based on the cargo's three-dimensional dimensions and material density. m l Forklift's own center of gravity position ( x v , y v , z v The value is a preset fixed value; the composite centroid is calculated using the following formula ( x cg , y cg , z cg ) position, where: x cg =( m v × x v + m l × x l ) / ( m v + m l )； y cg =( m v × y v + m l × y l ) / ( m v + m l )； z cg =( m v × z cgv + m l × x l ) / ( m v + m l )； wherein m v + m l total mass of the forklift truck m total ; S222: Based on the forklift truck overall vehicle mass, load mass, wheelbase, track parameters, establish the forklift truck roll dynamics equation: , wherein, M lat is the lateral roll moment, m v is the forklift's own mass, a y is the lateral acceleration, h cgv is the forklift's own center of gravity height, m l is the load mass, h cgl is the load center of gravity height; S223: Combined with the roll angle θ collected by the inertial sensor, the lateral load transfer rate is calculated: ， Where g is the acceleration of gravity, and B is the wheelbase.

9. The method of claim 8, wherein, In the step S222, the data processing module performs Kalman filtering on the lateral acceleration original data to obtain effective lateral acceleration a y Specifically, x k =A×x k-1 +K×u k-1 +w k-1 z k =H×x k +v k wherein, x k is k the filtered result at time instant t, i.e. the effective lateral acceleration a y , x k-1 is k the filtered result at time instant t-1, A is the state transition matrix and K is the control matrix, u k-1 is k the control quantity at time instant t-1, w k-1 is process noise, which is subject to a Gaussian distribution with mean 0 and variance z k is k the raw lateral acceleration data at time instant t, H is the observation matrix, v k is observation noise, which is subject to a Gaussian distribution with mean 0 and variance 10. The method of claim 9, wherein, In step S4, the size of the reverse compensation torque is dynamically adjusted according to the steering angle and the tilting direction: when the forklift is steering to the left and the goods are tilting to the left, the electronic steering controller applies a compensation torque to the right, and the size of the torque is positively correlated with the steering angle and the tilting angle of the goods; The size of the reverse compensation torque applied by the electronic steering controller is dynamically adjusted according to the steering angle of the forklift and the tilting direction of the goods, and specifically includes: Define the parameter steering angle α, set the left turn as positive, and the range is-180°~180°; define the goods tilting angle θ, set the left tilting as positive, and the range is-30°~30°, and the reverse compensation torque T; When α > 0 and θ > 0, the electronic steering controller applies a rightward compensating moment, satisfying the formula: T = - (K1 + K2) * α * θ k 1| α |+ k 2| θ |,wherein, k 1is a steering angle proportionality coefficient, k 2is a tilt angle proportionality coefficient; When a < 0 and 0 < 0, the electronic steering controller applies a compensating moment to the left, satisfying the formula: T = - (a + 0) * L k 1| α |+ k 2| θ |) When α With θ When the sign is opposite, the absolute value of the compensation torque T is calculated according to the above formula, with the direction opposite to the steering angle.

Images