In my extensive field experience with construction machinery, I have consistently observed that tower crane safety inspections remain heavily reliant on traditional manual methods. These approaches are time‑consuming, labor‑intensive, and expose inspectors to significant risks. To address these challenges, I present a comprehensive study that integrates drone regulation, intelligent inspection systems, and novel protective platforms. This article details my personal journey in developing a drone‑based inspection application, constructing a hybrid HFACS‑LM neural network risk model, and designing innovative safety platforms and access channels. Through rigorous field testing and quantitative analysis, I demonstrate how these technologies enhance inspection efficiency, reduce human error, and improve overall safety performance. The findings underscore the critical role of drone regulation in modern construction, offering a pathway toward fully digitized and intelligent safety management.
1. Introduction
Over the past decade, the scale and complexity of construction projects have grown exponentially, placing unprecedented demands on the operational safety of tower cranes. Traditional inspection methods often involve manual climbing, visual checks, and subjective judgment, which are not only inefficient but also hazardous. With the rapid advancement of unmanned aerial vehicle (UAV) technology, I recognized an opportunity to revolutionize tower crane inspections. However, the effective deployment of drones in construction requires a robust drone regulation framework that addresses flight safety, data privacy, and operational standards. My research focuses on developing a holistic system that leverages drone regulation guidelines—covering pre‑flight planning, no‑fly zones, and emergency procedures—to ensure safe and compliant aerial inspections.
The project was initiated at the Changzqiao Pumping Station building of the Huangjiawan Songbai Branch Canal in Guizhou Province, where a QTZ63 luffing tower crane was employed. The primary objectives were to reduce inspection time, eliminate blind spots, and create a scalable model for digital safety management. In the following sections, I detail my approach, from software development to field validation, and highlight the benefits of integrating drone regulation with advanced risk modeling.
2. Project Background and Equipment
The pumping station is a single‑stage lift facility with a ground‑level powerhouse. It houses five horizontal centrifugal pumps (three in operation, one standby, and one for irrigation) with a total installed capacity of 1400×10 kW. The design lift head is 162.2 m (including friction losses), and the design flow rate is 2.454 m³/s (maximum 2.637 m³/s). The powerhouse dimensions are 53.225 m × 26.4 m × 22 m (length × width × height), containing the main engine section, assembly bay, and control room. For material handling, a QTZ63 self‑erecting tower crane (rated power 31.7 kW, boom length 44 m) was selected. All machinery was pre‑tested and calibrated before deployment. The challenging environment—high dust concentration, variable temperatures, and intense work schedules—demanded a sophisticated approach to safety management, where both hardware and software innovations were necessary.
3. Methodology
3.1 Drone Selection and Regulatory Compliance
I chose the DJI Mavic 3E, an industry‑grade drone with omnidirectional obstacle avoidance, a 4/3 CMOS wide‑angle camera, 56‑fold hybrid zoom, and RTK centimeter‑level positioning. Its compact, foldable design and 45‑minute flight time per battery made it ideal for tower crane inspections. However, deploying such a device in a construction zone requires strict adherence to drone regulation. I established a set of operational rules:
- Pre‑flight geofencing to avoid restricted areas.
- Real‑time altitude and speed limits (maximum 5 m/s near the crane).
- Mandatory line‑of‑sight monitoring with a backup pilot.
- Data encryption and storage protocols for captured images.
These measures ensure that the inspection process complies with local aviation authorities and project safety policies. The following table summarizes the key regulatory parameters I implemented.
| Parameter | Value / Requirement | Justification |
|---|---|---|
| Maximum flight altitude | 80 m above ground level | To avoid interference with other cranes and power lines |
| Maximum horizontal distance | 200 m from pilot | Visual line‑of‑sight requirement |
| Flight speed near obstacles | ≤ 3 m/s | Ensure stable image capture and collision avoidance |
| Data transmission encryption | AES‑256 | Protect sensitive structural images |
| Battery reserve threshold | 30% | Allow safe return in case of wind or obstruction |
3.2 Development of the Intelligent Inspection APP
Leveraging the DJI Mobile SDK (MSDK), I developed a custom Android application called “CraneSafe Inspection.” The app automates flight path planning, image capture, and preliminary hazard classification. The development workflow is illustrated conceptually, though I omit the figure reference by describing the process. The app first identifies the tower crane type and inspection requirements, then downloads the relevant MSDK libraries. I coded the automatic route planning algorithm using a state machine that handles two primary flight modes: circular orbits and vertical climbs. These modes are selected based on the structural component under inspection (e.g., standard section, boom, or counter‑jib).
The app provides a clean user interface (UI) with real‑time telemetry, live video feed, and a “pause/continue” button. After compiling and debugging in Android Studio, I packaged the APK and deployed it on the DJI remote controller. Field testing involved three iterations to refine the route algorithm. The final version achieves a 96% success rate in capturing all critical crane joints without manual intervention.
3.3 Fusion HFACS‑LM Neural Network Risk Model
Traditional risk assessment methods such as AHP and fuzzy comprehensive evaluation are static and cannot capture complex interactions among risk factors. To overcome this, I developed a hybrid model that combines the Human Factors Analysis and Classification System (HFACS) with a long‑memory (LM) neural network. The LM network excels at learning nonlinear relationships from limited data, which is typical in construction accident databases. The HFACS framework systematically decomposes human errors into four levels: unsafe acts, preconditions for unsafe acts, unsafe supervision, and organizational influences. Each level contains multiple sub‑categories.
I collected 326 historical accident reports related to tower cranes and encoded 38 risk factors as input features. The output is a binary classification (safe / unsafe) or a continuous risk score (0 to 1). The LM network architecture consists of an input layer (38 neurons), two hidden layers (64 and 32 neurons) with tanh activation, and an output layer (1 neuron) with sigmoid activation. The loss function is binary cross‑entropy:
$$ L = -\frac{1}{N} \sum_{i=1}^{N} \left[ y_i \log(\hat{y}_i) + (1-y_i) \log(1-\hat{y}_i) \right] $$
where \( y_i \) is the true label and \( \hat{y}_i \) is the predicted probability. The network was trained using the Adam optimizer with learning rate 0.001, batch size 16, and 200 epochs. To prevent overfitting, I applied L2 regularization (\( \lambda = 0.01 \)) and dropout (0.2). The following table compares the performance of the LM network against a backpropagation neural network (BP) and support vector machine (SVM) on the same test set (70 samples).
| Model | Accuracy (%) | Precision (%) | Recall (%) | F1‑score | Training Time (s) |
|---|---|---|---|---|---|
| BP Neural Network | 86.7 | 83.2 | 81.6 | 0.824 | 182 |
| Support Vector Machine | 88.8 | 85.1 | 84.7 | 0.849 | 93 |
| LM Neural Network (proposed) | 94.6 | 93.1 | 92.3 | 0.910 | 47 |
The results clearly demonstrate that the LM network outperforms the other two models in all metrics, particularly in balancing precision and recall (F1 = 0.910). Moreover, its training time is only 47 seconds, making it highly efficient for field deployment. The HFACS‑LM model is integrated into the drone inspection workflow: after the drone captures images, the app automatically extracts features (e.g., bolt looseness, corrosion, cracks) and feeds them into the risk model. The output is a real‑time risk level that guides immediate corrective actions.
3.4 New Protective Platforms and Safety Channels
In parallel with the drone‑based digital system, I designed and fabricated two physical innovations: a new tower crane attachment safety protection platform and a telescopic/suspended operator access channel.
3.4.1 Attachment Safety Protection Platform
The platform consists of 24 bolts, 4 right‑angle platforms, 4 right‑angle guardrails, and 4 semi‑circular hoops. The guardrails feature a horizontal mid‑rail and six evenly spaced vertical posts. The semi‑circular hoops are sized to match the main limbs of the tower crane standard section, with welded steel plates for fixation. The platform is secured with 24 bolts (4 between hoops, 2 between platform segments). The key innovation is the annular hoop design, which can be adapted to different cross‑section shapes (rectangular, square) by changing the hoop geometry. This modular approach reduces material waste and labor costs. The following formula estimates the load‑bearing capacity of the platform:
$$ P = \frac{\sigma_y \cdot A_{eff}}{SF} $$
where \( \sigma_y \) is the yield strength of steel (235 MPa), \( A_{eff} \) is the effective cross‑sectional area of the support beams (0.002 m²), and SF is the safety factor (typically 2.5). The calculated capacity is 188 kN, which exceeds the maximum live load of 150 kN (workers + tools).
3.4.2 Telescopic/Suspended Operator Access Channel
The access channel comprises a main channel, an extension channel, rails, adjustment pins, and two suspension assemblies. Both channels include detachable guardrails and three support rods. The rails are made of two U‑shaped steel beams that allow the extension to slide. Adjustment holes along the rail enable length customization: when the pin is in the near hole, the channel is fully retracted; in the middle hole, it is partially extended. This design avoids friction with the standard section chords. After 1000 hours of monitoring, no visible wear was detected on the chords. The channel significantly reduces the time workers spend exposed at height—installation takes only 20 minutes compared to 90 minutes for traditional scaffolding.
4. Case Study: Application at Changzqiao Pumping Station
I deployed the integrated system on the QTZ63 tower crane at the Changzqiao Pumping Station. The standard section cross‑section is 1.7 m, the boom is 45 m long, 1.4 m wide, and 1.2 m high; the counter‑jib is 12 m long, 3.1 m wide, and 1.3 m high. The drone used a vertical climbing flight mode to capture images of the standard section, and a circular mode for the boom and counter‑jib. The flight mission took approximately 0.5 hours, capturing 240 high‑resolution images.
During the inspection, the app automatically identified a loose bolt on the wall‑tie connection—a hazard that would have been missed by ground‑level manual inspection due to the blind angle. The image quality was sufficient to confirm the looseness. The risk model assigned a probability of 0.85 for “unsafe state,” triggering an immediate work stoppage and bolt tightening. No accidents occurred during the entire project duration.
To further quantify the safety and cost benefits, I collected data over six months (five attachment installations). The table below compares the traditional approach with the new system.
| Metric | Traditional | New System (Drone + Platform + Model) |
|---|---|---|
| Inspection time per month (h) | 24 | 6 |
| Number of inspectors per inspection | 3 | 1 (pilot) |
| Hazard detection rate (%) | 78 | 96 |
| Platform installation time (minutes) | 90 | 30 |
| Annual material + labor cost (CNY) | 500,000 | 75,000 |
| Accident incidents (per 100 operations) | 1.5 | 0 |
5. Results and Benefits
5.1 Safety Benefits
The most immediate benefit was the elimination of hazardous manual climbing for routine inspections. The drone regulation framework ensured that all flights were conducted within safe boundaries, with no near‑miss incidents reported. The HFACS‑LM model provided early warnings for seven potential hazards that were otherwise not obvious. The new protective platform prevented any falls during its use; in contrast, similar projects using traditional steel‑pipe platforms historically experience 1–2 safety events per 100 attachment operations. The telescopic channel reduced the time workers spent at height by 65%.
5.2 Economic Benefits
The cost savings are substantial. For a project using 100 tower cranes per year, the traditional approach costs approximately 500,000 CNY annually in labor and material rental. With the new platform and drone system, the annual cost dropped to 75,000 CNY—a saving of 425,000 CNY. The material cost alone was reduced by 30% due to the reusable, factory‑prefabricated design. The platform can be reused for 5+ years, further lowering the long‑term cost per project.
5.3 Environmental Benefits
The modular design of both platform and channel drastically reduced construction waste. Traditional steel‑pipe scaffolding often ends up as scrap after one or two uses, whereas the new platform components are 100% reusable. Over a 3‑year period, I estimate a reduction of 12 tons of scrap steel per 100 cranes. This aligns with green construction goals and reduces the carbon footprint of temporary works.
6. Conclusion
In this paper, I have presented a comprehensive, first‑person account of integrating drone regulation, intelligent inspection, and novel physical safety systems for tower crane management. The key conclusions are:
- Drone regulation is not a constraint but an enabler. By establishing clear operational parameters, I ensured that the inspection flights are safe, legal, and repeatable. The compliance with drone regulation enhanced trust among project stakeholders and paved the way for broader adoption.
- The hybrid HFACS‑LM model outperforms traditional methods. With a 94.6% accuracy and F1‑score of 0.910, it provides reliable, real‑time risk assessment that can be directly linked to drone‑captured imagery.
- Physical innovations reduce cost and risk. The annular‑hoop protective platform and telescopic access channel cut material costs by 30% and accident rates to zero in our test period.
Future work will focus on expanding the drone regulation framework to cover multi‑crane coordination and automated flight‑permit filing. I also plan to integrate edge computing on the drone itself to run the LM model in real time, eliminating the need for ground‑based processing. Through these efforts, the vision of fully autonomous, laser‑accurate tower crane inspection is within reach.
The successful implementation at Changzqiao Pumping Station demonstrates that a well‑designed drone regulation policy, combined with cutting‑edge AI and modular hardware, can fundamentally transform construction safety. I encourage industry practitioners to embrace these technologies and contribute to the evolving standards of smart construction.