Device and method for monitoring blockage state of urban gutter inlet

Abstract

The invention discloses a device and method for monitoring the blocking state of an urban gutter inlet. The device comprises a data processing unit, a waterproof microphone, a humidity sensor, a liquid level sensor, an LED state indicator lamp, a protective shell and a data transmission interface, and the data processing unit is connected with the humidity sensor, the waterproof microphone and the liquid level sensor. The method comprises the steps of training a blockage judgment model of sound features, installing and setting a monitoring device, judging and processing blockage, early warning accumulated water, resetting dormancy and the like. According to the invention, the defects of low labor efficiency, high misjudgment of a single sensor and large interference of light environment on vision in the existing gutter inlet blockage monitoring are overcome.

Description

A device and method for monitoring the blockage status of urban storm drains Technical Field

[0001] This invention belongs to the field of urban municipal monitoring technology, specifically referring to a device and method for monitoring the blockage status of urban storm drains. Background Technology

[0002] Traditional methods for detecting blockages in urban storm drains are mainly manual, using pipe endoscopes or sonar for monitoring. This manual method is time-consuming, labor-intensive, and inefficient. Furthermore, due to the large size of cities, blockages cannot be detected in time, missing the best opportunity for dredging.

[0003] In addition, a single liquid level sensor is also a common monitoring method. This method is limited by environmental changes. When it is not raining, road watering, car wash wastewater, etc. may cause the liquid level to rise, which may be misjudged as a blockage. When it rains, if the drain is temporarily blocked by debris (such as fallen leaves), it is easy to trigger a false alarm. An even more prominent problem is that this method cannot work in the early stage of a blockage when water has not yet accumulated.

[0004] In addition, there are methods to determine the congestion situation based on video images, but such methods may have problems such as insufficient image clarity or blind spots at night or in areas with complex structures. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing storm drain blockage monitoring, such as low manual efficiency, high misjudgment rate of single sensors, and visual interference from lighting conditions. It provides a device and method for monitoring the blockage status of urban storm drains, thereby achieving efficient monitoring of storm drain blockage.

[0006] To achieve the above objectives, the present invention provides a device for monitoring the blockage status of urban storm drains, comprising a data processing unit, a waterproof microphone, a humidity sensor, a liquid level sensor, an LED status indicator, a protective housing, and a data transmission interface. The data processing unit is connected to the humidity sensor, the waterproof microphone, and the liquid level sensor respectively. The liquid level sensor, the waterproof microphone, and the humidity sensor are arranged side by side above the storm drain and are all mounted above the storm drain via brackets. The LED status indicator is connected to the data processing unit. A drainage outlet is provided at the bottom of the protective housing, and screw holes are provided on the edge of the protective housing for connection to the bracket. The data transmission interface remotely transmits the collected signals to an external data processing device.

[0007] As a further aspect of the present invention: the humidity sensor is a capacitive humidity sensor with an IP68 protection rating; the liquid level sensor is an ultrasonic liquid level sensor, the protective housing is made of ABS engineering plastic, and the overall protection rating is IP67; the bracket is made of stainless steel.

[0008] A method for monitoring the blockage status of urban storm drains includes the following steps: Step 1: Training a blockage discrimination model based on sound characteristics; Step 2: Installing and setting up the monitoring device; Step 3: Blockage judgment and handling; Step 4: Water accumulation early warning; Step 5: Dormant reset.

[0009] As a further aspect of the present invention: Step one includes the following steps: 1.1 Constructing an experimental platform indoors or utilizing existing outdoor rainwater inlets, and installing the device above the rainwater inlets; 1.2 Setting up the rainwater inlets for both unobstructed and blocked conditions; 1.3 Under simulated indoor or outdoor rainfall conditions or real outdoor rainfall conditions, activating humidity, liquid level, and rainwater inlet sound data acquisition, setting up unobstructed and blocked conditions, and conducting multiple sets of repeated experiments to obtain a large amount of data; 1.4 Reading the data and preprocessing it, using a bandpass filter to filter out noise from the acquisition device itself and high-frequency noise; 1.5 Extracting various acoustic signal features from the preprocessed audio frames, and performing wavelet transform or fast Fourier transform on the signals to obtain... 1.6 Obtain the frequency domain features of the signal spectrum, including the main frequency, spectral energy, bandwidth, and centroid frequency; 1.7 Standardize the extracted features to ensure consistent feature scale; Divide the collected data into training and testing sets, using 70% as the training set and 30% as the testing set, and select a linear SVM binary classification model as the core model; 1.8 Use the processed feature data as input and the storm drain blockage status as the target output, evaluate the machine learning model using metrics such as accuracy, precision, and recall, and evaluate and analyze the features involved in the binary classification model using a decision tree model; 1.9 Embed the final optimal model into the data processing unit or install it on an external data processing device.

[0010] As a further aspect of the present invention: Step two includes the following steps: The device is installed near the rainwater inlet in an area prone to urban flooding. A humidity sensor is used to collect ambient humidity data in real time. When it is detected that "the humidity rises from ≤30% RH to ≥80% RH within 1 minute and lasts for ≥2 minutes", it is determined to be a "rainfall scenario". The LED indicator light is in an orange warning state. At the same time, the waterproof microphone and liquid level sensor are activated. The waterproof microphone collects the sound signal from the rainwater inlet at a preset sampling rate. The liquid level sensor simultaneously collects the data on the change in water depth on the road surface around the rainwater inlet. The data is then fed into the data processing unit for further judgment.

[0011] As a further aspect of the present invention: Step three includes the following steps: using an on-site data processing unit or an external data processing device, combined with the aforementioned training model, to determine the blockage status of the rainwater inlet in real time; if the system identifies it as a blockage, it issues a blockage warning message and notifies relevant maintenance personnel to clean the rainwater inlet in a timely manner.

[0012] As a further aspect of the present invention: in step four, when the system identifies a blockage and the water depth reaches 15cm, it further issues a water accumulation warning.

[0013] As a further aspect of the present invention: In step five, when the humidity sensor detects that "the humidity has decreased to 30%RH for 1 hour", it determines that "the rainfall has ended", controls the microphone and liquid level sensor to return to sleep mode, and only retains the humidity sensor for low-power operation.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: 1) Combining multi-sensor fusion with machine learning improves monitoring accuracy and reduces false alarm rate: By fusing data from microphones, humidity sensors, and liquid level sensors, and combining this with machine learning models for accurate discrimination of acoustic features, the invention effectively avoids the problem of false alarms caused by single sensors, such as false alarms caused by liquid level rise due to road watering during non-rainfall periods or false alarms caused by temporary obstruction by fallen leaves during rainfall. 2) This invention utilizes the sound characteristics of different blockage states of rainwater inlets. Compared with image sensors, this invention has a significant advantage in low-light areas at night, as the sound is not affected by brightness. Simultaneously, the processing center can perform noise reduction on the sound data, improving accuracy. 3) Low-power design extends device battery life: The humidity sensor enables accurate triggering of rain scenarios. During non-rainfall periods, only the humidity sensor operates with low power, while the microphone and liquid level sensor remain in sleep mode, significantly reducing energy consumption and solving the problem of frequent battery replacements in outdoor scenarios without mains power, thus reducing maintenance costs. 4) High real-time performance and timely early warning: It can collect and process data in real time and determine the blockage status, issuing early warnings at the initial stage of blockage, facilitating timely dredging by maintenance personnel and preventing traffic congestion, flooding, and other problems caused by road surface water accumulation, thus avoiding missing the best dredging opportunity. 5) Non-intrusive monitoring and convenient installation and maintenance: The device is installed around the rainwater inlet without damaging the original drainage facilities. Its compact structure adapts to rainwater inlets of different sizes, has a high protection level, strong anti-interference ability, and low installation and maintenance costs. 6) Comprehensive feature extraction and reliable judgment: By extracting the time domain, frequency domain, and time-frequency domain features of the sound signal, it characterizes the rainwater inlet status from multiple dimensions and trains a machine learning model, further improving the reliability and accuracy of blockage judgment. Attached Figure Description

[0015] Figure 1 is a schematic diagram of the device structure in this invention.

[0016] Figure 2 is a signal diagram of the signal denoising process in an embodiment of the present invention, where (a) is the original time-domain signal diagram and (b) is the time-domain signal diagram after denoising.

[0017] Figure 3 is a signal diagram of the embodiment of the present invention under the condition of unobstructed rainwater inlet, wherein (a) is a time domain signal diagram and (b) is a spectrum signal diagram.

[0018] Figure 4 is a signal diagram of the embodiment of the present invention under the condition of blocked rainwater inlet, wherein (a) is a time domain signal diagram and (b) is a spectrum signal diagram.

[0019] Figure 5 is a visualization of the SVM decision boundary in an embodiment of the present invention.

[0020] Figure 6 is a location map in the feature space where the model in the embodiment of the present invention predicts "smooth".

[0021] In the diagram: 1. Waterproof microphone, 2. Humidity sensor, 3. Liquid level sensor, 4. LED status indicator, 5. Data processing unit, 6. Protective housing, 7. Screw hole, 8. Data transmission interface. Detailed Implementation

[0022] The invention will now be further described with reference to the accompanying drawings.

[0023] As shown in Figure 1, a device for monitoring the blockage status of urban storm drains includes a data processing unit 5, a waterproof microphone 1, a humidity sensor 2, a liquid level sensor 3, an LED status indicator 4, a protective housing 6, and a data transmission interface 8. The data processing unit 5 is connected to the humidity sensor 2, the waterproof microphone 1, and the liquid level sensor 3, respectively, and is used to receive sensor data and execute judgment algorithms. The waterproof microphone 1 is fixed above the storm drain via a mounting bracket and is used to collect the sound signal of water flow inside the storm drain. The humidity sensor 2 is a capacitive humidity sensor with an IP68 protection rating, which is mounted above the storm drain via a bracket and is used to detect changes in ambient humidity to trigger data acquisition. The system includes a start and end point; the liquid level sensor 3 is an ultrasonic liquid level sensor, mounted above the rainwater inlet via a bracket, used to detect the water depth around the rainwater inlet after blockage occurs; the LED status indicator 4 is connected to the data processing unit 5 to indicate whether the device is in working condition; the protective housing 6 is made of ABS engineering plastic with an overall protection rating of IP67, and a drainage outlet is provided at the bottom of the housing to prevent internal water accumulation; the protective housing 6 has screw holes 7 on its edge, through which it can be connected to the bracket, which is made of stainless steel and uses a waterproof sealing ring, and is fixed to the road surface or grass around the rainwater inlet with expansion screws to ensure the stability of the sensor and prevent damage from environmental influences. The data transmission interface 8 remotely transmits the collected signals to an external data processing device, which can be installed in a roadside distribution box, the central control room of a street light pole, or a cloud data center.

[0024] The method for monitoring the blockage status of urban storm drains includes the following steps: Step 1: Training of a blockage discrimination model based on sound features: 1.1 Build an experimental platform indoors or use an existing outdoor storm drain and install the device above the storm drain.

[0025] 1.2 The rainwater inlet was tested under both unobstructed and blocked conditions. Under unobstructed conditions, there were no obstructions. Under blocked conditions, different obstructions such as leaves and debris were used at the rainwater inlet to simulate the real situation.

[0026] 1.3 Under simulated indoor and outdoor conditions or real outdoor rainfall, activate the data acquisition for humidity, liquid level, and rainwater inlet sound. Set up unobstructed and blocked operating conditions, and conduct multiple sets of repeated experiments to obtain a large amount of data.

[0027] 1.4 Read the data and preprocess it. Use the bandpass filter of the data processing unit to filter out the noise of the acquisition device itself and high-frequency noise. Cut the long audio signal into short frames, each frame is 20-40 milliseconds, and add 50% overlap to facilitate frame-by-frame analysis. Use Hamming window to process each frame signal to reduce spectral leakage, remove invalid values ​​and outliers, align time, and unify data format.

[0028] 1.5 Extract various acoustic signal features from the preprocessed audio frames, such as the time-domain features of maximum amplitude, RMS, standard deviation, and peak factor. After performing wavelet transform or fast Fourier transform on the signal, obtain the signal spectrum and acquire the frequency-domain features of the dominant frequency, spectral energy, bandwidth, and centroid frequency.

[0029] 1.6 The extracted features were standardized to ensure consistent feature scale. The collected data was divided into training and test sets. Due to the limited amount of data obtained from indoor experiments, for datasets with small sample sizes, a typical approach was to use 70% as the training set and 30% as the test set. To ensure the reliability of small-sample training, provide clear decision boundaries, and avoid overfitting, a linear SVM binary classification model was selected as the core model.

[0030] 1.7 Using the processed feature data as input and the storm drain blockage status as the target output, the machine learning model is evaluated using the metrics of accuracy, precision, and recall. The decision tree model can be used to evaluate and analyze the features involved in the binary classification model.

[0031] 1.8 Embed the final optimal model into the data processing unit or install it on an external data processing device.

[0032] Step Two: Installation and Setup of the Monitoring Device: Install the device near the storm drains in flood-prone areas of the city. A humidity sensor collects real-time ambient humidity data. When the humidity rises from ≤30% RH to ≥80% RH within one minute, and the duration is ≥2 minutes, it is determined to be a "rainfall scenario," and the LED indicator turns orange as a warning. Simultaneously, the microphone and liquid level sensor are activated. The microphone collects sound signals from the storm drains at a preset sampling rate; the liquid level sensor simultaneously collects data on changes in the depth of water accumulation on the surrounding road surface. The data is then processed by the data processing unit for further analysis.

[0033] Step 3: Blockage Judgment and Handling: Using the on-site data processing unit or external data processing equipment, combined with the aforementioned training model, the blockage status of the rainwater inlet is judged in real time. If the system identifies a blockage, a blockage warning message is issued to notify relevant maintenance personnel to clean the rainwater inlet in a timely manner.

[0034] Step 4: Flood warning. When the system identifies a blockage and the water depth reaches 15cm, it will issue a flood warning.

[0035] Step 5: Sleep Reset. To further save energy, when the humidity sensor detects that "humidity has decreased to 30% RH for 1 hour", it determines that "rainfall has ended" and controls the microphone and liquid level sensor to return to sleep mode, keeping only the humidity sensor in low-power operation.

[0036] Example:

[0037] S1. Training of the blockage discrimination model based on sound features: A large amount of raw sound signal data was collected using an indoor test platform or in outdoor rainy weather. Unobstructed conditions were used to simulate smooth operation, while different obstructions, such as leaves, wooden boards and other debris, were used at the rainwater inlet to simulate blockage conditions. Finally, 70 sets of sound signals under smooth water flow conditions and 70 sets of sound signals under blocked outlet conditions were obtained.

[0038] The data is read and preprocessed, using a bandpass filter to remove noise from the acquisition device itself and high-frequency noise; the long audio signal is divided into short frames, each 20-40 milliseconds long, with 50% overlap for frame-by-frame analysis; a Hamming window is used to process each frame to reduce spectral leakage, remove invalid and outlier values, align time, and standardize the data format. See Figure 2.

[0039] After preprocessing, a Fast Fourier Transform (FFT) is performed on the signal. Under unobstructed conditions, its spectral characteristics typically show strong energy concentration in the low-frequency range, while high-frequency components are relatively few. The time-domain waveform of the signal is relatively regular, without sudden amplitude changes or pulses. However, under blocked conditions, the sound signal will significantly deviate from the spectral characteristics of normal operation, with a significant increase in high-frequency energy. The time-domain waveform will exhibit violent and irregular fluctuations. Sound features such as the maximum amplitude and dominant frequency are extracted from the signal to form a feature vector. The extracted features are standardized to ensure consistent feature scale. The processed feature data is used as input, and the "unobstructed" and "blocked" states of the drain outlet are used as the target output. The sound signal diagrams are shown in Figures 3 and 4.

[0040] The dataset was divided into training, validation, and test sets. The model was trained on the training set and its performance was evaluated on the validation set using the RBF kernel. The model was optimized by adjusting hyperparameters C=1.0 and γ=0.1. Finally, the generalization performance was evaluated on the test set. Evaluation metrics included accuracy, precision, recall, and F1 score. In this scenario, we are more concerned with identifying "blocked" states, aiming to minimize false negatives while controlling the false positive rate. The model primarily used cross-validation, employing a 5-fold cross-validation method to split the dataset. The training and validation sets comprised 85% of the total data (119 samples), and the test set comprised 15% (21 samples). The final trained model achieved an accuracy of 95.83%, precision of 90.00%, recall of 100.00%, and an F1 score of 0.92. The trained model was then embedded into the data processing unit, as shown in Figure 5.

[0041] S2. Installation and setup of the detection device: Select the storm drain inlet of the city road for installation. Fix the device to the ground with a bracket. Place the waterproof MEMS microphone above the storm drain grate and connect it to a Picoscope oscilloscope to collect sound signals. At the same time, install a capacitive humidity sensor and an ultrasonic liquid level sensor with IP68 protection rating. All sensors are connected to the data processing unit. The external enclosure is equipped with an ABS engineering plastic protective shell (IP67 protection rating, with a drain outlet at the bottom).

[0042] When outdoor rainfall begins, the humidity sensor detects that the humidity rises from 25% RH to 80% RH within 1 minute. After 2 minutes, the embedded microcontroller (MCU) of the data processing unit determines it as a "rainfall scenario" and triggers the microphone and liquid level sensor to start. The MEMS microphone records the sound information of rain in real time, the liquid level sensor transmits the water depth information in real time, and the LED indicator turns orange.

[0043] S3. Data Acquisition and Blockage Assessment: During rainfall, the microphone continuously collects sound signals from inside the drain outlet at a sampling frequency of 10kHz. The data processing unit uses digital signal processing software algorithms to perform bandpass filtering on the raw signal, removing noise below 100Hz and above 8kHz, dividing it into frames with a frame length of 30ms, 50% overlap, and applying a Hamming window. The collected sound signals are input into a trained SVM model, which quickly outputs prediction results. In real blockage situations, the model's prediction of "blockage" achieves a confidence level of 92.87%.

[0044] S4. Early Warning and Response: By combining the device with edge computing equipment such as AI chips, when the model determines "blockage" based on the sound of water flow, the LED indicator turns red, and an early warning message is sent to the monitoring center in the cloud via the wireless communication module. After receiving the early warning, maintenance personnel can take timely measures, go to the site to clear the blockage, and prevent flooding in advance, based on the water depth information provided by the liquid level sensor. Figure 6 shows that through a large number of experiments, the model has obtained a large number of characteristic spatial locations under the conditions of unobstructed and blocked rainwater inlets. When a new signal is located at this location, it can be directly determined whether it is unobstructed or blocked.

[0045] S5. Rainfall End and Sleep Mode: After the rainfall ends, the humidity sensor detects that the humidity drops to 35%RH for 1 hour. The data processing unit controls the microphone and liquid level sensor to enter sleep mode, the LED indicator turns green, and only the humidity sensor and communication unit operate at low power.

Claims

1. A device for monitoring the blockage status of urban storm drains, characterized in that, Includes a data processing unit (5), a waterproof microphone (1), a humidity sensor (2), a liquid level sensor (3), an LED status indicator (4), a protective housing (6), and a data transmission interface (8). The data processing unit (5) is connected to the humidity sensor (2), the waterproof microphone (1), and the liquid level sensor (3) respectively. The liquid level sensor (3), the waterproof microphone (1), and the humidity sensor (2) are arranged side by side above the rainwater inlet and are all installed above the rainwater inlet by a bracket. The LED status indicator (4) is connected to the data processing unit (5). A drain outlet is opened at the bottom of the protective housing (6). The edge of the protective housing (6) is provided with screw holes (7) and is connected to the bracket through the screw holes (7). The data transmission interface (8) remotely transmits the collected signals to an external data processing device.

2. The device for monitoring the blockage status of urban storm drains according to claim 1, characterized in that, The humidity sensor (2) is a capacitive humidity sensor with IP68 protection rating; the liquid level sensor (3) is an ultrasonic liquid level sensor; the protective housing (6) is made of ABS engineering plastic with an overall protection rating of IP67; and the bracket is made of stainless steel.

3. The method for monitoring the blockage status of urban storm drains according to claim 2, characterized in that, Includes the following steps: Step Step 1: Training a congestion discrimination model based on sound features; Step 2: Installing and setting up monitoring devices; Step 3: Congestion judgment and handling; Step 4: Water accumulation early warning; Step 5: Dormant reset.

4. The method according to claim 3, characterized in that, Step one includes the following steps: 1.1 Constructing an indoor experimental platform or utilizing existing outdoor rainwater inlets, and installing the device above the rainwater inlets; 1.2 Setting up experimental conditions at the rainwater inlets for both unobstructed and blocked conditions; 1.3 Under simulated indoor or outdoor rainfall conditions or real outdoor rainfall conditions, activating humidity, liquid level, and rainwater inlet sound data acquisition, setting up unobstructed and blocked conditions, and conducting multiple sets of repeated experiments to obtain a large amount of data; 1.4 Reading the data and preprocessing it, using a bandpass filter to filter out noise from the acquisition device itself and high-frequency noise; 1.5 Extracting various acoustic signal features from the preprocessed audio frames, performing wavelet transform or fast Fourier transform on the signals to obtain the signal spectrum, and obtaining... 1.6 Extract frequency domain features including main frequency, spectral energy, bandwidth, and centroid frequency; 1.7 Standardize the extracted features to ensure consistent feature scale; Divide the collected data into training and testing sets, using 70% as the training set and 30% as the testing set, and select a linear SVM binary classification model as the core model; 1.8 Use the processed feature data as input and the storm drain blockage status as the target output, evaluate the machine learning model using accuracy, precision, and recall metrics, and evaluate and analyze the features involved in the binary classification model using a decision tree model; 1.9 Embed the final optimal model into the data processing unit or install it on an external data processing device.

5. The method according to claim 3, characterized in that, Step two includes the following steps: The device is installed near the rainwater inlet in an area prone to urban flooding. A humidity sensor is used to collect ambient humidity data in real time. When it is detected that "the humidity rises from ≤30%RH to ≥80%RH within 1 minute and lasts for ≥2 minutes", it is determined to be a "rainfall scenario". The LED indicator light is in an orange warning state. At the same time, the waterproof microphone and liquid level sensor are triggered to start. The waterproof microphone collects the sound signal of the rainwater inlet at a preset sampling rate. The liquid level sensor simultaneously collects the data on the change of water depth on the road surface around the rainwater inlet. The data is then fed into the data processing unit for further judgment.

6. The method according to claim 3, characterized in that, Step three includes the following steps: using the on-site data processing unit or external data processing equipment, combined with the aforementioned training model, to determine the blockage status of the rainwater inlet in real time. If the system identifies a blockage, it issues a blockage warning message and notifies relevant maintenance personnel to clean the rainwater inlet in a timely manner.

7. The method according to claim 3, characterized in that, In step four, when the system identifies a blockage and the water depth reaches 15cm, a water accumulation warning is issued.

8. The method according to claim 3, characterized in that, In step five, when the humidity sensor detects that "humidity has decreased to 30%RH for 1 hour", it determines that "rainfall has ended", controls the microphone and liquid level sensor to return to sleep mode, and only keeps the humidity sensor in low power operation.

Images