Electric drill with abnormity alarm function and alarm method thereof

Abstract

The invention belongs to the technical field of electric drills, and provides an electric drill with an abnormal alarm function and an alarm method thereof. The method comprises the following steps: constructing and obtaining a first multi-dimensional feature vector based on torque fluctuation data, current fluctuation data and vibration data associated with a drill bit, and predicting and obtaining an abnormal probability by using a hierarchical prediction mechanism based on the first multi-dimensional feature vector; and when the abnormal probability is higher than a probability threshold value, outputting a hard foreign matter alarm signal. On one hand, real-time early warning of hard foreign matter contact during drilling of the electric drill is achieved by fusing data of the multi-dimensional sensor, damage such as tipping and breakage caused by collision of a drill bit with foreign matter is effectively avoided, the tool maintenance and replacement cost is reduced, the operation continuity is guaranteed, and the use safety and reliability of the electric drill are improved; and on the other hand, more accurate prediction is realized through a grading prediction mechanism and a dynamic threshold mechanism, the false alarm rate can be effectively reduced, and interference to operators is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electric drills, and in particular to an electric drill with an abnormality alarm function and an alarm method thereof. Background Art

[0002] In modern industrial production and daily household maintenance scenarios, electric drills, as efficient and convenient drilling tools, are widely used in processing a variety of materials such as wood, metal, and plastic. However, during the drilling process, the drill bit often accidentally contacts foreign objects hidden within the processed material, such as metal nails, stones, and other hard objects.

[0003] Currently, traditional electric drills lack effective monitoring and early warning mechanisms for drill bit contact with hard foreign objects. When a drill bit encounters such an object, it often experiences excessive impact and resistance, leading to severe damage such as chipping and breakage. This not only increases tool repair and replacement costs, but also disrupts normal operations, reduces work efficiency, and can even cause safety accidents due to drill bit damage.

[0004] Therefore, there is an urgent need for a method that can output an alarm message in time when the drill bit drills a foreign object that may cause damage to the drill bit, so as to improve the safety and reliability of the use of the electric drill. Summary of the Invention

[0005] To this end, the present invention provides an alarm method for an electric drill, an electric drill with an abnormality alarm function, an electronic device, a computer storage medium, and a computer program product to solve at least one of the above technical problems.

[0006] In a first aspect, the present invention provides an alarm method for an electric drill, the method comprising the following steps: constructing a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit; and using a hierarchical prediction mechanism to predict an abnormality probability based on the first multidimensional feature vector; wherein the hierarchical prediction mechanism is related to the number of holes drilled by the electric drill after this startup; and outputting a hard foreign body alarm signal when the abnormality probability is higher than a probability threshold; wherein the probability threshold is derived based on the drill bit properties.

[0007] According to a second aspect of the present invention, an electric drill with an abnormal alarm function is provided, comprising an abnormality prediction module and an alarm decision module; the abnormality prediction module is configured to construct a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit, and to predict the abnormality probability using a graded prediction mechanism based on the first multidimensional feature vector; wherein the graded prediction mechanism is related to the number of holes drilled by the electric drill after this startup; and the alarm decision module is configured to output a hard foreign body alarm signal when the abnormality probability is higher than a probability threshold; wherein the probability threshold is derived based on the properties of the drill bit.

[0008] In a third aspect of the present invention, an electronic device is provided, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program is executed by the processor to implement the method described in any one of the preceding items.

[0009] According to a fourth aspect of the present invention, a computer storage medium is provided, wherein the computer storage medium stores a computer program executable by a processor, wherein the computer program is executed by the processor to implement the method as described in any of the preceding items.

[0010] According to a fifth aspect of the present invention, a computer program product is provided. The computer program product comprises a computer program executable by a processor, wherein the computer program is executed by the processor to implement the method as described in any one of the preceding items.

[0011] The solution of the present invention, on the one hand, realizes real-time early warning of contact with hard foreign objects when the electric drill is drilling by fusing multi-dimensional sensor data, effectively avoiding damage such as chipping and breakage of the drill bit caused by hitting foreign objects, reducing tool maintenance and replacement costs, ensuring operation continuity, and improving the safety and reliability of electric drill use; on the other hand, more accurate predictions are achieved through a hierarchical prediction mechanism and a dynamic threshold mechanism, which can effectively reduce the false alarm rate and reduce interference with operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0013] Figure 1 The present invention is a flowchart of an alarm method for an electric drill disclosed in an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of using a hierarchical prediction mechanism to predict abnormality probabilities according to an embodiment of the present invention.

[0015] Figure 3 It is a schematic diagram of the structure of a lightweight neural network disclosed in an embodiment of the present invention.

[0016] Figure 4 The present invention is a structural diagram of an electric drill with an abnormality alarm function disclosed in an embodiment of the present invention.

[0017] Figure 5This is another structural schematic diagram of an electric drill with an abnormality alarm function disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following specific embodiments illustrate the implementation of this application. Those familiar with the art can easily understand the other advantages and functions of this application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0020] like Figure 1 As shown, an embodiment of the present invention discloses an alarm method for an electric drill, the method comprising the following steps: S10, constructing a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit, and predicting an abnormality probability based on the first multidimensional feature vector using a hierarchical prediction mechanism; wherein the hierarchical prediction mechanism is related to the number of drilling times of the electric drill after it is turned on this time.

[0021] This step achieves an accurate assessment of foreign object contact risk through multi-dimensional data fusion and a hierarchical prediction mechanism. Specifically, the sensor array first collects the following three key drill-related data in real time: Torque fluctuation data is acquired through a strain-gauge torque sensor. When the drill contacts a hard foreign object, the sudden increase in rotational resistance causes the torque to exceed the rated value by 20%-50% (e.g., a rated torque of 30 N·m may suddenly increase to over 45 N·m); current fluctuation data is monitored through a current transformer. To maintain the motor's speed, the current increases significantly (e.g., a normal 2A current may surge to over 3.5A); and vibration data is collected through a triaxial accelerometer. Impact with a hard foreign object generates a high-frequency vibration component of 200-500Hz with increased amplitude (e.g., a normal 0.1mm amplitude may increase to over 0.5mm).

[0022] The following multiple dimensions of features can be extracted from the above three types of key data: (1) Torque-related features: Torque mean: The average value of torque over a period of time, reflecting the basic load.

[0023] Torque variance: measures the degree of dispersion of torque fluctuations.

[0024] Maximum torque: The maximum torque value during the sampling period.

[0025] Minimum torque value: The minimum torque value during the sampling period.

[0026] Torque change rate: the average change in torque per unit time.

[0027] Torque Peak Factor: The ratio of peak torque to root mean square torque.

[0028] Torque kurtosis: describes the thickness of the tail of the torque distribution and reflects the impact characteristics.

[0029] (2) Current-related characteristics: Current mean: the average value of the motor's operating current.

[0030] Current standard deviation: reflects the stability of current fluctuations.

[0031] Current peak: The maximum current value during the sampling period.

[0032] Current rise slope: the rate of increase of current from the stable value to the peak value.

[0033] Current Fallback Rate: The rate at which current recovers after the load is removed.

[0034] Current harmonic content: The proportion of high-frequency harmonic energy obtained through FFT analysis.

[0035] (3) Vibration-related characteristics: Mean vibration acceleration: the average value of the three-axis vibration acceleration.

[0036] Vibration acceleration standard deviation: measures the fluctuation of vibration amplitude.

[0037] Vibration peak value: maximum vibration acceleration value.

[0038] Root mean square (RMS) vibration: an effective indicator of vibration energy.

[0039] 200-500Hz high-frequency energy ratio: energy in the characteristic frequency band generated by the collision of hard foreign objects.

[0040] Main frequency of vibration: the frequency component with the largest power spectrum density.

[0041] Vibration frequency entropy: reflects the complexity of vibration frequency distribution.

[0042] Three-axis vibration coupling coefficient: correlation index of X, Y, and Z axis vibration signals.

[0043] In specific implementation, the principal component analysis method (APH) can be used to filter out the features that best represent the drill bit drilling into hard foreign matter from the features of the above dimensions, such as torque mean, torque variance, current peak, 200-500Hz high-frequency energy ratio, vibration root mean square value, etc., which can improve the timeliness of prediction.

[0044] Subsequently, a hierarchical prediction mechanism is used to process the first multidimensional feature vector. This hierarchical prediction mechanism is determined based on the total number of holes drilled since the drill was powered on. Upon startup, the drill enters a new operating cycle, which likely corresponds to a different operating scenario or object. During the first N drilling operations, since sufficient drilling data for that operating scenario or object is not yet available, a more sensitive alarm mechanism is required. After N drilling operations, a more accurate alarm mechanism is implemented to reduce the false alarm rate. This will be analyzed in detail later and will not be repeated here.

[0045] S20: When the abnormal probability is higher than a probability threshold, output a hard foreign body alarm signal; wherein the probability threshold is derived based on drill bit properties.

[0046] This step achieves precise alarms through a dynamic threshold mechanism and multi-source data fusion. Specifically, when the abnormal probability output by the hierarchical prediction mechanism exceeds the probability threshold, the drill immediately sends a hard foreign object alarm signal via a buzzer (e.g., ≥85dB), a graphic display (e.g., a red warning icon), and a wireless communication module (Bluetooth / WiFi) to the terminal device, prompting the operator to stop and retract the drill immediately to reduce the risk of drill bit damage.

[0047] The dynamic determination of the probability threshold is based on drill bit attribute data, including drill material (high-speed steel / ceramic), diameter (3mm / 6mm), and edge wear (obtained through periodic scanning with an industrial camera or inferred based on recorded drilling time).

[0048] The solution of the present invention, on the one hand, realizes real-time early warning of contact with hard foreign objects when the electric drill is drilling by fusing multi-dimensional sensor data, effectively avoiding damage such as chipping and breakage of the drill bit caused by hitting foreign objects, reducing tool maintenance and replacement costs, ensuring operation continuity, and improving the safety and reliability of electric drill use; on the other hand, more accurate predictions are achieved through a hierarchical prediction mechanism and a dynamic threshold mechanism, which can effectively reduce the false alarm rate and reduce interference with operators.

[0049] Further, if Figure 2 As shown, based on the first multidimensional feature vector, a hierarchical prediction mechanism is used to predict the abnormality probability, including: determining the cumulative number of drilling times after the electric drill is turned on, if the drilling time is lower than the number threshold, using the first prediction mechanism to predict the abnormality probability; otherwise, using the second prediction mechanism to predict the abnormality probability; wherein, the first prediction mechanism is a rule-based mechanism, the second prediction mechanism is a deep algorithm-based mechanism, and the warning sensitivity of the first prediction mechanism is higher than that of the second prediction mechanism.

[0050] The electric drill of the present invention can be used in factory environments, such as those used to produce and assemble furniture and mechanical equipment. In these environments, operators use the drill to perform repetitive batch drilling operations on different types of objects over varying operating cycles, determined by the power cycle. Furthermore, the hardness of objects (batches) within each operating cycle, as well as the hardness, content, and distribution of impurities, can vary. This makes it difficult to use unified rules or thresholds for early warning, as false alarms are likely to occur.

[0051] For this factory scenario, the present invention provides a customized, special warning mode for the electric drill. This mode can be enabled manually or, alternatively, enabled by default when the drill is manufactured for this factory scenario. During each operation cycle, the operator first uses the drill with this special warning mode enabled to drill a batch of objects, or to perform a drilling test (for example, the operator randomly drills multiple holes on samples (corresponding to the drilling objects in this batch)). During this process, the drill accumulates the number of holes drilled (each cycle from the start of the drill to its shutdown is counted as one drilling cycle).

[0052] When the total number of holes drilled since startup falls below a threshold, the first rule-based prediction mechanism is activated. This mechanism triggers an alarm if any feature in the first multidimensional feature vector is detected to be abnormal—for example, a sudden increase in torque, a surge in current, or the proportion of high-frequency energy between 200 and 500 Hz exceeding a preset threshold. This highly sensitive mechanism is suitable for rapid early warning in initial operations.

[0053] When the total number of drilled holes reaches a threshold, the system automatically switches to a second prediction mechanism based on a deep learning algorithm. This prediction mechanism retrains the deep learning model using the drilling log data before the threshold is reached. This retrained deep learning model is then used to analyze the first multidimensional feature vector collected in step S10 to determine the probability of drilling into a hard foreign object. Compared to rule-based prediction mechanisms, this prediction mechanism significantly improves accuracy and reduces false alarm rates.

[0054] It's understandable that the deep learning model has been pre-trained, and retraining it using drilling log data is fine-tuning based on small sample data to adapt it to the characteristics of the target during this operation cycle. Furthermore, drilling operations before the threshold are used to accumulate small sample data for fine-tuning the deep learning model. Furthermore, during this early stage, because the characteristics of the target during this operation cycle are not yet fully understood, a more sensitive first prediction mechanism is employed to reduce the probability of drill bit damage.

[0055] Furthermore, the probability threshold corresponding to the first mechanism can be set lower than the probability threshold under the second mechanism, so that the warning sensitivity of the first prediction mechanism is higher than that of the second prediction mechanism. It should be noted that under the first mechanism, when any indicator is higher than the preset indicator threshold, the abnormal probability can be directly set to be higher than the probability threshold, that is, the hard foreign object alarm signal is triggered. Moreover, this indicator threshold is lower than the conventional alarm indicator threshold, thereby adjusting its alarm sensitivity to be higher than usual.

[0056] In addition, the number threshold can be set manually, for example, 8 times, 10 times, 15 times, etc.; it can also be based on the minimum requirement of small sample data for fine-tuning training of the deep learning model, and the minimum requirement can be based on the actual training rules of the deep learning model, which will not be described in detail.

[0057] Furthermore, the deep algorithm is based on a lightweight convolutional neural network, and the method also includes: during the operation interval after the drilling number exceeds the number threshold, based on the multiple drilling record data after this startup, the target feature pattern is matched in the feature patterns pre-stored in the feature mask library; wherein, each pre-stored feature pattern corresponds to objects of different materials and high-frequency feature combinations; the network layer in the lightweight convolutional neural network except the network layer corresponding to the high-frequency feature combination is set to a frozen state, and the multiple drilling record data are used to fine-tune the lightweight convolutional neural network after the setting and processing.

[0058] Once the drilling count exceeds a threshold, the second prediction mechanism automatically activates. However, this requires fine-tuning the deep learning model using multiple previously recorded drilling records (which can be some or all of the recorded data). To prevent damage to the drill bit, the alarm for drilling into hard foreign objects must be as fast as possible. Lightweight convolutional neural networks employ an inverse residual structure. Through a bottleneck structure of "dilation-convolution-compression," they reduce the number of parameters while preserving feature expression capabilities, reducing computational complexity by 75% compared to traditional CNNs. Therefore, the present invention employs a lightweight convolutional neural network to implement the second prediction mechanism.

[0059] During the interval between drills, when the number of drillings exceeds a threshold, i.e., during the drill's work intervals, fine-tuning training of the lightweight convolutional neural network is initiated. Specifically, the present invention pre-installs a feature mask library in the drill's memory, storing multiple feature patterns. Each feature pattern corresponds to an object of a specific material, such as wood (solid wood: pine, oak, etc.; artificial wood: plywood, MDF, particleboard), plastic (hard plastic: PVC, ABS; soft plastic: polyethylene), or composite materials (plastic-steel / aluminum-plastic materials, fiberglass board, carbon fiber materials, resin-based composite materials, etc.). Furthermore, each feature pattern is associated with a corresponding high-frequency feature combination. The high-frequency features in this combination are the most relevant characterization parameters for when the drill hits a hard foreign object. In other words, the confidence level of the probability of the drill hitting a hard foreign object, predicted based on this high-frequency feature combination, is significantly higher or the highest.

[0060] Based on the matching calculation of multiple drilling record data after this startup and the feature patterns pre-stored in the feature mask library, the feature pattern with the highest similarity can be obtained, which is used as the target feature pattern, and then the high-frequency feature combination corresponding to the target feature pattern is adopted.

[0061] In the lightweight convolutional neural network, a corresponding network layer is set for each feature, and each network layer has been trained during pre-training. However, during the fine-tuning training process, only the network layers corresponding to the high-frequency feature combination are trained. Specifically, the network layers corresponding to the high-frequency feature combination continue to remain in an activated state, while the network layers that are insensitive to the high-frequency feature combination are set to a frozen state using the network layer freezing technology. In this way, during the fine-tuning training process, only the network layers corresponding to the high-frequency feature combination (such as torque mean, torque variance, current peak, 200-500Hz high-frequency energy ratio, vibration root mean square value) can be updated with parameters, significantly reducing the amount of model updates for fine-tuning training, thereby accelerating the speed of fine-tuning training while ensuring the effectiveness of fine-tuning training.

[0062] It's understandable that the purpose of fine-tuning training is to enable the lightweight convolutional neural network to learn the characteristics of the object in the operation cycle, namely the hardness of the object, the hardness of the internal hard objects, and their distribution characteristics. This reduces the probability of issuing useless alarms due to excessive hardness of the object, excessive hardness of the internal hard objects, or a high distribution of hard objects (but not to the extent that they would damage the drill bit). Furthermore, the multiple drilling records obtained by screening should be records that did not issue alarms.

[0063] The lightweight convolutional neural network is used to determine the probability that the first multidimensional feature vector detected under this drilling condition belongs to the abnormal condition of drilling into a hard foreign object, that is, the abnormal probability. The specific description is as follows: Figure 3As shown in the figure, the lightweight convolutional neural network preferably adopts the MobileNetV3 architecture, including an input layer, a convolutional layer, three serially connected depth-wise separable convolutional modules, a global average pooling layer, a fully connected layer, a dropout layer, and an output layer.

[0064] Input layer: Receives feature data in the form of one-dimensional signals, with the number of channels consistent with the feature dimension.

[0065] The convolution layer (Conv1) uses a 1×1 convolution kernel, 16 channels, a stride of 1, and a ReLU6 activation function.

[0066] Three depth-wise separable convolution modules (Block1, Block2, Block3) are connected in series. Each module has the same structure: it first passes through depthwise convolution (Depthwise Conv) with a 3×3 convolution kernel. The number of channels is consistent with the output of the previous layer (Block1 is 16 channels, Block2 is 24 channels, and Block3 is 32 channels), with a stride of 1, and then undergoes batch normalization (BatchNorm) and ReLU6 activation.

[0067] Then, through pointwise convolution, a 1×1 convolution kernel is used, and the number of channels is increased to 24, 32, and 48 respectively, with a step size of 1.

[0068] Then, the SE module is connected to perform global average pooling first, and then the channel is compressed to 8 through 1×1 convolution. After ReLU activation, 1×1 convolution is used to restore the output channel number of the current module (24, 32, 48), and finally the channel attention weight is generated through Sigmoid.

[0069] The output at the end of the module is batch normalized and HardSwish activated.

[0070] The global average pooling layer compresses the features into a vector.

[0071] Connect the fully connected layer to map the 48-dimensional features to 32 dimensions and use ReLU6 activation.

[0072] The dropout layer (dropout rate 0.2) prevents overfitting.

[0073] The output layer maps the 32-dimensional features to 1-dimensional through full connection, and outputs the abnormality probability (range 0-1) after Sigmoid activation.

[0074] The pre-training process is as follows: 1. Data preparation: Dataset partitioning: 70% training set, 15% validation set, and test set. Data augmentation: Time domain: Adding Gaussian noise (SNR = 20dB), time axis scaling (±15%); Frequency domain: Random frequency shift (±5Hz), spectral masking (mask length = 5%). The above datasets include both normal operating condition sample data and foreign object condition sample data (wherein the normal operating condition label for the normal operating condition sample data and the foreign object condition label for the foreign object condition sample data are manually or automatically annotated). The ratio of the two can be freely set, and the specific details are not detailed here.

[0075] 2. Training parameters: Optimizer: Adam, initial learning rate 0.001, cosine annealing learning rate schedule (epochs = 10, minimum value = 0.00001).

[0076] Loss function: Focal Loss (weight factor α=0.75, focusing parameter γ=2) to address data imbalance.

[0077] Training configuration: Batch Size = 64, Epoch = 100, early stopping strategy (stop if the validation set loss does not decrease after 5 epochs).

[0078] The following training objectives are adopted: 1. Main objectives: Prediction accuracy: Achieve anomaly detection accuracy of more than 92% on the test set; False alarm rate: The false alarm rate under normal working conditions is less than 3%; Missing alarm rate: The missing alarm rate under foreign object working conditions is less than 5%; Inference speed: Single sample inference time ≤ 15ms.

[0079] 2. Auxiliary Objective: Model Compression: Compress the model size to ≤1MB through quantization (8-bit integers) and pruning (removing connections with weights < 0.01); Generalization: Maintain a detection accuracy of over 85% on drilling data across batches and materials.

[0080] 3. Evaluation indicators can use weighted F1 score as the core evaluation indicator to balance the detection performance of different categories (normal / abnormal): ; Precision is the precision rate, Recall is the recall rate, and they are calculated for abnormal categories.

[0081] After pre-training, the lightweight convolutional neural network is able to distinguish between true anomaly features caused by hard foreign objects and false anomaly features caused by the hardness of objects of different materials. Subsequent fine-tuning training (the specific process is generally the same and will not be repeated here) can further improve this ability to distinguish specific objects.

[0082] Furthermore, when fine-tuning the lightweight convolutional neural network, the electric drill can also output a reminder message of fine-tuning training to the operator to remind him to wait until the fine-tuning training is completed before using the electric drill to perform drilling operations, for example, through the display screen on the electric drill. The present invention does not impose specific restrictions on this.

[0083] Furthermore, the target feature pattern is obtained by matching the feature patterns pre-stored in the feature mask library based on the multiple drilling record data after this startup, including: constructing a second multidimensional feature vector based on the multiple drilling record data after this startup, performing similarity calculation between the second multidimensional feature vector and the feature patterns pre-stored in the feature mask library, and determining the feature pattern with the highest similarity as the target feature pattern; calculating a first number of all feature patterns whose similarity is higher than a similarity threshold, obtaining a second number based on the matching of the first number, randomly screening the second number of features from all features associated with the lightweight convolutional neural network, and adding the screened features to the high-frequency feature combination corresponding to the target feature pattern.

[0084] In batch drilling scenarios in factories, the objects being drilled may contain mixed materials (e.g., metal embedded in wood) and material differences between batches (e.g., large variations in wood hardness). This makes it difficult to accurately capture actual working conditions based solely on high-frequency feature combinations. For example, when the drilling object features are between two materials (e.g., concrete containing quartz sand), or when the drilling object is a new composite material containing hard particles, the deep learning model may be unable to identify potential foreign object risks, resulting in missed detections. Furthermore, fixed high-frequency feature combinations are difficult to adapt to subtle differences between batches (e.g., variations in metal sheet thickness), potentially leading to increased false alarm rates.

[0085] Therefore, the present invention incorporates a mechanism for dynamically inserting random features to enhance the deep learning model's adaptability to non-standard operating conditions, improving generalization capabilities while maintaining prediction speed. Specifically, features are extracted from multiple drilling logs (the data used to output alarm information) after the current startup to construct a second multidimensional feature vector. In addition to including the same types of features as the first multidimensional feature vector, the second multidimensional feature vector may also include an appropriate number of other features, such as torque change rate, current harmonic content, and vibration frequency entropy.

[0086] Next, a similarity calculation (e.g., using a cosine similarity algorithm) is performed between the second multidimensional feature vector and the third multidimensional feature vector corresponding to each pre-stored feature pattern in the feature mask library. The third multidimensional feature vector is a typical feature vector under normal drilling conditions (i.e., without drilling into hard foreign objects), derived from the high-frequency feature combination corresponding to the feature pattern. The feature pattern with the highest similarity is determined as the target feature pattern. For example, if the feature pattern with the highest similarity is "metal," the default high-frequency feature combination is set to current peak value, vibration RMS, etc.

[0087] Next, a first number of all feature patterns whose similarity between the second multidimensional feature vector and each third multidimensional feature vector exceeds a similarity threshold (e.g., 0.6) is counted. For example, if metal (0.7), concrete (0.65), and wood (0.5) are similar, the first number is 2.

[0088] The second quantity is obtained based on matching the first quantity with a preset matching rule. The matching rule is, for example: when the first quantity = 1, the second quantity = 0 (only using the high-frequency feature combination of the target feature pattern); when the first quantity = 2, the second quantity = 2 (supplementing 2 random features); when the first quantity ≥ 3, the second quantity = 3 (supplementing 3 random features).

[0089] It can be understood that when the first number is small (for example, 1), it means that the characteristics of the drilling object are very clear and no additional features need to be added to avoid computational redundancy; when the first number is larger (for example, ≥2), it means that the characteristics of the drilling object are more complex and more random features need to be added to cover potential non-pre-stored features and enhance the adaptability of the deep learning model to mixed, special, and unknown working conditions.

[0090] Then, a second number of features are randomly selected from all features associated with the lightweight convolutional neural network (such as torque kurtosis and current harmonic content) and added to the high-frequency feature combination of the target feature pattern. For example, if the target feature pattern is "metal" and the original high-frequency feature combination is [current peak value, vibration RMS], the random features "torque change rate" and "current harmonic content" are added to form a new high-frequency feature combination of [current peak value, vibration RMS, torque change rate, current harmonic content].

[0091] By randomly inserting features, the deep learning model can learn edge features not covered by the pre-existing patterns (such as the impact of the oxide layer in metal sheets on torque), reducing missed detections due to material variation.

[0092] Furthermore, the method also includes: determining the probability threshold based on the drill bit attributes, specifically: calculating the basic probability threshold through logistic regression based on the historical drilling data of the same type of drill bits; the historical drilling data includes the characteristic vectors of normal working conditions and working conditions of contacting hard foreign matter and the actual working condition labels; and correcting the basic probability threshold based on the cutting edge wear of the drill bit to obtain the probability threshold.

[0093] Different types of drill bits (of varying materials and diameters) have vastly different tolerances and performance when encountering hard foreign matter. For example, ceramic drill bits are hard but brittle, while high-speed steel drill bits offer excellent toughness but relatively weak wear resistance. Using a uniform probability threshold to determine whether a hard foreign object has been drilled can lead to false positives or missed reports. By collecting historical drilling data for the same type of drill bit, which contains feature vectors and corresponding actual operating condition labels for normal drilling conditions and conditions involving contact with hard foreign matter, a logistic regression algorithm can be used to find a reasonable probability threshold—the base probability threshold—to distinguish between normal and abnormal operating conditions based on the past performance of that type of drill bit.

[0094] At the same time, cutting edge wear is inevitable during drill use, and greater wear deteriorates the drill's cutting performance and strength. A new drill bit has excellent strength and performance, and can withstand large fluctuations without damage. In this case, a relatively high probability threshold can be set to reduce unnecessary alarms. However, a severely worn drill bit can be damaged even by small hard foreign objects. To promptly alert operators and reduce the risk of drill damage, the probability threshold needs to be lowered, making the alarm system more sensitive.

[0095] By setting the correction coefficient corresponding to different cutting edge wear intervals, the threshold can be dynamically adjusted according to the real-time wear status of the drill bit, so that the alarm system can adapt to changes in drill bit performance, while ensuring judgment accuracy and protecting the drill bit to the greatest extent. The specific examples are shown in the following table:

[0096] like Figure 4 As shown, an embodiment of the present invention further provides an electric drill 100 with an abnormal alarm function, including an abnormality prediction module 101 and an alarm decision module 102; the abnormality prediction module 101 is used to: construct a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit, and use a hierarchical prediction mechanism to predict the abnormality probability based on the first multidimensional feature vector; wherein the hierarchical prediction mechanism is related to the number of drilling times after the electric drill is turned on this time; the alarm decision module 102 is used to: output a hard foreign body alarm signal when the abnormality probability is higher than a probability threshold; wherein the probability threshold is derived based on the drill bit properties.

[0097] Furthermore, the abnormality prediction module 101 is specifically used to: determine the cumulative number of drilling times after the electric drill is turned on; if the number of drilling times is lower than the number threshold, use the first prediction mechanism to predict the abnormality probability; otherwise, use the second prediction mechanism to predict the abnormality probability; wherein, the first prediction mechanism is a rule-based mechanism, and the second prediction mechanism is a deep algorithm-based mechanism, and the warning sensitivity of the first prediction mechanism is higher than that of the second prediction mechanism.

[0098] Furthermore, the deep algorithm is based on a lightweight convolutional neural network, and the abnormality prediction module 101 is also specifically used to: during the operation interval after the drilling number is higher than the number threshold, based on the multiple drilling record data after this startup, match the target feature pattern in the feature patterns pre-stored in the feature mask library; wherein, each pre-stored feature pattern corresponds to an object of different materials and a high-frequency feature combination; set the network layer in the lightweight convolutional neural network except the network layer corresponding to the high-frequency feature combination to a frozen state, and use the multiple drilling record data to fine-tune the lightweight convolutional neural network after the setting and processing.

[0099] Furthermore, the abnormality prediction module 101 is also specifically used to: construct a second multidimensional feature vector based on multiple drilling record data after this startup, calculate the similarity between the second multidimensional feature vector and each feature pattern pre-stored in the feature mask library, and determine the feature pattern with the highest similarity as the target feature pattern; calculate a first number of all feature patterns with a similarity higher than a similarity threshold, obtain a second number based on the matching of the first number, randomly select the second number of features from all features associated with the lightweight convolutional neural network, and add the selected features to the high-frequency feature combination corresponding to the target feature pattern.

[0100] Furthermore, if Figure 5 As shown, it also includes a threshold determination module 103, which is specifically used to determine the probability threshold based on the drill bit attributes, specifically: based on the historical drilling data of the same type of drill bits, a basic probability threshold is calculated by logistic regression; the historical drilling data includes the feature vectors of normal working conditions and working conditions of contacting hard foreign matter and the actual working condition labels; the basic probability threshold is corrected based on the edge wear of the drill bit to obtain the probability threshold.

[0101] An embodiment of the present invention further provides an electronic device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program is executed by the processor to implement any of the methods described above.

[0102] An embodiment of the present invention further provides a computer storage medium storing a computer program executable by a processor, wherein the computer program is executed by the processor to implement any of the methods described above.

[0103] An embodiment of the present invention further provides a computer program product, which includes a computer program executable by a processor, wherein the computer program is executed by the processor to implement any of the methods described above.

[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0105] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An alarm method for an electric drill, characterized in that: The method includes the following steps: constructing a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit; and using a hierarchical prediction mechanism to predict an abnormality probability based on the first multidimensional feature vector; wherein the hierarchical prediction mechanism is related to the number of holes drilled by the electric drill after it is turned on this time; and outputting a hard foreign matter alarm signal when the abnormality probability is higher than a probability threshold; wherein the probability threshold is derived based on the drill bit properties.

2. The alarm method for an electric drill according to claim 1, characterized in that: Based on the first multidimensional feature vector, a hierarchical prediction mechanism is used to predict the abnormality probability, including: determining the cumulative number of drilling times after the electric drill is turned on; if the number of drilling times is lower than a threshold, a first prediction mechanism is used to predict the abnormality probability; otherwise, a second prediction mechanism is used to predict the abnormality probability; wherein, the first prediction mechanism is a rule-based mechanism, the second prediction mechanism is a deep algorithm-based mechanism, and the warning sensitivity of the first prediction mechanism is higher than that of the second prediction mechanism.

3. The alarm method for an electric drill according to claim 2, characterized in that: The deep algorithm is based on a lightweight convolutional neural network, and the method also includes: during the operation interval after the number of drilling times exceeds the number threshold, based on the multiple drilling record data after this startup, the target feature pattern is matched in the feature patterns pre-stored in the feature mask library; wherein, each pre-stored feature pattern corresponds to an object of different materials and a high-frequency feature combination; the network layer in the lightweight convolutional neural network except the network layer corresponding to the high-frequency feature combination is set to a frozen state, and the multiple drilling record data are used to fine-tune the lightweight convolutional neural network after the setting and processing.

4. The alarm method for an electric drill according to claim 3, characterized in that: A target feature pattern is obtained by matching the feature patterns pre-stored in the feature mask library based on the multiple drilling record data after the current startup, including: constructing a second multidimensional feature vector based on the multiple drilling record data after the current startup, performing similarity calculation between the second multidimensional feature vector and the feature patterns pre-stored in the feature mask library, and determining the feature pattern with the highest similarity as the target feature pattern; calculating a first number of all feature patterns with a similarity higher than a similarity threshold, obtaining a second number based on the matching of the first number, randomly screening the second number of features from all features associated with the lightweight convolutional neural network, and adding the screened features to the high-frequency feature combination corresponding to the target feature pattern.

5. The alarm method for an electric drill according to claim 1, characterized in that: The method further includes: determining the probability threshold based on the drill bit attributes, specifically: calculating a basic probability threshold through logistic regression based on historical drilling data of drill bits of the same type; the historical drilling data includes feature vectors of normal working conditions and working conditions in contact with hard foreign matter, as well as actual working condition labels; and correcting the basic probability threshold based on the amount of cutting edge wear of the drill bit to obtain the probability threshold.

6. An electric drill with an abnormal alarm function, characterized in that: It includes an abnormality prediction module and an alarm decision module; the abnormality prediction module is used to: construct a first multidimensional feature vector based on torque fluctuation data, current fluctuation data, and vibration data associated with the drill bit, and use a hierarchical prediction mechanism to predict the abnormality probability based on the first multidimensional feature vector; wherein the hierarchical prediction mechanism is related to the number of holes drilled after the electric drill is turned on this time; the alarm decision module is used to: output a hard foreign body alarm signal when the abnormality probability is higher than a probability threshold; wherein the probability threshold is derived based on the drill bit properties.

7. The electric drill with abnormality alarm function according to claim 6, characterized in that: The abnormality prediction module is specifically used to: determine the cumulative number of drilling times after the electric drill is turned on; if the number of drilling times is lower than the number threshold, use the first prediction mechanism to predict the abnormality probability; otherwise, use the second prediction mechanism to predict the abnormality probability; wherein, the first prediction mechanism is a rule-based mechanism, the second prediction mechanism is a deep algorithm-based mechanism, and the warning sensitivity of the first prediction mechanism is higher than that of the second prediction mechanism.

8. An electronic device, characterized in that: The electronic device includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the computer program is executed by the processor to implement the method according to any one of claims 1 to 5.

9. A computer storage medium, characterized in that: The computer storage medium stores a computer program that can be executed by a processor, and the computer program is executed by the processor to implement the method according to any one of claims 1 to 5.

10. A computer program product, characterized in that: The computer program product comprises a computer program executable by a processor, wherein the computer program is executed by the processor to implement the method according to any one of claims 1 to 5.

Images