Magnetic disk drive and control method

Abstract

Properly controlling the operation of the magnetic head. [Solution] In the magnetic disk device of the embodiment, the controller, while writing to the first track among the plurality of tracks, determines using a determination model whether the position of the magnetic head moves from the target position on the first track to the side of the second track which is different from the first track among the plurality of tracks, and whether the radial distance from the target position to the position of the magnetic head exceeds a threshold. If the distance does not exceed the threshold, the controller performs writing to the first track, and if the distance exceeds the threshold, the controller does not perform writing to the first track. The determination model is a trained model that takes multiple position error information obtained by reading the servo information from each of the plurality of servo regions as input and outputs whether the distance exceeds the threshold while the magnetic head moves through the plurality of servo regions.

Description

Technical Field

[0001] Embodiments of the present invention relate to a magnetic disk device and a control method.

Background Art

[0002] In a magnetic disk device, a magnetic head writes and reads data to and from a magnetic disk. It is desirable to appropriately control the operation of the magnetic head.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0004] One embodiment aims to provide a magnetic disk device capable of appropriately controlling the operation of a magnetic head.

Means for Solving the Problems

[0005] The magnetic disk device of the embodiment comprises a magnetic disk having a plurality of tracks including a data area and a servo area on which servo information is recorded, and a controller, wherein during the execution of writing to the first track of the plurality of tracks, the controller uses a determination model to determine whether the position of the magnetic head has shifted from a target position on the first track to a second track different from the first track, and whether the radial distance from the target position to the position of the magnetic head has exceeded a threshold, and if the distance does not exceed the threshold, it performs writing to the first track, and if the distance exceeds the threshold, it does not perform writing to the first track, and the determination model is a trained model that takes a plurality of position error pieces obtained by reading the servo information from each of the plurality of servo areas as input and outputs whether the distance has exceeded the threshold while the magnetic head moves through the plurality of servo areas. [Brief explanation of the drawing]

[0006] [Figure 1] Figure 1 is a schematic diagram showing an example of the configuration of a magnetic disk drive according to the first embodiment. [Figure 2] Figure 2 is a schematic diagram showing an example of the configuration of a magnetic disk in the first embodiment. [Figure 3] Figure 3 is a schematic diagram illustrating an example of the operation of the magnetic head according to the first embodiment during data writing. [Figure 4] Figure 4 shows an example of the input and output of the decision model according to the first embodiment. [Figure 5] Figure 5 shows an example of the configuration of a decision model composed of a multilayer perceptron neural network according to the first embodiment. [Figure 6] Figure 6 is a diagram illustrating an example of training data for training the decision model according to the first embodiment. [Figure 7] Figure 7 shows an example of the configuration of a decision model composed of a recurrent neural network according to the first embodiment. [Figure 8] Figure 8 is a flowchart showing an example of the control process procedure according to the first embodiment. [Figure 9] Figure 9 is a flowchart showing an example of the learning process for the decision model according to the first embodiment. [Figure 10] Figure 10 shows an example of the difference in effects between the first embodiment and the comparative example. [Figure 11] Figure 11 is a control block diagram showing an example of the configuration of the simulator used in the second embodiment. [Figure 12] Figure 12 shows an example of the input and output of the decision model according to the second embodiment. [Figure 13] Figure 13 is a flowchart showing an example of the control process procedure according to the second embodiment. [Figure 14] Figure 14 is a flowchart showing an example of the learning process for the decision model according to the second embodiment. [Modes for carrying out the invention]

[0007] The magnetic disk device and control method according to the embodiments will be described in detail below with reference to the attached drawings. However, the present invention is not limited to these embodiments.

[0008] (First embodiment) Figure 1 is a schematic diagram showing an example of the configuration of the magnetic disk device 1 of the first embodiment.

[0009] The magnetic disk drive 1 is connected to the host 2. The magnetic disk drive 1 can receive access commands, such as write commands and read commands, from the host 2.

[0010] The magnetic disk drive 1 includes a magnetic disk 11 having a recording surface formed on its surface. The magnetic disk drive 1 writes and reads data to / from the magnetic disk 11 (more precisely, the recording surface of the magnetic disk 11) according to an access command. Note that the magnetic disk drive 1 may have a plurality of magnetic disks 11, but in the embodiment, for simplicity of explanation and illustration, the magnetic disk drive 1 is assumed to include one magnetic disk 11.

[0011] Writing and reading of data are performed via the magnetic head 22. Specifically, in addition to the magnetic disk 11, the magnetic disk drive 1 includes a spindle motor 12, a motor driver IC (Integrated Circuit) 21, a magnetic head 22, an actuator arm 15, a voice coil motor (VCM) 16, a lamp 13, a head IC 24, a read / write channel (RWC) 25, a RAM 27, a FROM (Flash Read Only Memory) 28, a buffer memory 29, a hard disk controller (HDC) 23, and a processor 26.

[0012] The magnetic disk 11 is rotated at a predetermined rotational speed by a spindle motor 12 attached to the rotation axis of the magnetic disk 11. The spindle motor 12 is driven by the motor driver IC 21.

[0013] The motor driver IC 21 controls the rotation of the spindle motor 12 and the rotation of the VCM 16.

[0014] The magnetic head 22 writes and reads data to / from the magnetic disk 11 by means of a write element 22w and a read element 22r provided therein. Further, the magnetic head 22 is attached to the tip of the actuator arm 15. The magnetic head 22 is moved along the radial direction of the magnetic disk 11 by the VCM 16 driven by the motor driver IC 21.

[0015] When the rotation of the magnetic disk 11 is stopped, the magnetic head 22 is moved onto the ramp 13. The ramp 13 is configured to hold the magnetic head 22 in a position away from the magnetic disk 11.

[0016] During reading, the head IC24 amplifies the signal read by the magnetic head 22 from the magnetic disk 11 and outputs it, supplying it to the RWC25. The head IC24 also amplifies the signal corresponding to the data to be written, supplied by the RWC25, and supplies it to the magnetic head 22.

[0017] HDC23 controls the transmission and reception of data between it and host 2 via the I / F bus, controls the buffer memory 29, and performs error correction processing on the read data.

[0018] The buffer memory 29 is used as a buffer for data transmitted to and from the host 2. For example, the buffer memory 29 is used to temporarily store data written to or read from the magnetic disk 11.

[0019] The buffer memory 29 is composed of, for example, volatile memory capable of high-speed operation. The type of memory that constitutes the buffer memory 29 is not limited to a specific type. The buffer memory 29 may be composed of, for example, DRAM (Dynamic Random Access Memory), SRAM (Static Random Access Memory), or a combination thereof.

[0020] The RWC25 modulates the data to be written, supplied from the HDC23, and sends it to the head IC24. The RWC25 also demodulates the signal read from the magnetic disk 11 and supplied from the head IC24, and outputs it as digital data to the HDC23.

[0021] RWC25 and HDC23 are connected to FRAM (Flash Random Access Memory) 35. FRAM 35 is a read / write non-volatile memory. The judgment model 351 and simulator 352 are stored in FRAM 35. Simulator 352 simulates position error information, which will be described later. Details of the judgment model 351 will be described later.

[0022] The processor 26 is, for example, a CPU (Central Processing Unit). RAM 27, FROM (Flash Read Only Memory) 28, and buffer memory 29 are connected to the processor 26.

[0023] FROM28 is a non-volatile memory. FROM28 stores firmware (program data) and various operating parameters. The firmware may also be stored on the magnetic disk 11.

[0024] RAM27 is composed of, for example, DRAM, SRAM, or a combination thereof. RAM27 is used by the processor 26 as operating memory. RAM27 is used as an area where firmware is loaded and where various management data is stored.

[0025] The processor 26 controls the magnetic disk device 1 according to the firmware stored in FROM 28 or the magnetic disk 11. For example, the processor 26 loads the firmware from FROM 28 or the magnetic disk 11 into RAM 27 and controls the motor driver IC 21, head IC 24, RWC 25, HDC 23, etc., according to the loaded firmware.

[0026] The configuration including RWC25, processor 26, FRAM35, and HDC23 can also be considered as a controller 30. The controller 30 may also include other elements (for example, RAM27, FROM28, buffer memory29, or RWC25).

[0027] Figure 2 is a schematic diagram showing an example of the configuration of the magnetic disk 11 in the first embodiment.

[0028] During the manufacturing process, servo information is written to the magnetic disk 11, for example, by a servo writer or by a self-servo write (SSW). As shown in Figure 2, a radial arrangement of servo regions 42 is shown as an example of the arrangement of servo regions on which servo information is written.

[0029] The servo information includes sector / cylinder information, burst patterns, and postcodes. The sector / cylinder information can provide the circumferential servo address (servo sector address) and radial servo address (track address) of the magnetic disk 11. During operation of the magnetic disk drive 1, the sector / cylinder information is used to control the seek operation that moves the magnetic head 22 to the target track.

[0030] The burst pattern is data used to detect the amount of radial displacement of the magnetic head 22 relative to the center of the track 41 (hereinafter abbreviated as the track center), and consists of a repeating pattern with a predetermined period. For example, the cylinder address is given as an integer value, and by demodulating the burst pattern, it is possible to obtain a decimal offset amount relative to the position indicated by the cylinder address.

[0031] Here, the track position defined by the burst pattern may deviate from the actual track position due to errors in writing servo information, etc. This positional deviation occurs repeatedly in the same way with a period of one rotation of the magnetic disk (and spindle motor), and is therefore called RRO (Repeatable Ru-Out). In the manufacturing process, the RRO is learned for each track, and the learned RRO value is written to the magnetic disk 11 as a postcode. Then, when the magnetic disk device 1 is used and the magnetic head 22 is positioned on the target track, control is performed to cancel out the positional deviation caused by the RRO based on the postcode.

[0032] As shown in Figure 2, multiple concentric tracks 41 are configured by the burst pattern and postcode. A servo region 42 on one track 41 can be referred to as a servo sector. Between the servo regions 42 (i.e., servo sectors) around each track 41, there are data regions 43 on which data can be written. Multiple data sectors are formed sequentially within the data region 43. Data is written and read from each data sector by the magnetic head 22.

[0033] The signal read by the magnetic head 22 (more precisely, the lead element 22r) includes servo information read from the servo sector and data read from the data sector. The servo information read by the magnetic head 22 is demodulated by the head IC 24 into a position error signal (hereinafter also referred to as "position error information" or "demodulated value") indicating the relative position of the magnetic head 22 from the track, and supplied to the controller 30. Because demodulation noise is superimposed during demodulation, this position error signal differs from the actual relative position of the magnetic head 22 (hereinafter also referred to as "actual position" or "true value"). The controller 30 works in cooperation with the motor driver IC 21 to perform positioning control of the magnetic head 22 based on the supplied position error signal.

[0034] For example, the controller 30 and motor driver IC 21 perform feedback control based on the position error signal to bring the difference between the position of the target track and the current position of the magnetic head 22 closer to zero.

[0035] The movement of the magnetic head 22 broadly includes a seek operation and a track-following operation. The seek operation is the operation of moving the magnetic head 22 radially across the magnetic disk 11 toward the target track. The track-following operation is the operation of maintaining the position of the magnetic head 22 on the target track after the magnetic head 22 has been moved near the target track by the seek operation. In the track-following operation, the position of the magnetic head 22 is adjusted by the feedback control described above so that the magnetic head 22 moves relative to the target track. The state in which the position of the magnetic head 22 is maintained on the target track is referred to as the on-track state.

[0036] Data writing and data reading are performed when the magnetic head 22 is on track. That is, the controller 30 determines whether the magnetic head 22 is on track, and if the magnetic head 22 is on track, it uses the magnetic head 22 to write or read data.

[0037] Whether the magnetic head 22 is on-track or not is determined based on a threshold set relative to the track position. For example, the threshold used to determine the on-track state during light operation is called WOS (Write Offtrack Slice).

[0038] Figure 3 is a schematic diagram illustrating an example of the operation of the magnetic head 22 according to the first embodiment during data writing. When referring to the radial position, the inner circumference of the magnetic disk 11 is considered the positive side, and the outer circumference of the magnetic disk 11 is considered the negative side. The designer can arbitrarily decide whether to consider the inner circumference or the outer circumference of the magnetic disk 11 as the positive side with respect to the radial position.

[0039] Figure 3 shows multiple servo sectors (SrvSct), specifically servo sectors #k-4 to #k+2. It also shows the center of track #n as one of several tracks. Furthermore, it displays the trajectory of the magnetic head 22 when data is written to track #n.

[0040] When writing data to track #n, the track-following operation adjusts the position of the magnetic head 22 so that the difference between the position of the magnetic head 22 and the center of track #n approaches zero. However, the position of the magnetic head 22 may fluctuate from the target position, such as the read or write position, due to various disturbances. As a result, as shown in Figure 3, the trajectory of the magnetic head 22 fluctuates from the center of track #n.

[0041] For each track, a predetermined range of variation for the magnetic head 22 during data writing is set. The line defining the boundary of the variation range is the WOS (Working Speed). As shown in Figure 3, the positive WOS is set at a position fixed by a fixed value L from the center of track #n, and the negative WOS is set at a position fixed by a fixed value L from the center of track #n. Here, the fixed value L is arbitrarily determined by experimental data and product specifications.

[0042] During the writing of data to track #n, servo information is read each time the magnetic head 22 passes through a servo sector, and position error information (position error signal) generated from the read servo information is supplied to the controller 30. Each time the controller 30 acquires position error information, it estimates whether the magnetic head 22 is within the range of permitted variation based on the acquired position error information.

[0043] Specifically, the controller 30, while writing to a predetermined track among the multiple tracks 41 (an example of a first track), determines whether the position of the magnetic head 22 has shifted from the target position on the predetermined track to the side of a different track among the multiple tracks 41 (an example of a second track) (either the side in the direction of rotation of the magnetic disk 21 or the side in the opposite direction of rotation), and whether the radial distance from the target position to the position of the magnetic head 22 has exceeded a threshold, i.e., WOS, using multiple position error information obtained by reading servo information from each of a predetermined number of multiple servo regions 42 and a determination model 351 stored in the FRAM 35. Here, obtaining multiple position error information obtained by reading servo information from each of a predetermined number of multiple servo regions 42 from among the multiple servo regions 42 present on the track 41 is referred to as sampling multiple position error information.

[0044] The controller 30 then performs a write operation on the first track if the distance from the target position to the magnetic head 22 does not exceed the WOS (Write-Off Score). On the other hand, the controller 30 does not perform a write operation on the first track if the distance from the target position to the magnetic head 22 exceeds the WOS. Here, the situation where the distance from the target position to the magnetic head 22 exceeds the WOS is referred to as "off-track".

[0045] Figure 4 shows an example of the input and output of the judgment model 351 according to the first embodiment. As shown in Figure 4, the judgment model 351 takes as input multiple sampled position error information (i.e., demodulated values) obtained by reading servo information from each of the multiple servo regions 42, and outputs whether the light is on or off as an output value. In the example in Figure 4, the judgment model 351 takes as input n+1 demodulated values ​​(position error information) and outputs a single output value indicating whether the light is on or off.

[0046] Here, whether writing is permitted or not indicates whether the distance from the target position to the position of the magnetic head 22 exceeds WOS while the magnetic head 22 moves through the data area 43 from the servo area 42 of the current sector to the servo area 42 of the next sector. The determination model 351 outputs an output value (e.g., "1") indicating that writing is not permitted if it determines that the above distance exceeds WOS. The determination model 351 outputs an output value (e.g., "0") indicating that writing is permitted if it determines that the above distance does not exceed WOS.

[0047] The judgment model 351 is a pre-trained model that has been trained using machine learning such as deep learning, using multiple position error information (i.e., demodulated values) obtained by reading servo information from each of the multiple servo regions 42 (sampled), and true values ​​which are the true position error information (i.e., actual positions) between the multiple servo regions 42. Here, since the true values ​​cannot be obtained during operation, they can be obtained by analyzing the recording surface of the magnetic disk 11 after data has been written, or by a simulator 352 that simulates position error information. The labels created from these true values ​​are used as training data when training the judgment model 351.

[0048] The decision model 351 is composed of, for example, a multi-layer perceptron (MLP) neural network. Figure 5 shows an example of the configuration of a decision model 351 composed of a multilayer perceptron neural network according to the first embodiment.

[0049] As shown in Figure 5, the judgment model 351 receives multiple demodulated values, i.e., multiple position error information, into the input layer, passes through an intermediate layer (not shown), and outputs an output value from the output layer indicating whether or not off-tracking is present. The output value for off-tracking is determined when the distance from the target position to the position of the magnetic head 22 exceeds WOS (Work-Off Score), indicating that off-tracking is present (e.g., "1", i.e., equivalent to writing not being possible), and when the distance does not exceed WOS, it is determined when off-tracking is absent (e.g., "0", i.e., equivalent to writing being possible).

[0050] Figure 6 is a diagram illustrating an example of training data for training the decision model 351 according to the first embodiment. In the graph shown in Figure 6, the vertical axis represents the radial position of the magnetic disk 11, and the horizontal axis represents time.

[0051] In the graph of Figure 6, the dotted line 613a represents the positive WOS, and the dotted line 613b represents the negative WOS. The symbols 612, 612a, and 612b indicate multiple position error information (demodulated values) obtained by reading servo information from each of the multiple servo regions 42, i.e., sampled position error information (demodulated values). The solid line 611 indicates the true value, which is the position information (i.e., actual position) between the multiple servo regions 42.

[0052] In this embodiment, the controller 30 performs the training of the decision model 351. In this case, the true value data shown in Figure 6 is prepared in advance. The controller 30 then inputs the multiple demodulated values ​​612, 612a, and 612b shown in Figure 6 to the decision model 351. Here, as shown in Figure 6, the true value at the position of symbol 614 between demodulated values ​​612a and 612b exceeds WOS, and off-track occurs. Here, symbol 615 indicates the off-track occurrence data.

[0053] Therefore, the controller 30 uses the labels created from the demodulated values ​​and true values ​​as training data and calculates the connection weights between each layer constituting the neural network of the decision model 351 using methods such as backpropagation, thereby training the decision model 351 so that the output value indicates that there is off-track activity.

[0054] Furthermore, the decision model 351 is not limited to using, for example, a multilayer perceptron neural network. For example, the decision model 351 may be constructed using a recurrent neural network (RNN).

[0055] Figure 7 shows an example of the configuration of the decision model 351, which is composed of a recurrent neural network according to the first embodiment. As shown in Figure 7, the judgment model 351 receives multiple demodulated values, i.e., multiple position error information, as input to the input layer. After passing through an intermediate layer (not shown), the output layer outputs whether or not off-tracking is present as an output value. As shown in Figure 7, the output of each layer is output to the next layer as well as input to its own layer. The output value, whether or not off-tracking is present, is the same as that of the judgment model 351 using the multilayer perceptron neural network configuration shown in Figure 5.

[0056] Furthermore, the decision model 351 can also be constructed using a Long Short-Term Memory (LSTM) network.

[0057] Next, the control process by the magnetic disk device 1 configured according to this embodiment will be described. Figure 8 is a flowchart showing an example of the control process procedure according to the first embodiment.

[0058] First, the controller 30 causes the magnetic head 22 to read the servo information of the current servo region 42, obtains position error information (demodulated value), and updates the input table for the judgment model 351 (S101). Next, the controller 30 inputs the multiple acquired position error information (demodulated values) contained in the input table to the judgment model 351 (S102).

[0059] Next, the controller 30 obtains the output value from the judgment model 351 (S103). Then, the controller 30 determines whether the output value from the judgment model 351 indicates that writing is permitted (S104). If the output value indicates that writing is permitted (S104: Yes), the controller 30 determines that no off-track has occurred and writes the data to the data area up to the next servo sector (S105).

[0060] On the other hand, if the output value in S104 indicates that writing is not possible (S104: No), the controller 30 determines that an off-track event has occurred and does not perform data writing.

[0061] The controller 30 then determines whether processing has been completed for all servo regions 42 of track 41 (S106). If processing has not been completed for all servo regions 42 of track 41 (S106: No), the controller 30 repeatedly executes the processes from S101 to S105.

[0062] On the other hand, if processing is completed for all servo regions 42 of track 41 (S106:Yes), the process ends.

[0063] Next, we will explain the training process of the judgment model 351 by the controller 30. Figure 9 is a flowchart showing an example of the learning process procedure for the decision model 351 according to the first embodiment.

[0064] First, the controller 30 obtains the true value of the position error information using the method described above (S201). Next, the controller 30 causes the magnetic head 22 to read the servo information for each of the multiple servo regions 42 and acquires multiple position error information (demodulated values) (S202). Then, the controller 30 inputs the acquired multiple position error information (demodulated values) into the judgment model 351 (S203).

[0065] Next, the controller 30 obtains the output value from the judgment model 351 (S204). Then, the controller 30 uses the true value obtained in S201 to train the judgment model 351 (S205).

[0066] The controller 30 then determines whether the learning has converged (S206). If the learning has not converged (S206: No), the controller 30 repeatedly executes the processes from S201 to S205.

[0067] On the other hand, if it is determined that the learning has converged (S206:Yes), the process terminates.

[0068] In the comparative example, the system used methods such as multiple regression to determine whether the position at the next sampling point would exceed WOS (Work-Over Stop) based on noisy position error information (i.e., demodulated values) obtained at each sampling period of multiple servo information, the speed calculated by the control system based on the demodulated values, and the estimated position information at the next sampling point. As a result, in the comparative example, the determination accuracy was insufficient, or it was difficult to detect off-track even when the position between the sampled servo regions exceeded WOS, i.e., when off-track occurred.

[0069] In contrast, in the magnetic disk device according to this embodiment, the controller 30 determines whether the position of the magnetic head 22 has shifted from the target position on the first track to the side of the second track, which is different from the first track, during the execution of writing to the first track among the multiple tracks 41, and whether the distance from the target position to the position of the magnetic head 22 has exceeded WOS. At this time, the determination is made using multiple position error information obtained by reading servo information from each of the multiple servo regions 42, and a determination model 351, which is a trained model that takes the multiple position error information as input and outputs whether the above distance has exceeded WOS while the magnetic head 22 moves through the multiple servo regions 42. If the above distance does not exceed WOS, writing to the first track is performed, and if the above distance exceeds WOS, writing to the first track is not performed.

[0070] Therefore, in this embodiment, the WOS (Wave-Off Score) can be exceeded at a position between the servo regions 42 using the determination model 351. This allows for high-precision detection of off-track and appropriate control of the operation of the magnetic head 22.

[0071] Figure 10 shows an example of the difference in effects between the first embodiment and the comparative example. In Figure 10, the left side shows a comparative example using the multiple regression method, and the right side shows an example of this embodiment using the judgment model 351. The horizontal axis shows the evaluation metrics for the comparative example and this embodiment, namely, accuracy, recall, precision, and specificity. The vertical axis shows the percentages for each evaluation metric.

[0072] Here, accuracy is an indicator that shows the proportion of correct predictions out of all predictions made by the judgment model 351. Furthermore, recall is an indicator that represents the proportion of times the model correctly predicted a positive outcome among the actual positive outcomes. In this embodiment, recall represents the proportion of times the judgment model 351 correctly detected off-track outcomes among the actual off-track outcomes; a higher recall rate means fewer off-track outcomes were missed.

[0073] Precision is an index that represents the proportion of predictions that the model predicted to be positive that were actually correct. In this embodiment, Precision is the proportion of predictions that the decision model 351 predicted to be off-track that were actually off-track.

[0074] Specificity is an index that represents the proportion of predictions that the model predicted to be false that were actually false. In this embodiment, it is the proportion of predictions that the decision model 351 predicted to be not off-track that were actually not off-track.

[0075] In this embodiment, compared to the comparative example, an improvement in accuracy and a high recall rate are observed, indicating that the number of missed off-track events between servo regions 42 has decreased, and that off-track events can be predicted with high accuracy.

[0076] Furthermore, in the magnetic disk device 1 according to this embodiment, the determination model 351 is a trained model learned using multiple position error information and true values ​​which are the actual positions between multiple servo regions 42. Therefore, according to this embodiment, by using a trained model, off-track detection can be performed with higher accuracy, and the operation of the magnetic head 22 can be controlled more appropriately.

[0077] (Second embodiment) In the first embodiment, the method for generating training data was not particularly limited, but in this second embodiment, training data is generated using a simulator.

[0078] The configuration of the magnetic disk device 1 and magnetic disk 11 according to this embodiment is the same as in the first embodiment.

[0079] Figure 11 is a control block diagram showing an example of the configuration of the simulator 352 used in the second embodiment. The simulator 352 of this embodiment comprises a controller C[z], a zero-order hold circuit ZOH, a controlled object P(s), and a sampler S with a sampling period T. T It primarily features and .

[0080] Controller C[z] receives the difference between the target position and the demodulated value (position error information) that has been fed back, and outputs the estimated speed of the magnetic head 22, the estimated position of the magnetic head 22 in the servo region of the next sampling, and the control output. The zero-order hold circuit ZOH converts the discrete output of the control output from controller C[z] into a continuous value. Sampler S T This device takes the true value obtained by adding the output from P(s) and the disturbance d1(t) as input, and outputs it as a discrete value. d1(t) is a disturbance such as the server rack on which the magnetic disk drive 1 is installed. d2(t) is noise relative to the demodulated value.

[0081] The controller 30 of this embodiment determines whether the distance from the target position to the position of the magnetic head 22 exceeds the previous WOS using multiple position error information, the estimated speed of the magnetic head 22 output from the simulator's controller C[z], the estimated position of the magnetic head 22 in the next sampled servo region output from the controller C[z], and the determination model 351.

[0082] The determination model 351 of this embodiment is a trained model that has been learned using multiple position error information (demodulated values), the true value calculated by the simulator, the estimated speed of the magnetic head 22 output from the simulator's controller C[z], and the estimated position of the magnetic head 22 in the next sampled servo region, also output from the controller C[z]. The other functions and configuration of the determination model 351 are the same as in the first embodiment.

[0083] Figure 12 shows an example of the input and output of the judgment model 351 according to the second embodiment. As shown in Figure 12, the judgment model 351 takes as input multiple position error information (i.e., demodulated values) obtained by reading servo information from each of the multiple servo regions 42, the estimated position, and the estimated speed, and outputs whether the lights are on or off as an output value. Here, the determination of whether the lights are on or off is the same as in the first embodiment.

[0084] Next, the control process by the magnetic disk device 1 configured according to this embodiment will be described. Figure 13 is a flowchart showing an example of the control process procedure according to the second embodiment.

[0085] First, the controller 30, similar to the first embodiment, causes the magnetic head 22 to read the servo information of each of the multiple servo regions 42 and acquires multiple position error information (demodulated values) (S101).

[0086] Next, the controller 30 obtains the estimated speed of the magnetic head 22 output from the simulator's controller C[z] and the estimated position of the magnetic head 22 in the next sampled servo region, also output from the controller C[z] (S301). Then, the controller 30 inputs the acquired position error information (demodulated values), the estimated position, and the estimated speed into the judgment model 351 (S302).

[0087] Next, the controller 30 obtains the output value from the judgment model 351 (S103). The subsequent processing from S104 to S106 is carried out in the same manner as in the first embodiment.

[0088] Next, we will explain the training process of the judgment model 351 by the controller 30. Figure 14 is a flowchart showing an example of the learning process procedure for the decision model 351 according to the second embodiment. The acquisition of the true value of the position error information (S201) and the acquisition of multiple position error information (demodulated values) (S202) are performed in the same manner as in the first embodiment.

[0089] Next, the controller 30 obtains the estimated speed of the magnetic head 22 output from the simulator's controller C[z] and the estimated position of the magnetic head 22 in the next sampled servo region, also output from the controller C[z] (S401). Then, the controller 30 inputs the acquired position error information (demodulated values), the estimated position, and the estimated speed into the judgment model 351 (S402).

[0090] Next, the controller 30 obtains the output value from the judgment model 351 (S204). The subsequent processing in S205 and S206 is carried out in the same manner as in the first embodiment.

[0091] In the magnetic disk device 1 according to this embodiment, the controller 30 determines whether the distance from the target position to the position of the magnetic head 22 exceeds WOS using multiple position error information, the estimated speed of the magnetic head estimated by the simulator, the estimated position of the magnetic head in the next sampled servo region 42, and the determination model 351.

[0092] Therefore, according to this embodiment, since the WOS is exceeded at a position between multiple servo regions 42 using the determination model 351, off-track detection can be performed with high accuracy, and the operation of the magnetic head 22 can be appropriately controlled.

[0093] Furthermore, in the magnetic disk device 1 according to this embodiment, the trained model is learned using multiple position error information (demodulated values), the true value calculated by the simulator, the estimated speed, and the estimated position. Therefore, according to this embodiment, by using the trained model, off-track detection can be performed with higher accuracy, and the operation of the magnetic head 22 can be controlled more appropriately.

[0094] In the above embodiment, the learning process for the judgment model 351 was performed by the controller 30 of the magnetic disk drive 1, but this is not the only option. The judgment model 351 can be installed in a location other than the magnetic disk drive 1, and the learning process for the judgment model 351 can be performed on a computer or the like to create a trained model, which can then be mounted on the magnetic disk drive 1.

[0095] In the magnetic disk device 1 according to the above embodiment, the case where the writing method is CMR (Conventional Magnetic Recording) was used as an example for explanation, but the same can be applied to the case of SMR (Shingled Magnetic Recording).

[0096] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of Symbols]

[0097] 1,1a,1b Magnetic disk drive, 11 Magnetic disk, 12 Spindle motor, 13 Lamp, 15 Actuator arm, 21 Motor driver IC, 22 Magnetic head, 22r Read element, 22w Write element, 23 HDC, 24 Head IC, 25 RWC, 26 Processor, 27 RAM, 28 FROM, 29 Buffer memory, 30 Controller, 42 Servo area, 43 Data area.

Claims

1. A magnetic disk having multiple tracks, including a data area and a servo area on which servo information is recorded, A magnetic head that performs read / write operations on the aforementioned multiple tracks, Equipped with a controller, The aforementioned controller, During the execution of a light on the first track among the plurality of tracks, the position of the magnetic head shifts from the target position on the first track to the side of a second track different from the first track among the plurality of tracks, and a determination model is used to determine whether the radial distance from the target position to the position of the magnetic head exceeds a threshold. If the distance does not exceed the threshold, light is applied to the first track. If the distance exceeds the threshold, the light to the first track is not performed. The determination model is a trained model that takes multiple position error pieces obtained by reading the servo information from each of the multiple servo regions as input and outputs whether or not the distance exceeds the threshold while the magnetic head moves through the multiple servo regions. Magnetic disk drive.

2. The aforementioned determination model is a trained model that has been learned using the plurality of position error information and a first true value which is the actual position of the magnetic head between the plurality of servo regions. The magnetic disk device according to claim 1.

3. The controller determines whether the distance exceeds the threshold using the plurality of position error information, the estimated speed of the magnetic head estimated by the simulator, the estimated position of the magnetic head in the next servo region, and the determination model. The magnetic disk device according to claim 2.

4. The aforementioned decision model is a trained model that has been trained using the plurality of position error information, the second true value calculated by the simulator, the estimated speed, and the estimated position. The magnetic disk device according to claim 3.

5. The aforementioned decision model is composed of a multilayer perceptron (MLP) neural network. The magnetic disk device according to claim 1.

6. The aforementioned decision model is composed of a recurrent neural network (RNN). The magnetic disk device according to claim 1.

7. The aforementioned decision model is composed of a Long Short Term Memory (LSTM) network. The magnetic disk device according to claim 1.

8. A control method for controlling a magnetic disk device, The magnetic disk device is A magnetic disk having multiple tracks, including a data area and a servo area on which servo information is recorded, The system includes a magnetic head that performs read / write operations on the aforementioned multiple tracks, The control method described above is During the execution of a light on the first track among the plurality of tracks, the position of the magnetic head shifts from the target position on the first track to the side of a second track different from the first track among the plurality of tracks, and a determination model is used to determine whether the radial distance from the target position to the position of the magnetic head exceeds a threshold. If the distance does not exceed the threshold, light is applied to the first track. This includes not performing a light on the first track if the distance exceeds the threshold, The determination model is a trained model that takes multiple position error pieces obtained by reading the servo information from each of the multiple servo regions as input and outputs whether or not the distance exceeds the threshold while the magnetic head moves through the multiple servo regions. Control method.

Images