Inspection method, inspection device, and inspection system

Abstract

An inspection method according to one aspect of the present disclosure is for inspecting a wiring board P by applying an inspection voltage or an inspection current to an inspection target X. The inspection target X is a part of the wiring board P that comprises an insulator and a conductor and on which an electric component is mounted. The inspection target is subject to a plurality of inspections, and the inspections include a first inspection for inspecting the conduction state of the conductor or the insulator, and a second inspection for inspecting time-series data of the conductive state of the conductor or the insulator. The result of the first inspection and the result of the second inspection are recorded in a recording device. A third inspection is further included, which is carried out after the first inspection and the second inspection, in a state where processing is performed and an electric component is mounted in a subsequent process. The result of the third inspection is recorded in association with the result of the first inspection and the result of the second inspection.

Description

Inspection method, inspection device, and inspection system

[0001] The present disclosure relates to an inspection method, an inspection device, and an inspection system.

[0002] In a wiring board having multiple wiring patterns, an inspection is performed to determine whether the continuity and insulation of the wiring board are good or bad by applying an inspection voltage or passing an inspection current to each wiring pattern and measuring the resistance value from the voltage and current values ​​generated in the wiring pattern.

[0003] Although measuring such resistance values ​​alone may satisfy the required resistance, abnormalities such as abnormal transient responses may be overlooked. For this reason, a known inspection method acquires transient responses and determines a fault if the peak current value during inspection is equal to or greater than a threshold (see Patent Document 1). This inspection method can detect sparks, which are a phenomenon in which a potential difference between circuit patterns causes dielectric breakdown and a large current to instantaneously flow between the circuit patterns.

[0004] Furthermore, with improvements in inspection methods and methods for managing inspection results, inspection results are now classified during inspection and a database is constructed. Machine learning models are used to classify the inspection results, thereby improving inspection accuracy. For example, a method using a machine learning model for inspection is known in which attributes of images of wiring boards are classified by a trained classifier, and images belonging to a specific classification are further analyzed to improve defect detection accuracy (Patent Document 2).

[0005] JP 2021-169952 A JP 2021-148678 A

[0006] Known defects such as sparks can be detected based on predefined rules. However, it is not realistic to predefine many defect events and implement their detection methods individually, and there is a significant possibility that defects will be overlooked. Therefore, it becomes necessary to define other detection methods.

[0007] Furthermore, in a database that classifies and stores the results of image inspection, defective images in the process are classified, but it is not possible to improve the accuracy of the inspection results of the inspection equipment in the process in relation to the inspection results of the inspection equipment in the subsequent process, or to re-learn them.

[0008] The present disclosure has been made in light of the above circumstances, and aims to provide an inspection method, an inspection device, and an inspection system that reduce the possibility of overlooking defects when inspecting wiring boards.

[0009] (1) An inspection method according to one aspect of the present disclosure is an inspection method for inspecting a wiring board by applying an inspection voltage or passing an inspection current through an object to be inspected, the object to be inspected being a part of a wiring board made of an insulator and a conductor and on which an electrical component is mounted, the inspection method employing a plurality of inspections, the inspections including a first inspection for inspecting the electrical conduction state of the conductor or the insulator, and a second inspection for inspecting time-series data of the electrical conduction state of the conductor or the insulator, the results of the first inspection and the second inspection being recorded in a recording device, the inspection method further comprising a third inspection that is performed after the first inspection and the second inspection, with the electrical component mounted and processed in a subsequent process, the result of the third inspection being recorded in association with the results of the first inspection and the second inspection.

[0010] (2) In the inspection method of (1), the results of the second inspection recorded in the recording device may be rewritten with information on the defective state of the wiring board determined in the third inspection.

[0011] (3) In the inspection method of (2), the second inspection may use a machine learning model that has been trained to pass time-series data that has been determined to be a non-defective product based on the results of the first inspection.

[0012] (4) In the inspection method of (3), the second inspection may involve re-training the machine learning model using time series data excluding time series data containing defect information determined in the third inspection.

[0013] (5) In the inspection method of (4), the recording device may record at least one of the production information items including temperature, humidity, radio wave conditions, monitoring images, noise, vibration, number of inspections, number of contacts, and part usage time, which share the time information of the first inspection or the second inspection.

[0014] (6) In the inspection method of (5), the machine learning model used in the second inspection may be updated according to the third inspection result, and the first inspection conditions of the inspection device that inspects the wiring board may be optimized by referring to the third inspection result and the production information.

[0015] (7) In the inspection method of (6), the recording device may record the results of the first inspection and the second inspection according to the wiring pattern and insulating area of ​​the conductor, and record design information regarding the wiring pattern and the insulating area.

[0016] (8) In the inspection method of (7), the design information may include, in addition to the design of the wiring pattern and the insulating area, production information regarding at least one of the distance between wirings, through-hole information, resist area information, materials used, manufacturing factory, transportation distance, manufacturing time, parasitic capacitance, and inspection equipment number.

[0017] (9) In any of the inspection methods (1) to (8), the results of the second inspection may be rewritten using defective state information classified by an interface provided in the device performing the third inspection.

[0018] (10) An inspection system according to another aspect of the present disclosure is an inspection system that inspects a wiring board by applying an inspection voltage or passing an inspection current through an object to be inspected, wherein the object to be inspected is a part of a wiring board made of an insulator and a conductor and on which an electrical component is mounted, and the inspection system includes: an inspection device that includes a first inspection unit that inspects the electrical state of the conductor or the insulator; a second inspection unit that inspects time-series data on the electrical state of the conductor or the insulator; and a recording device that records the results of the first inspection unit and the second inspection unit; and another inspection device that includes a third inspection unit that inspects the wiring board after inspection by the first inspection unit and the second inspection unit, after processing in a subsequent process and with an electrical component mounted thereon, and the results of the third inspection unit are recorded in association with the results of the first inspection and the second inspection unit.

[0019] (11) An inspection device according to yet another aspect of the present disclosure is an inspection device that inspects a wiring board based on waveform data of an inspection object obtained from the inspection object by applying an inspection voltage or passing an inspection current, and uses a plurality of parameterized inspection conditions. For each inspection condition, the inspection device includes: a machine learning model that collects waveform data obtained from the inspection object in advance and trains the model based on the collected waveform data; an acquisition unit that acquires the waveform data and the inspection conditions obtained from the inspection object of the wiring board to be inspected; and a judgment unit that inputs the waveform data and the inspection conditions acquired by the acquisition unit into the machine learning model and judges whether the inspection object is good or bad based on the output result.

[0020] In the inspection method and system, in the voltage or current inspection of a wiring board, the results of a first inspection that inspects the current flow state and a second inspection that inspects time-series data on the current flow state of a conductor or insulator are recorded in a recording device and stored in the recording device together with the results of a third inspection, which is a subsequent process, so that the time-series data on the current flow state can be linked to the inspection results of the subsequent process.The time-series data on the current flow state of a wiring board may contain information on potential defects in the wiring board that cannot be determined by the quality of the current flow state, and by linking it to the inspection results of the subsequent process, it becomes possible to determine defects that cannot be determined by the quality of the current flow state.

[0021] FIG. 1 is an explanatory diagram of a configuration of a wiring board and Gerber data. FIG. 2 is an explanatory diagram showing an example of an inspection target. FIG. 3 is a block diagram of an inspection device included in an inspection system according to an embodiment of the present disclosure. FIG. 4 is a configuration diagram of another inspection device included in an inspection system according to an embodiment of the present disclosure. FIG. 5 is a configuration diagram of a first inspection device. FIG. 6 is a functional block diagram of the first inspection device of FIG. 5. FIG. 7 is a configuration diagram of the switch circuit of FIG. 6 when inspecting the inspection target of FIG. 3. FIG. 8 is a configuration diagram of an inspection device according to an embodiment of the present disclosure. FIG. 9 is a configuration diagram of a machine learning model used in the inspection device of FIG. 8. FIG. 10 is an explanatory diagram explaining processing of the autoencoder of FIG. 9. FIG. 11 is a schematic cross-sectional view of a wiring board explaining a defect example in the wiring board. FIG. 12 is an image showing an example of a defect classification method in another inspection device. FIG. 13 is an explanatory diagram explaining an example of the influence of a defect on the waveform diagnosis result of the inspection device. FIG. 14 is an explanatory diagram explaining an example, different from FIG. 13, of the influence of a defect on the waveform diagnosis result of the inspection device. FIG. 15 is an explanatory diagram showing a method of operating an inspection system using the inspection device of FIG. 3 and another inspection device of FIG. 4. FIG. 16 is an image diagram showing an example of a user interface of the inspection system of FIG. 15. FIG. 17 is an explanatory diagram showing an application example of a method of operating the inspection system. FIG. 18 is an explanatory diagram showing an example of an automatically generated inspection report. FIG. 19 is an example of a display of inspection results. FIG. 20 is an explanatory diagram explaining data clustering using Gerber data. FIG. 21 is a diagram showing the use of inspection results when designing a board. FIG. 22 is a graph explaining a method of detecting deterioration over time of consumables.

[0022] Hereinafter, an embodiment of the present disclosure will be described in detail with reference to the drawings as appropriate.

[0023] An inspection method according to one embodiment of the present disclosure is an inspection method for inspecting a wiring board by applying an inspection voltage or passing an inspection current to an object to be inspected, and can be performed using an inspection system for inspecting a wiring board by applying an inspection voltage or passing an inspection current to an object to be inspected.

[0024] <Wiring Board> Electrical components are mounted on the wiring board. Examples of the wiring board include flexible printed circuits, printed circuit boards, and HDI (High Density Interconnect) boards that use high-density wiring technology.

[0025] 1, the wiring board P is a double-sided board including materials used in the manufacturing process, and is composed of, from the top down, component side silk P1, component side resist P2, component side wiring pattern P3 (conductor), substrate P4 including vias (conductor), solder side resist P5, solder side wiring pattern P6 (conductor), and solder side silk P7.

[0026] The wiring board P may include a multi-layered substrate, and may have vias for connecting conductors, ground layers connected by vias to some of the layer structure, and dummy wiring to suppress warping of the board. The flexible board has wiring (conductors) made of copper foil patterned with resist on a polyimide film (insulator), and a reinforcing plate is attached using a polyimide film or adhesive to increase rigidity.

[0027] 2, the inspection object X is a part of the wiring substrate P, such as a wiring pattern or an insulator between wiring patterns. There may be one inspection object X, but typically there are multiple inspection objects X, and FIG. 2 shows four inspection objects X: a first inspection object X1, a second inspection object X2, a third inspection object X3, and a fourth inspection object X4.

[0028] The test voltage applied to or test current passed through the test object X is not particularly limited as long as it can test the quality of the wiring board P through the test object X, but examples include direct current, alternating current, frequency signals such as sine waves, rectangular waves, trapezoidal waves, and frequency signals (pulse waves).

[0029] Furthermore, examples of things measured in the inspection of the wiring board P include temporal changes in input signals, current, voltage, power, waveform signals, frequency signals, their temporal and time-series fluctuations, the DC resistance, impedance, inductance of the object to be measured, their temporal and time-series fluctuations, average value, variance, coefficient of variation, or temporal and time-series fluctuations of the average value, variance, and coefficient of variation, frequency signals, values ​​obtained by Fourier transforming their temporal and time-series fluctuations, measured time information, etc. These measurements can be performed through continuity tests, insulation tests (spark detection), impedance measurements, circuit network analysis, TDR (time domain reflectometry), high-frequency tests, etc.

[0030] [Inspection Method] The inspection method includes a first inspection for inspecting the electrical conduction state of the conductor or the insulator, and a second inspection for inspecting time-series data of the electrical conduction state of the conductor or the insulator, the results of the first inspection and the second inspection being recorded in a recording device, and further includes a third inspection for inspecting the wiring substrate P with the electrical components mounted thereon after the first inspection and the second inspection, and the result of the third inspection is linked to the result of the second inspection and recorded in the recording device.

[0031] The inspection method will be described in detail below, along with the configuration of the inspection system.

[0032] [Inspection System] The inspection system includes an inspection device 10 (see FIG. 3 ) including a first inspection unit that inspects the electrical state of the conductor or the insulator, a second inspection unit that inspects time-series data on the electrical state of the conductor or the insulator, and a recording device that records the results of the first inspection unit and the second inspection unit, and another inspection device 11 (see FIG. 4 ) that includes a third inspection unit that inspects the wiring substrate P after the inspections by the first inspection unit and the second inspection unit, after the wiring substrate P has been processed in a subsequent process and has electric components mounted thereon, and the results of the third inspection unit are recorded in association with the results of the first inspection and the second inspection unit.

[0033] The inspection device 10 includes an electrical current inspection unit 101, a frequency inspection unit 102, a waveform inspection unit 103, an inspection condition information acquisition unit 104, a design information acquisition unit 105, an image signal acquisition unit 106, an environmental information acquisition unit 107, a communication unit 108, an input / output unit (I / O) 109, a display unit 110, a clock generator, and a recording device for recording various inspection results.

[0034] <First Inspection Unit> The first inspection performed by the first inspection unit corresponds to an electrical inspection and a frequency inspection. For the electrical inspection, for example, a first inspection device 12 shown in FIG. 5 can be used.

[0035] The first inspection device 12 includes a wiring board holder 21 , a measurement unit 22 (upper measurement unit 22 A and lower measurement unit 22 B), a determination unit 23 , and a display unit 24 .

[0036] The wiring board holder 21 holds a wiring board sheet S on which a plurality of wiring boards P (4 × 4 = 16 sheets in FIG. 5) are formed, and the upper measurement unit 22A and the lower measurement unit 22B are driven to be movable at least in directions parallel to the surfaces of the wiring boards P, i.e., in the x, y, and θ directions shown in FIG. 5. The wiring board holder 21 may be equipped with an automatic sheet feeder that automatically takes in and discharges the wiring board sheets.

[0037] The upper measurement unit 22A is located above the front surface of the wiring substrate P and has a measurement unit main body 22a, a plurality of pogo pins or wire probes 22b, and a camera 22c. The lower measurement unit 22B is located below the back surface of the wiring substrate P and has a measurement unit main body 22a, a plurality of pogo pins or wire probes 22b, and a camera 22c.

[0038] As shown in FIG. 6, the measurement unit main body 22a of the upper measurement unit 22A and the lower measurement unit 22B includes a switch circuit 221 that constitutes a circuit for applying a test voltage or flowing a test current to the test object X of the wiring board P, a voltmeter 222 for measuring the voltage applied to the test object X, an ammeter 223 for measuring the current flowing through the test object X, and a variable voltage generator 224 for controlling the test voltage.

[0039] The multiple pogo pins or wire probes 22b of the upper measurement unit 22A and the lower measurement unit 22B are arranged on the lower surface (the surface facing the wiring board P) or upper surface (the surface facing the back surface of the wiring board P) of the measurement unit main body 22a. The measurement unit main bodies 22a of the upper measurement unit 22A and the lower measurement unit 22B are configured to be movable in the x-direction, y-direction, and also in the up-down direction (z-direction in FIG. 5). By bringing the measurement unit main body 22a close to the surface of the wiring board P, the multiple pogo pins or wire probes 22b of the upper measurement unit 22A and the lower measurement unit 22B come into contact with the front and back surfaces of the wiring board P so as to sandwich them between them, and are configured to make electrical contact when they come into contact with conductors on the surface of the wiring board P. When measuring conductors on the wiring board, in addition to two-contact testing, a four-terminal method can be used.

[0040] When the measurement unit main body 22a is brought close to the surface of the wiring board P, the camera 22c captures an image of the wiring board P near the position where at least the plurality of pogo pins or wire probes 22b come into contact with the surface of the wiring board P. This image makes it possible to easily bring the plurality of pogo pins or wire probes 22b into contact with the desired positions on the surface of the wiring board P. In other words, while checking the image from the camera 22c, the wiring board sheet S is moved so as to make contact with the inspection object X, and the plurality of pogo pins or wire probes 22b are brought into contact with the surface of the wiring board P, thereby enabling the inspection of the target inspection object X. Furthermore, the image of each wiring board may be saved together with the electrical inspection results, and a pass / fail judgment may be made simultaneously by image inspection.

[0041] The camera 22c may be, for example, an imaging element such as a CCD (Charge Coupled Device) sensor or an infrared camera. The imaging information obtained by capturing an image of the state of the wiring board before or after the first or second inspection can be recorded in the recording device. The imaging information may include an infrared image.

[0042] In the first inspection device 2 shown in Figure 5, measurement units are provided on both surfaces (front surfaces) of the wiring substrate P, but in cases where the wiring substrate P is a single-sided substrate, they may be provided on one surface. Also, although the upper measurement unit 22A and the lower measurement unit 22B are movable in the x, y, and θ directions, the wiring substrate holder 21 may be driven to be movable in the x, y, and θ directions, and the upper measurement unit 22A and the lower measurement unit 22B may be configured to be movable in the vertical direction (z direction in Figure 5). Also, a so-called flying probe system may be used in which several probes move independently to perform inspection.

[0043] The determination unit 23 is connected to the measurement unit main bodies 22a and 23a by a cable K, and controls the measurement unit main bodies 22a and 23a, receives the measurement results of the measurement unit main bodies 22a and 23a, and determines whether the current supply state is good or bad.

[0044] 6, the determination unit 23 includes a switch changeover unit 231, an applied power source setting unit 232, a voltage detection unit 233, a voltage rise completion detection unit 234, a current detection unit 235, a current increase detection unit 236, a detection time setting unit 237, and a control unit 238. Although there is only one determination unit 23 for convenience of illustration, a determination unit may be provided for each measurement unit.

[0045] The switch changeover unit 231 controls the switch circuit 221. The switch changeover unit 231 selects the inspection object X of the wiring board P and configures a circuit required for the inspection.

[0046] The applied power setting unit 232 controls the applied power of the variable voltage generator 224 , and the applied power of the variable voltage generator 224 is set via the control unit 238 .

[0047] The voltage detection unit 233 converts the analog signal measured by the voltmeter 222 into a digital signal, and includes an A / D converter.

[0048] The voltage increase completion detection unit 234 determines that the voltage has increased when the voltage signal output from the voltage detection unit 233 reaches the voltage value set by the applied power supply setting unit 232, and issues a voltage increase completion signal.

[0049] The current detection unit 235 converts the analog signal measured by the ammeter 223 into a digital signal and includes an A / D converter. The output signal of this A / D converter is also connected to a waveform inspection unit, which will be described later.

[0050] The current increase detection unit 236 extracts the current signal output from the current detection unit 235 at a predetermined sampling interval, and if the current sampled value increases or decreases by more than a predetermined current value compared to the previous sampled value, it determines that there is an abnormality in the current and issues a current abnormality signal.

[0051] The detection time setting unit 237 sets a peak current value for detecting whether or not a spark occurs. The detection time is provided in advance from an external source. The control unit 238 determines whether or not a current increase signal is generated by the current increase detection unit 236 within this detection time after the voltage increase completion signal is generated by the voltage increase completion detection unit 234. Here, a spark refers to a momentary flow of electricity that occurs between circuit patterns separated by a space (air) due to a potential difference between the circuit patterns that occurs when inspecting the insulation state between the circuit patterns on the wiring board P.

[0052] The control unit 238 has a CPU, a memory, an interface for data communication, etc., and controls the determination unit 23. Specifically, the control unit 238 performs a continuity test and an insulation test on the test object X of the wiring substrate P. The insulation test includes a test for the occurrence of sparks.

[0053] The display unit 24 displays the inspection conditions imposed on the inspection object X, the pass / fail result of the inspection object X determined by the control unit 238, and the like.

[0054] <First Inspection> The first inspection (continuity inspection and insulation inspection) performed using the first inspection unit will be described below using the inspection object X shown in Fig. 2 as an example. The first inspection is performed before electrical components are mounted on the wiring substrate P.

[0055] 7 shows an example of the configuration of the switch circuit 221 when inspecting the inspection objects X (first inspection object X1, second inspection object X2, third inspection object X3, and fourth inspection object X4) shown in Fig. 2. Note that the switch circuit 221 is an example and may be different if the configuration of the inspection object X is different, and does not prevent the adoption of other configurations even if the inspection object X is the same.

[0056] The switch circuit 221 has a +-side main switch A0 connected to the + side of the variable voltage generator 224 and a --side main switch B0 connected to the - side (ground side). The +-side main switch A0 is connected to the circuit patterns, which are the conductor portions of the four test objects X, via switches A1 to A4, and is configured to select one of the first test object X1, the second test object X2, the third test object X3, and the fourth test object X4 using switches A1 to A4. The --side main switch B0 is similarly configured to select one of the first test object X1, the second test object X2, the third test object X3, and the fourth test object X4 using switches B1 to B4. The positions in the circuit patterns connected by switches B1 to B4 may be the same as or different from the positions in the circuit patterns connected by switches A1 to A4. In the switch circuit 221 of FIG. 3, they are connected to the same positions.

[0057] Furthermore, the negative main switch B0 is configured to select one of the first test object X1, the second test object X2, the third test object X3, and the fourth test object X4 via the switches C1 to C42. Unlike the switches B1 to B4, the positions in the circuit pattern connected by the switches C1 to C42 are different from the positions in the circuit pattern connected by the switches A1 to A4, and it is preferable that the path length along the circuit pattern between these positions is longer. In other words, as shown in FIG. 6, it is preferable that different ends of a band-shaped circuit pattern are selected by the combination of the switches A1 to A4 and the test objects X corresponding to the switches C1 to C42. Note that if the circuit pattern has branches, such as the third test object X3 and the fourth test object X4, it is preferable to provide a switch for each branch. In particular, during continuity testing, if a continuity defect exists in a branch not provided with the switches C1 to C42, the defect may be overlooked.

[0058] (Continuity Test) A continuity test is performed by selecting (turning ON) one of the switches A1 to A4 and one of the switches C1 to C42 corresponding to the test object X. For example, for the first test object X1, the switches A1 and C1 connected to it are selected. For the third test object X3, the switch A1 and either the switch C31 or the switch C32 are selected. To ensure a reliable continuity test, it is desirable to test both the combination of the switches A1 and C31 and the combination of the switches A1 and C32.

[0059] Selection of switches A1 to A4 and switches C1 to C42 is performed with the + side main switch A0 and the - side main switch B0 in the non-selected (OFF) state, and then the + side main switch A0 and the - side main switch B0 are turned ON. The selection of subsequent switch procedures is similar unless otherwise noted. The switching of these switches is controlled by the switch switching unit 231.

[0060] After the above switch selection, for example, a predetermined voltage is applied from variable voltage generator 224, and continuity can be determined by whether or not a current of a predetermined value or greater flows in ammeter 223. Alternatively, the resistance value as viewed from the upstream side of switch circuit 221 (the side of the + side main switch A0 and the - side main switch B0) can be measured. In this case, if the resistance is lower than the predetermined resistance value, continuity is determined to be good, and if it is higher, continuity is determined to be poor. In either method, the predetermined voltage, predetermined current value, and predetermined resistance value may be changed or may remain the same depending on the type of circuit under test X.

[0061] The test results are stored in a recording device, which will be described later.

[0062] (Insulation Test) The insulation test is performed by controlling the ON / OFF of switches A1 to A4 and switches B1 to B4 as shown in Table 1. In Table 1, blank spaces mean OFF.

[0063]

[0064] In the first test, a predetermined voltage is applied to the variable voltage device 224, and if continuity (current increase signal) or spark generation is observed, it is determined that there is an insulation defect in at least one location between the first test object X1 and the third test object X3, between the second test object X2 and the third test object X3, between the first test object X1 and the fourth test object X4, or between the second test object X2 and the fourth test object X4. Conversely, if neither continuity nor spark generation is observed, it can be said that there is no insulation defect between these.

[0065] Similarly, the second test determines whether there is an insulation defect between the first test object X1 and the second test object X2, and the third test determines whether there is an insulation defect between the third test object X3 and the fourth test object X4.

[0066] By following the above procedure, it is possible to carry out inspections with a small number of trials for all combinations of the four inspection targets X. Furthermore, similar to the continuity inspection, the inspection results are stored in a recording device.

[0067] (Variation) Instead of conducting continuity and insulation tests using DC voltage and current, the first inspection unit can also use the frequency inspection unit 102 to perform tests using a frequency signal (pulse wave) as an input signal. In this case, the upper measurement unit 22A and the lower measurement unit 22B have inspection probes that are connected to a network analyzer or the like. The upper measurement unit 22A and the lower measurement unit 22B move so as to sandwich the wiring board to be inspected, bringing multiple inspection probes into contact with the test object, inputting a pulse wave inspection signal from one end and detecting an output signal from the other end. By analyzing the detected signals, it is possible to inspect frequency characteristics such as the quality of the board and how well the board can handle high-speed signals. The pass / fail judgment is made by measuring the frequency characteristics of the passing power and reflected power of the path. In addition, it is possible to measure characteristics such as insertion loss, return loss, VSWR (voltage standing wave ratio), isolation, characteristic impedance, TDR (time domain reflectometry) characteristics, transfer characteristic S parameters, isolation between the circuit and terminal, gain, insertion phase difference, and group delay of a conductor pattern on a circuit board to be tested, such as a high-frequency circuit board. To determine whether the circuit is compatible with high-speed signals, test conditions such as test time and threshold value are defined for each characteristic, and the result of the first test can be used to determine whether the circuit is compatible with high-speed signals. Simultaneously with saving the results of the first test, it is also possible to test the time waveform of the pulse wave used in the test (a second test, described below) and save the test results.

[0068] That is, a test frequency signal can be applied instead of a test voltage or test current.

[0069] Furthermore, for DC signals, the input signal can be a periodic square wave or trapezoidal wave to clarify the degree of change. In this case, a first test can be performed simultaneously, in which the pass / fail of the test object is determined by comparing the voltage and current changes at the rising edge of the square wave or trapezoidal wave with a threshold value, and a second test (described later) can be performed by using the time waveform changes at the rising and falling edge as the diagnostic target.

[0070] The result of the first inspection (pass / fail) is saved along with the sheet number of the sheet on which the wiring board P to be inspected is mounted, the wiring board number, inspection conditions, inspection circuit pattern number, adjacent circuit pattern number, clearance setting, wiring length, wiring width, minimum clearance length, maximum bending angle, a wiring board inspection management number that is uniquely determined for each inspection, inspection type (continuity, insulation, frequency), time, inspection voltage, etc. The wiring board inspection management number may be a combination of a uniquely determined wiring board number and an inspection serial number.

[0071] Defects discovered by the first inspection include poor conductivity due to foreign matter in the insulating region, poor conductivity due to disconnections, poor through-hole connections, unintended conductivity through spaces created by peeling of the insulating region, poor connections due to peeling caused by insufficient removal of cutting debris (smear), and increased resistance due to disconnection of some of the multiple vias connected to the ground layer formed in the intermediate layer of a multilayer board.Even if no abnormalities are found when inspecting the wiring board before components are mounted, defects such as peeling may grow during subsequent reflow processes, resulting in growth defects (migration defects).

[0072] <Second Inspection Unit> The second inspection corresponds to a waveform inspection. The second inspection unit performs the second inspection of the inspection method, i.e., inspects time-series data of the current-carrying state of the conductor or insulator. The second inspection unit may be a second inspection device 13 shown in FIG. 8.

[0073] 8 is an inspection device according to another aspect of the present disclosure, which inspects a wiring substrate P based on waveform data D of an inspection object X on the wiring substrate P obtained by applying an inspection voltage or passing an inspection current therethrough. The second inspection device 13 includes a measurement unit 31, an acquisition unit 32, a determination unit 33, a machine learning model 34, and a display unit 35. The measurement unit 31, the determination unit 33, and the display unit 35 of the second inspection device 13 may also serve as part or all of the measurement unit 22, the determination unit 23, and the display unit 24 of the first inspection device 12.

[0074] <Second Inspection> The following describes the second inspection (inspection of time-series data on the current-carrying state of the conductor or insulator of the inspection object X) performed using the second inspection unit. The second inspection is performed before electrical components are mounted on the wiring substrate P.

[0075] The second inspection method inspects the wiring board P based on waveform data of the inspection object X obtained from the inspection object X by applying an inspection voltage or passing an inspection current through the inspection object X. The second inspection will be described when the first inspection for inspecting the electrical continuity of the conductor or insulator described above is performed. The inspection signal emitted from the inspection power supply / signal transmission unit of the electrical continuity inspection unit is input to the inspection power supply / signal reception unit of the electrical continuity inspection unit 101 and also input to the waveform inspection result acquisition unit of the waveform inspection unit. The inspection signal passing through the inspection object of the wiring board is input to the inspection power supply / signal reception unit of the electrical continuity inspection unit and also input to the waveform inspection result acquisition unit of the waveform inspection unit. The waveform inspection unit, taking into account the inspection conditions (inspection time, voltage, etc.) obtained from the electrical continuity inspection condition setting unit, normalizes (preprocesses) the input waveform data in the waveform diagnosis data generation unit, preparing it for input to the waveform inspection result determination unit.

[0076] In the inspection method for the second inspection, a plurality of inspection conditions C are used, and for each inspection condition C, waveform data D obtained from an inspection object X for data collection is collected in advance. A machine learning model 34 that has been trained in advance for each inspection condition C based on the collected waveform data D is used, and the waveform data D obtained from the inspection object X of the wiring substrate P to be inspected for pass / fail and the inspection condition C are input into the machine learning model 34, and the pass / fail of the inspection object X is determined from the output result.

[0077] The inspection conditions C are determined based on design information of the circuit pattern and wiring pattern of the wiring board P, and are stored in the recording device, for example, as part of an inspection program used when inspecting the inspection object X by the second inspection device 13. More specifically, one inspection condition C from among the plurality of inspection conditions C is assigned to each of the inspection objects X among the plurality of inspection objects X, and is stored in the recording device.

[0078] The inspection method of the second inspection is characterized in that one inspection condition C from a plurality of inspection conditions C is assigned to each inspection object X for data collection, waveform data D obtained from the inspection object X is collected in advance for each inspection condition C, and a machine learning model 34 that has been trained in advance for each inspection condition C is constructed. In this case, the measurement unit 31 described later inspects the inspection object X using the inspection program, so that the waveform data D of the inspection object X can be collected in advance. The collected waveform data D of each inspection object X is linked to the inspection condition C used for the inspection by the inspection program.

[0079] As described above, the inspection conditions C used in the inspection program of the second inspection device 13 are used as the inspection conditions C when generating the machine learning model 34 that has been trained in advance for each inspection condition C, so there is no need for the operator to separately specify the inspection conditions C when generating the machine learning model 34. Therefore, when generating the machine learning model 34, human error when specifying the inspection conditions C can be avoided.

[0080] In the inspection method of the second inspection, the measurement unit 31 uses the inspection program to inspect the inspection object X of the wiring substrate P to be inspected for pass / fail and obtains waveform data D (this waveform data D is also linked to the inspection condition C in the same manner as described above). Thereafter, the waveform data D obtained from the inspection object X of the wiring substrate P to be inspected for pass / fail and the inspection condition C are input to the machine learning model 34, which has been trained in advance for each inspection condition C based on the waveform data D collected in advance, and the pass / fail of the inspection object X is determined from the output result.

[0081] The inspection method for the second inspection will be described in more detail below in association with the description of each component of the second inspection device 13.

[0082] (Measurement Unit) The measurement unit 31 measures electrical characteristics obtained from the test object X on the wiring substrate P by applying a test voltage or passing a test current. The measurement unit 31 can be controlled by, for example, a program (more specifically, the test program). The waveform data D obtained by measurement by the measurement unit 31 and the test conditions C linked to it are stored in a recording device.

[0083] Specifically, the measurement unit 31 has a mechanism for bringing a predetermined test probe into contact with the test object X, and is configured to apply a test voltage (current) through the test probe to measure required electrical characteristics. Fig. 8 shows a case in which the measurement unit 31 applies a test voltage to the test object X of the wiring board P to obtain time-series (time axis t) waveform data D (D1, D2) of the current I to be measured. Note that the measurement unit 31 may also be configured to pass a test current through the test object X of the wiring board P to obtain the voltage waveform data D.

[0084] When the input signals of the test voltage or test current are waveform signals such as sine waves, rectangular waves, or trapezoidal waves, a test can be performed using a network analyzer to test the input characteristics of the frequency signals, and it is also possible to perform a Fourier transform of the input signals and measure the TDR.

[0085] The object to be measured may be a voltage or a current, or may be other electrical quantities such as capacitance or inductance. Furthermore, the waveform data D of the test object X may be the transient response (time-series waveform data) of the object to be measured, or may be frequency characteristic data. Time-series data is preferable as the waveform data D. Furthermore, from the viewpoint of easily acquiring the transient response, the object to be measured for the transient response is preferably a voltage or a current.

[0086] The measurement results are stored in a recording device as waveform data D of the inspection object X in association with the inspection conditions C, or are transferred directly to the acquisition unit 32 .

[0087] Here, the inspection object X is, for example, a wiring pattern (inspection circuit pattern) provided on a wiring substrate P. On the wiring substrate P, a large number of various inspection objects X are arranged.

[0088] The inspection conditions C include at least conditions (parameters) that may affect the measurement results and their arguments (parameter values). Examples of the conditions (parameters) include the shape of the inspection circuit pattern on the wiring board P to which the inspection voltage or inspection current is applied or passed during measurement; the combination of the inspection circuit patterns in the case of an insulation inspection between multiple inspection circuit patterns; the waveform of the voltage or current to be inspected (shape and absolute value of the voltage or current); the measurement range of the waveform data D; and some or all of the ambient temperature and humidity. Other conditions (parameters) preferably include some or all of the customer name, wiring board number, piece number within the wiring board, wiring pattern number provided within the piece, and terminal number of the wiring pattern. The inspection method for the second inspection uses multiple inspection conditions C, and at least one of the above conditions (parameters) or their arguments (parameter values) differs between the multiple inspection conditions C.

[0089] In particular, it is preferable that the inspection conditions C include any one of the conditions (parameters) and their arguments (parameter values) of the shape of the inspection circuit pattern of the wiring board P, the waveform of the voltage or current to be inspected, and the measurement range of the waveform data D, and it is more preferable that the inspection conditions C include all of these. By including some or all of these in the inspection conditions C, the inspection accuracy of the inspection method of the second inspection can be improved.

[0090] It is preferable to create an inspection program to be used when inspecting the inspection object X by the second inspection device 13, including all of the above-mentioned conditions (parameters) and their arguments (parameter values). It is also preferable to associate all of the above-mentioned conditions (parameters) and their arguments (parameter values) with the waveform data D of the inspection object X, and store the data in a recording device. This allows each piece of waveform data D to be associated with each inspection condition C.

[0091] (Acquisition Unit) The acquisition unit 32 acquires waveform data D obtained from the inspection object X of the wiring board P to be inspected for pass / fail, and the inspection conditions C. The acquisition unit 32 can be controlled by, for example, a program.

[0092] Specifically, the acquisition unit 32 receives the waveform data D of the inspection object X from the measurement unit 31 in association with the inspection condition C.

[0093] (Determining Unit) The determining unit 33 inputs the waveform data D acquired by the acquiring unit 32 and the inspection conditions C to the machine learning model 34, and determines whether the inspection object X is good or bad based on the output result.

[0094] In the inspection method of the second inspection, as shown in FIG. 9, an autoencoder 34a is used as the machine learning model 34 used to determine whether the inspection object X is good or bad.

[0095] The autoencoder 34a is one of the well-known machine learning techniques that uses a neural network.

[0096] (Machine Learning Model) In the second inspection device 13, the machine learning model 34 is composed of an autoencoder 34a and a weight W. That is, as the machine learning model 34, waveform data D obtained from an inspection target X for data collection is collected in advance for each inspection condition C, and the machine learning model 34 is trained in advance for each inspection condition C based on the collected waveform data D.

[0097] Explaining the learning of the machine learning model 34 in detail with reference to FIG. 10 , a plurality of waveform data D obtained from a test object X for data collection is collected in advance under certain test conditions C. At this time, it is preferable that the waveform data D collected in advance be waveform data judged to be normal. When using an autoencoder 34a, if the autoencoder is trained to accurately reproduce waveform data D judged to be normal, when an abnormal waveform is input, the waveform cannot be accurately reproduced. Therefore, by training only on waveform data D judged to be normal, it becomes easier to distinguish from abnormal waveform data D, and the test accuracy of the test method of the second test can be improved.

[0098] In reality, the waveform data D obtained from the test object X for data collection may include abnormal waveform data in addition to normal waveform data. For this reason, it is necessary to extract only normal waveform data as the waveform data D used for training the machine learning model 34, that is, it is necessary to determine whether the waveform data D is normal. To determine whether the waveform data D is normal, for example, a method of clustering multiple waveform data D and determining the majority as normal data, or a method of visually determining whether each piece is normal and classifying it can be used.

[0099] The processing in the autoencoder 34a is performed so that the waveform data D matches the reconstructed waveform data E that is reconstructed from the waveform data D through the autoencoder 34a. Therefore, no training data is required in the processing of the autoencoder 34a.

[0100] The weight W after learning of the machine learning model 34 is output from the autoencoder 34a and stored in association with the inspection condition C. Here, the "weight" corresponds to the "weight" in deep learning, and is an important parameter for extracting features. The weight W can be extracted and stored by a program, and can be stored in, for example, a recording device provided in the inspection system.

[0101] The necessary weight W differs for each inspection condition C. Therefore, the machine learning model 34 is trained for each inspection condition C. Therefore, in the inspection method for the second inspection, the weight W is trained for each inspection condition C.

[0102] (Determining Unit) The operation of the determining unit 33 will be described with reference to FIG.

[0103] First, the weight W is switched according to the inspection conditions C of the wiring substrate P to be inspected for pass / fail. Specifically, the weight W corresponding to the inspection conditions C of the waveform data D acquired by the acquisition unit 32 is selected and applied as the weight W of the autoencoder 34 a.

[0104] Next, the waveform data D and its inspection conditions C are input to the autoencoder 34a, which outputs reconstructed waveform data E. If the input waveform data D is normal waveform data D1, the reconstructed waveform data E1 will be close to the normal waveform data D1. On the other hand, if the input waveform data D is abnormal waveform data D2, the autoencoder 34a attempts to approximate the normal reconstructed waveform data E, and the reconstructed waveform data E2 will differ from the input abnormal waveform data D2.

[0105] The inspection method of the second inspection utilizes this property to determine pass / fail. That is, the error between the waveform data D input to the autoencoder 34a and the reconstructed waveform data E output is used. This error is small when the input waveform data D is normal waveform data D1 and large when the input waveform data D is abnormal waveform data D2. Therefore, if the error exceeds a predetermined threshold, it can be determined to be defective. The inspection method of the second inspection can detect even an unknown defect if there is a difference from the normal waveform data D, and generally, if there is a defect, there is often a difference from the normal waveform data D. Therefore, the inspection method of the second inspection can dramatically reduce the possibility of overlooking a defect compared to conventional inspection methods that can only detect known defect phenomena.

[0106] The error can be calculated using a known method such as the least squares method.

[0107] If the input waveform data D is determined to be normal, additional learning may be performed on the machine learning model 34 using that waveform data D. The additional learning may be performed sequentially, or may be performed all at once after a certain number of waveform data D have been accumulated. Furthermore, waveform data D determined to be abnormal may be accumulated and used for analyzing failure modes.

[0108] The determination made by the determination unit 33 can be executed by a program. That is, the program is a program for causing a computer to execute an inspection to inspect a wiring substrate P based on waveform data D of the inspection object X obtained from the inspection object X by applying an inspection voltage or passing an inspection current, a plurality of inspection conditions C are used for the inspection of the wiring substrate P, waveform data D obtained from the inspection object X for data collection is collected in advance for each inspection condition C, and a machine learning model 34 that has been trained in advance for each inspection condition C based on the collected waveform data D is used, the waveform data D obtained from the inspection object X of the wiring substrate P to be inspected for pass / fail and the inspection condition C are input to the machine learning model 34, and the pass / fail of the inspection object X is determined from the output result.

[0109] A signal representing the result of the determination made by the determination unit 33 is output to the display unit 35. The signal may be configured to be output to a device other than the display unit 35.

[0110] (Display Unit) The display unit 35 displays the pass / fail result of the inspection object X judged by the judging unit 33. The display unit 35 can be controlled by a program, for example.

[0111] The display unit 35 may be, for example, a known display or an LED (Light Emitting Diode). When a display is used, it is possible to simultaneously display a large amount of information, such as the judgment result, waveform information, and the error between the waveform data D input to the autoencoder 34a and the reconstructed waveform data E output. On the other hand, when an LED is used, pass / fail can be easily indicated by, for example, turning on / off or by changing the color of the light, which makes it easy to use for sorting pass / fail products in product manufacturing.

[0112] The test results are stored in a recording device.

[0113] <Clock Generator> The clock generator generates a timestamp (time information) when data is stored in the recording device.

[0114] A known clock can be used as the clock generator. A radio-controlled clock is preferable for accurate date and time. The date and time indicated by the clock generator is sent to the recording device as digital information.

[0115] <Recording Device> The recording device records the results of the first inspection unit and the results of the second inspection unit. That is, in this inspection method, the results of the first inspection and the results of the second inspection are recorded in the recording device.

[0116] As explained in the first inspection section, during the first and second inspections, multiple wiring boards P are arranged on a single wiring board sheet S. The wiring board sheet S is cut after electrical components are mounted and reflowed. A sheet control number is assigned to the wiring board sheet S, and a wiring board serial number is assigned near each wiring board P, and they are distinguished by an identification system with a high information density per area, such as a two-dimensional barcode. A continuity test, an insulation test, and a waveform test (time-series data test) are performed for each wiring board serial number, and a test number is assigned for each test.

[0117] The data is written to the recording device as soon as the continuity test, insulation test, and waveform test (time series data test) are completed. For example, when the first test is performed, the test results are recorded as shown in Table 2, along with the sheet control number, wiring board serial number, and test target number, and the results of the second test, the waveform test. Furthermore, time information obtained by a clock generator is also recorded with these test results.

[0118]

[0119] The inspection device 10 can store the results of the first and second inspections, as well as production information aligned with the inspection results on a time axis. The production information includes the number of inspections, the number of contacts, the component usage time, contact voltage, the moving average value of contact voltage, the probe load, the probe stroke amount, the probe contact impedance, the measurement cycle, the measurement suspension period, image information of the wiring board P, the infrared image of the wiring board P, the calibration time, and the operating status of other inspection devices 11 present on the same premises. In addition to the equipment described above for measuring and storing these values, the inspection device 10 also includes a counter for measuring the number of inspections, a load meter for measuring the load, a sensor for detecting when measurement has been suspended, a network connector for connecting to a process control system, and the like.

[0120] The inspection device 10 can store the results of the first and second inspections, as well as environmental information, time-aligned with the inspection results. Environmental information refers to external energy that the inspection device receives, such as temperature, humidity, external air temperature, climate, air pressure, radio wave conditions, surveillance image information, noise, and vibration, which may affect the inspection. To measure environmental information, the inspection device or factory is equipped with sensors such as a thermometer / hygrometer, radio wave measuring device, surveillance camera, microphone, and acceleration sensor, as well as an interface for inputting external information such as weather and air pressure.

[0121] The inspection device 10 can store management information along with the results of the first and second inspections. This information does not need to be matched with the inspection time information, such as the inspection device number, the parts replacement history of the inspection device, the name of the factory where the inspection was performed, the materials used (adhesive, insulator, conductor, protective sheet, resist used), the name of the factory where the wiring board P was produced, the transportation route, transportation means, number of days for transportation, and manufacturing date. The inspection device 10 is equipped with input means, display means, input / output means for transferring data, a network connector for connecting to a process management system, etc., and acquires and stores this information.

[0122] When designing a wiring board P, each layer, including the silk layer, resist layer, wiring pattern layer, drill, and solder surface, is designed using circuit design CAD, and the designed wiring is saved as data that can reproduce the design. The design data is handled in a general-purpose format, such as Gerber or extended Gerber. When determining the inspection conditions for the board inspection system, the Gerber data is used to develop a design drawing, and the wiring length, width, clearance with adjacent wiring, bend angle, etc. are measured while referring to the drawing, and inspection conditions such as the inspection method, the inspection point where the probe contacts, and the inspection voltage and time are determined while taking these into consideration. In this process, a wiring number is defined for each wiring, and design information such as the wiring length, wiring width, bend angle, clearance with adjacent wiring, adjacent wiring number, characteristics, classification, distance between wiring, through-hole information, resist area information, materials used, manufacturing plant, transportation distance, manufacturing period, and parasitic capacitance are determined for each wiring, and saved together with the results of the first and second inspections.

[0123] Design information can be acquired manually by the user by unpacking Gerber data, or online or offline from the manufacturing plant at the time of product shipment. Alternatively, information can be acquired using software that defines board inspection conditions. The software reads Gerber data, which is a list of data, and can interpret insulators and conductors, measure wiring length, wiring width, and clearance from adjacent wiring, calculate wiring area for each wiring, automatically assign wiring numbers, and set vias. It can also provide recommended positions for inspection points where pogo pins or wire probes will contact, recommended inspection voltages, recommended inspection times, and inspection sequences for each wiring number. Furthermore, when multiple conductors are close to each other on a printed circuit board or wiring cable, minute capacitance components are generated between each conductor and the electrode. This can lead to short circuits or malfunctions due to the addition of foreign matter or peeling. Preliminary parasitic capacitance calculations and values ​​measured through continuity tests can be recorded as design information. Similarly, parasitic inductance and parasitic resistance can also be recorded.

[0124] The acquired design information is recorded in association with the corresponding inspection management number and the inspected wiring number when the inspection conditions are set or after the inspection. Not only information related to the wiring number of the inspection target X, but also information on adjacent wiring can be linked and recorded. Information on wiring can be stored simultaneously with data and classification that represent the characteristics of the wiring pattern based on Gerber.

[0125] The above-mentioned inspection results, production information, environmental information, design information, etc. can be stored in a recording device, which may be a hard disk drive or memory provided in the inspection device 10, a recording device on a production management system connected to the inspection device 10 via an in-house network, a recording device on the cloud, or a recording device provided in another inspection device 11 described later.

[0126] <Third Inspection Unit> The third inspection is performed after the first and second inspections, and inspects the wiring substrate P with the electrical components mounted thereon. For the third inspection, for example, another inspection device 11 shown in FIG. 4 can be used.

[0127] The other inspection device 11 includes a solder printing device 121 and a post-solder printing inspection device 122. The solder printing device 121 prints solder paste on electrode portions on a wiring board P by screen printing. The post-solder printing inspection device 122 measures the printed solder paste on the wiring board P that has been processed in the process of the solder printing device 121 two-dimensionally or three-dimensionally, for example, using images, and judges the quality of the printing from the results. This inspection of the quality of the printing corresponds to an example of the third inspection.

[0128] The other inspection device 11 includes a mounter 131 and a post-mount inspection device 132. The mounter 131 picks up electrical components to be mounted on the wiring board P and places the electrical components on the solder paste at the mounting location. The post-mount inspection device 132 measures the electrical components on the wiring board P processed in the process by the mounter 131 in two or three dimensions, for example, using images, and determines whether the electrical components are properly mounted based on the results. This inspection of the quality of the mount corresponds to an example of the third inspection.

[0129] The other inspection device 11 includes a reflow furnace 141 and a post-reflow inspection device 142. The reflow furnace 141 heats and melts the solder paste, then cools it, and solders and joins the electrical components onto the wiring board P. The post-reflow inspection device 142 measures the solder portion of the wiring board P after reflow, which has been processed in the reflow furnace 141, two-dimensionally or three-dimensionally, for example, using images, and judges whether the reflow is good or bad from the results. This inspection of the quality of the reflow corresponds to an example of the third inspection.

[0130] The other inspection device 11 includes a finished product inspection device 152. The finished product inspection device 152 inspects the wiring board P after it has been finally assembled and has electrical components mounted thereon. In this finished product inspection, the appearance and electrical characteristics of the wiring board P are evaluated to determine whether the wiring board P is good or bad. This inspection of the quality of the wiring board P corresponds to an example of the third inspection.

[0131] The other inspection device 11 includes an inspection management terminal 162. The inspection management terminal 162 is used for tasks such as creating and distributing inspection programs that define the content of inspections performed by each inspection device, and registering and configuring various information. The inspection management terminal 162 is implemented, for example, by a general-purpose computer. The inspection management terminal 162 may be configured as a single computer, or as shown in FIG. 4, as multiple computers. Alternatively, some or all of the inspection management functions may be implemented in the computer possessed by each inspection device. First, the inspection results performed for each inspection are recorded in a recording device possessed by each inspection device. The inspection management terminal may exist for each factory, each process, or on the cloud. The inspection management terminal and each inspection device are connected via a physical interface or a network including wireless communication, and information recorded in each inspection device is shared through inspection result relationships. The storage device that records the first and second inspection results described above is connected to the inspection management device that manages the inspection results of the third inspection and each inspection device via a physical interface or a network including wireless communication. Examples of information that can be related include board management numbers, wiring numbers, and inspection management numbers. A relational database may be created in which the information that each inspection device has that can be related is compiled into a data sheet. The inspection management terminal can use the relational database to obtain necessary information from each inspection management terminal and inspection device.

[0132] The results of the third inspection are compiled by the inspection management device 162, and the recording device records the results of the third inspection unit.

[0133] The results of the third inspection, for example, for board inspection and mounted board inspection, include the serial numbers of the boards and components, inspection conditions, inspection results and inspection time, and inspection image data from the image inspection, as shown in Table 3. The third inspection results may also include the results of the continuity inspection and waveform inspection, which are inspection results from the previous process. Information on the first inspection results and second inspection results sent from the inspection management system is selectively recorded.

[0134]

[0135] At this time, the results of the third inspection are recorded in association with the results of the first inspection and the results of the second inspection. The results of the first inspection and the results of the second inspection can be associated with each other, for example, by using a two-dimensional barcode attached to the wiring board P and including identification information such as a sheet control number and a wiring board serial number. The inspection management system that manages the results of the third inspection prepares items (columns) for saving the results of the first inspection and the second inspection, and can also add the results of the first inspection and the second inspection after the fact as necessary.

[0136] For example, the columns "Post-Process Inspection Results" and "Post-Process Inspection No." shown in Table 2 can be used to add the results of a third inspection and link the results of the first inspection, the second inspection, and the third inspection. This allows the results of the post-process inspection performed by another inspection device 11 to be directly referenced from the records of the results of the first inspection and the second inspection performed by the inspection device 10, and also allows more detailed post-process inspection results to be known by referring to the results of the third inspection recorded in the other inspection device 11 using the post-process inspection No.

[0137] The post-reflow inspection machine classifies defects found during inspection and inputs the defect classification along with the inspection number, board number, production information, and environmental information. For example, in the case of post-reflow image inspection, the inspector visually inspects the images and operates the inspection device's GUI screen to register the defect information. Evaluation and classification can also be performed using a classifier built into the inspection device and trained using training data, or a combination of a classifier and visual inspection can be used. Inspectors can link the post-reflow image inspection results with the inspection results of the previous process. Figure 12 shows a GUI screen on which inspectors select and save the inspection result classification while reviewing the image inspection results. The image inspection device can send the inspection results to a database recording the inspections of the previous process. The inspector selects the inspection results of the previous process displayed on the GUI and presses the send button, linking the inspection results of the previous process with the inspection results of the previous process. The inspection results of the previous process can also be sent to the inspection device manufacturer.

[0138] Figures 13 and 14 show examples of defects discovered during post-process inspections. Figure 13 shows a migration defect in which copper ions have precipitated along the glass fibers between the via and conductor, while Figure 14 shows delamination caused by insufficient adhesion between the conductor and substrate, and by water vapor generated within the substrate due to the heat of component mounting. Other examples include plating breaks due to thermal expansion and contraction, and breaks due to abnormal via formation. In cases where moisture that has entered small gaps within the substrate is heated and vaporized by the mounting heat of processes such as reflow, causing the gaps to grow larger, it is difficult to discover the defect using a board continuity test in a pre-process inspection.

[0139] <Defect Analysis> As shown in Figure 11, defects observed on the wiring board P include glass cloth peeling PD1 in the insulating region PS1, disconnection PD2, peeling PD3, and corrosion PD4 in the wiring PS2, and peeling PD5, corrosion PD6, and plating defects PD7 in the through-hole PS3. For example, if a connection defect is discovered during a post-reflow continuity test after electronic component mounting, a detailed analysis is performed to examine the board cross section. The analysis involves identifying the defective wiring number and the location of the analysis point based on the design information, followed by preprocessing. The preprocessing involves immersing the test sample in epoxy resin, removing air bubbles using a vacuum pump or similar device, and allowing the resin to penetrate. The resin is then hardened and cut using a precision cutting device to include the identified observation point. Figure 14 shows delamination occurring on the cut cross section, and a more detailed cause of the defect is identified through visual inspection using an optical microscope. As described above, the defective conditions detected by the post-process inspection equipment and the detailed information obtained by the operator analyzing the samples classified as defective using an analytical device are recorded in the recording device of the inspection equipment and the recording device of the manufacturing management system. The record includes the board management number, the inspection management number, the detected defective conditions, and the detailed information.

[0140] These defect status information may be classified using an interface provided in the other inspection device 11 performing the third inspection, as shown in FIG. 12 . For example, an inspection management device 162 of the other inspection device 11 may be used as an interface to select the defect type from a pull-down menu 162b in an inspection image 162a of the defect location. The defect status information classified by the interface may then be used to rewrite the results of the second inspection. The input defect status and detailed information of the wiring board may be managed in a recording device provided in the inspection device, a recording device in the manufacturing management system, or a recording device in the cloud. Linking to the inspection results of the previous process is also possible. For example, when an operator in charge of detailed analysis of a connection defect inputs the inspection results into the manufacturing management system, the operator inputs the wiring board management number and wiring number to be analyzed. The manufacturing management system can then identify the inspection results of the previous process (first inspection result, second inspection result) from the input wiring board management number and wiring number and display the inspection results to the operator. The manufacturing control system can also notify the system that records the inspection results of the previous process (first inspection result, second inspection result) and update the inspection results of the previous process with the inspection results of the next process. The notification and update can be made to the manager who manages the inspection results of the previous process, who can then update the results. Notification and update of the defect status and detailed information by the control system are also simultaneously made to the board inspection equipment manufacturer.

[0141] That is, the second inspection result is rewritten using the defect status information classified by an interface provided in the device that performs the third inspection.

[0142] The inspection device 10 can detect potential defects even in an object that has been judged to be good in a continuity test. Possible defects include peeling of glass cloth PD1 in the insulating region PS1, disconnection PD2 in the wiring PS2, peeling due to etching defects PD3, peeling due to insufficient desmearing PD5, and corrosion PD4 and PD6 due to etching residue. Examples of complex defects include peeling PD5 that has occurred in the Cu inner lining of the through-hole PS3 and poor through-hole connection due to corrosion PD6.

[0143] 13 and 14 show examples of the effect of defects on the waveform diagnosis results of the inspection system 10. FIG. 13 illustrates the case where copper ion deposition PD8 (migration) occurs between via PS4 and conductor PS5. The characteristic response waveform shown in FIG. 13 is obtained in the second inspection. FIG. 14 illustrates delamination PD9 of wiring PS2, which may grow and eventually lead to delamination. The characteristic response waveform shown in FIG. 13 is obtained in the second inspection. The pre-process inspection worker or inspection system receives input of the defect condition and detailed information that occurred in the post-process (third inspection) and reconfirms the first and second inspection results for the relevant board number and wiring number. Even if the reconfirmation reveals a good continuity test, there may be cases where a defect occurred in the post-process, resulting in characteristic features appearing in the second inspection results. Defects identified in the third inspection include growth defects such as ion deposition, interlayer delamination due to water vapor, and material deterioration due to environmental conditions.

[0144] <Operation of the Inspection System> FIG. 15 shows an example of operation of the inspection system.

[0145] The second inspection device is equipped with an autoencoder as a means of anomaly detection. As mentioned above, autoencoders are used for dimensionality reduction and feature extraction, which are one of the unsupervised machine learning techniques that use neural networks. However, they can also detect anomalies by comparing input and output data. During the autoencoder's learning process, the weights of the neural network are adjusted so that the input and output match. Through this learning process, a neural network is formed that extracts only the important information necessary for reconstruction from the data and then efficiently generates the original data.

[0146] For example, the inspection device 10 is delivered to a certain user, and the first inspection and the second inspection are started together with the prototype of the wiring substrate P. The results are accumulated, for example, in a server 41 owned by the user, and first and second inspection result databases 42 containing the inspection results and waveform data are constructed.

[0147] The inspection equipment manufacturer trains the neural network so that normal data obtained from the above-mentioned prototype results pass through. For example, the inspection equipment manufacturer collects waveforms determined to be normal from the first inspection results obtained from the above-mentioned prototype, and determines whether there is an abnormality by building a neural network through which normal waveforms pass.

[0148] If a defect is identified during the prototype stage and a characteristic feature is evident in the waveform diagnosis results, the waveform data excluding the waveform containing the defect can be redefined as normal data, and the autoencoder neural network can be retrained. Using this updated waveform diagnosis database 43, anomaly detection can be performed using, for example, the machine learning model 34 (see FIG. 8 ) used in the second inspection. By retraining the machine learning model 34 in this manner, potential defects that occur before mass production and are difficult to identify through continuity testing can be detected, thereby optimizing the machine learning model 34 for the user's product, the wiring board P. As shown in FIG. 15 , the inspection equipment manufacturer performs clustering 44 on the inspection results of the inspection equipment performed during the prototype stage. Even the normal waveforms of the first inspection results collected during the prototype stage contain diagnostic results that may indicate a defect. Waveforms that may indicate a defect can be identified by empirical rules based on past diagnostic results, determining the similarity of diagnostic results, and comparing and analyzing the shapes of diagnostic waveforms. For example, the analysis performed by clustering can involve inputting an inspection signal to the inspection target during a continuity test in a board inspection, and using the time-series data of measurement values ​​output from the inspection target board as the analysis target for waveform diagnosis. Alternatively, the measurement values ​​output from the inspection target board can be passed through an autoencoder provided in the waveform diagnosis database 43, and the waveform output from the autoencoder can be used as the analysis target for waveform diagnosis. Both of these methods are also possible. The waveform diagnosis database 43 is not simply a database, but a system equipped with a processor, a recording device, an interface, a display, input means, and software for updating the machine learning model. The machine learning model 34 used in the mass production stage is updated and shared using the waveform diagnosis database 43. The waveform diagnosis database 43 in the present invention has an inspection result database function for recording inspection results, an inspection result database update function for forming relationships with previous and subsequent processes and obtaining and updating results from other inspection devices, a machine learning model update function for updating the machine learning model 34 by referencing stored inspection results, including inspection results from other inspection devices (third inspection), so as to exclude defective data and pass normal data, and a waveform diagnosis function (second inspection) for diagnosing the waveform of the inspection target using the updated machine learning model 34.The waveform diagnosis database 43 used in the mass production stage is sufficient if it has at least an inspection result database function and a waveform diagnosis function, and updates its inspection function by obtaining an updated machine learning model from an inspection device having a machine learning model update function. In FIG. 15 , the inspection equipment manufacturer owns the waveform diagnosis database 43 having the machine learning model update function and the inspection result database update function, and updates the inspection database by obtaining inspection results from another inspection device (third inspection). However, the present invention is not limited to this configuration. The waveform diagnosis database 43 having the machine learning model update function and the inspection result database update function may be installed in the prototype inspection device 10 or mass production inspection device of a user (e.g., a circuit board manufacturer). Furthermore, the mass production inspection device used in the mass production stage may have, in addition to the waveform diagnosis function and the inspection result database function, an inspection result database update function that forms a relationship with the preceding and following processes. In this case, the third inspection results obtained by the mass production inspection device used in the mass production stage are notified to the waveform diagnosis database 43 having the machine learning model update function and the inspection result database update function of the inspection equipment manufacturer, and the inspection results are rewritten using the inspection result database update function owned by the inspection equipment manufacturer.

[0149] This retrained machine learning model 34 can be used in mass production.

[0150] When mass production begins at the user, the wiring board P is similarly inspected using the inspection device 10. After the first and second inspections have been performed by the inspection device 10, the wiring board P undergoes the mounting of electrical components and the like in a post-process 45 along with a third inspection, and is then assembled as a final product 46 before being shipped to a market 47.

[0151] Although a part may be judged as non-defective in the first and second inspections, it may be judged as defective by another inspection device 11 that performs the third inspection, as described above. If a defect is identified in a subsequent process 45, the defective state and detailed information are stored in a waveform diagnosis database 43, production data is traced 48, and defect analysis 49 is performed. If a characteristic is found in the waveform diagnosis results, the waveform data group from which the defective waveform data has been removed is redefined as normal data in the defective waveform data removal process 4A, and the autoencoder's neural network can be retrained. The retrained autoencoder is stored in the waveform diagnosis database 43, subjected to anomaly detection training 4B, and transmitted to inspection equipment deployed in mass production. In other words, if a waveform similar to the waveform judged defective in a subsequent inspection is input to an inspection device in the mass production stage using the updated and transmitted autoencoder, the input data and output data will no longer match, and by identifying the difference, it will be possible to identify similar defective states. In this way, the results of the second inspection recorded in the recording device can be rewritten using the defective condition information of the wiring substrate P determined in the third inspection, and the autoencoder's neural network (machine learning model) can be re-trained using time series data excluding the time series data containing the defective information determined in the third inspection.

[0152] The waveform diagnosis database 43 has a clustering function and can link and record design information with first and second inspection results. The waveform diagnosis database 43 acquires prototype information, including Gerber data, and can interpret insulators and conductors, measure wiring length, wiring width, and clearance from adjacent wiring, calculate wiring area for each wiring, automatically assign wiring numbers, and set vias. Furthermore, it can provide recommended positions for inspection points where contacts such as pogo pins or wire probes make contact, recommended inspection voltages, recommended inspection times, and an inspection sequence for each wiring number. While such clustering of inspection target data can be performed by an operator manually extracting Gerber data, which is a list of data, to create a wiring diagram, it can also be performed using software that defines board inspection conditions. Clustering of design information does not need to be performed simultaneously with inspection; it can be performed before inspection, when setting inspection conditions, or after inspection. The set information such as the wiring number is used in the inspection of the subsequent process, and any defect information that occurs in the inspection of the subsequent process is assigned the wiring number and the defect status and detailed information are notified and sent to the inspection DB of the previous process.

[0153] The defect analysis 49 performed by the inspection equipment manufacturer not only checks the waveform data, but also uses the management information saved together with the first and second inspection results, and production data obtained afterwards. Defects can occur due to changes in the materials used or the ambient temperature, and wiring boards with lot numbers and sheet numbers related to those materials or similar temperatures may be subject to reinspection.

[0154] For example, if the same user is also performing a third inspection, the defect condition information from the other inspection device 11 can be transferred to a server 41 owned by the user, and the results of the second inspection recorded in the first and second inspection result databases 42 can be overwritten. Alternatively, as shown in Fig. 4, a dealer 171 or a support center 172 of the other inspection device 11 can be connected to the other inspection device 11 via an internet line or the like, and defect information and detailed information of defects occurring in the final product can be notified and transmitted to the inspection device manufacturer and the board manufacturer via the dealer 171 or the support center 172, and the third inspection result can be added to the first inspection result and the second inspection result.

[0155] Furthermore, there may be cases where the product is not judged to be defective by other inspection equipment 11 and is discovered as defective in the market 47. Even if such a defect occurs, if the inspection method in question is used, the results of the third inspection can be traced.

[0156] In either case of failure, the results of the second inspection or the first inspection can be traced and a failure analysis 49 can be performed.

[0157] In the second inspection, the machine learning model 34 can be retrained using time series data excluding time series data containing the defect information determined in the third inspection. Specifically, the waveform diagnosis database 43 is reconstructed excluding time series data containing the defect information, and the machine learning model 34 is retrained according to this waveform diagnosis database 43. As described above, an inspection object X that is determined to be a good product in the second inspection but is determined to be a defect in the third inspection should preferably have been determined to be a defect in the second inspection. Furthermore, while the inspection object X described above is determined to be a good product when the current machine learning model 34 is used, the time series data often has different properties from the time series data obtained from an inspection object X that is actually determined to be a good product. Therefore, by retraining the machine learning model 34 using time series data excluding such time series data containing the defect information, it may be possible to determine the object as a defect.

[0158] Furthermore, the machine learning model used in the second inspection can be updated based on the results of the third inspection, and the inspection conditions of the inspection device 10 can be optimized by referring to the results of the third inspection and production information. Some defects may be more likely to occur under specific conditions, such as when inspected at a specific measurement interval. In such cases, the inspection device 10 stores the results of the first and second inspections, as well as information on the measurement interval, aligned with the inspection results on a time axis. This allows for defect analysis to identify the specific conditions that make the defect more likely to occur. If such conditions exist, the occurrence of defects can be prevented by, for example, setting inspection conditions that make defects less likely to occur.

[0159] The updated machine learning model 34 and inspection conditions need to be reflected in the mass-production inspection system 10 at the appropriate timing. For example, the updated machine learning model 34 and inspection conditions can be accepted through a user interface of the inspection system 10, such as that shown in FIG. 16 . Using such a user interface can avoid the inconvenience of unintentionally updating the machine learning model 34 and inspection conditions. FIG. 16 shows a screen of an inspection management system that manages a third inspection system or multiple third inspection systems. The inspection system manufacturer can recommend changes to the inspection conditions for the first inspection based on not only the updated machine learning model but also the identified defect information. This screen recommends changing the voltage application time for inspection No. 1 from 0 seconds to 0 seconds. For example, in order to accurately perform waveform diagnosis (the second inspection) when an unintended defect occurs in a downstream process (the third inspection), it may be desirable to change the inspection conditions or inspection points of the first inspection. In such a case, the waveform diagnosis database 43 lists not only the second inspection conditions but also the conditions to be changed for the first inspection results, and can propose the inspection conditions to be changed to the user as recommended inspection conditions.

[0160] That is, an inspection method for inspecting a wiring board by applying an inspection voltage or passing an inspection current through an object to be inspected, the object to be inspected being a wiring board made of insulators and conductors and carrying electrical components, using a plurality of inspections, the inspections including a first inspection having first inspection conditions for inspecting the electrical conduction state of the conductors or insulators and a second inspection having second inspection conditions for inspecting time-series data of the electrical conduction state of the conductors or insulators, the results of the first inspection and the second inspection being recorded in a storage device, the wiring board with the electrical components mounted thereon is processed in a subsequent process, a third inspection is performed, and part of the first inspection conditions or the second inspection conditions is changed based on information on the defective state of the wiring board determined in the third inspection. The first inspection conditions can be predetermined, including inspection time, voltage, current, and frequency, and these inspection conditions are changed depending on the results of the third inspection.

[0161] When the machine learning model 34 and the inspection conditions are updated, the defect detection capability can be improved and the occurrence of defects can be prevented for subsequent production lots.

[0162] On the other hand, for lots produced before the machine learning model 34 or the inspection conditions were updated, it is possible to extract inspected objects that should be judged as defective and final products that incorporate the substrates that are judged as non-defective by referencing the inspection results from the first and second inspection result databases 42. Therefore, it is also possible to perform a recall if necessary.

[0163] As shown in Figure 17, the inspection equipment manufacturer can also manage the inspection results of multiple production factories. Figure 17 shows an example of operation across three production factories A, B, and C, but the number of production factories that are operated across the factory is not limited to three.

[0164] In the example shown in Figure 17, inspection results from production factories A, B, and C of different board manufacturers are aggregated in a cloud-based fault diagnosis database (Fault Diagnosis DB in Figure 17) operated by an inspection equipment manufacturer. Note that the fault diagnosis database stores inspection result data such as images and waveform data, which can easily become enormous in volume. From this perspective, storing the data in the cloud is preferable. In this case, defect information linked to the design information of products and wiring patterns from different manufacturers can be aggregated. By clustering wiring patterns, information on wiring patterns prone to defects can be collected and analyzed across production factories, manufacturers, and products. Inspection equipment manufacturers can improve inspection accuracy by introducing machine learning models updated with the collected defect information into the mass production stage. Furthermore, the cloud-based fault diagnosis database can collect defect information and receive third-party inspection results from each manufacturer, providing a service that links this information. This allows board manufacturers to outsource inspection result management.

[0165] When operating across multiple production factories, defects that occur can be classified into those specific to the production factory and those common to all production factories. Defects that are common to all production factories can be used to update the machine learning models 34 of other production factories, and can also be incorporated into the construction of an initial model (a model before re-learning) of the machine learning model 34 in the prototype stage. This can improve the accuracy of the machine learning model 34.

[0166] <Inspection report> The inspection system stores the linked first inspection results, second inspection results, and third inspection results in a recording device, so it can report the range of inspection results obtained by inspecting multiple wiring boards P, the inspection results of the relevant wiring boards P within the range, deviation values, etc.

[0167] It is also possible to automatically collect and compile various information necessary to investigate the cause of defects in the wiring board P and present the results. For example, FIG. 18 shows an example of a waveform diagnosis report 52 for the second inspection created from a corresponding defect report 51 for the image inspection in the third inspection. The defect report 51 is an image inspection result report that includes a defective board image 51a generated from the third inspection results. The defect report 51 includes display items 51b related to the program name, drawing information, manufacturing date and time, inspection date and time, board management number, wiring number, various related inspection results 51c, area information on the defective part on the wiring board, and an image 51d surrounding the defective part, all of which are recorded in the device. The inspection management system can create the report immediately after the defect occurs. An operator can also add comments in the comment field 51e. The waveform diagnosis report 52 is a continuity and waveform inspection result report automatically generated from the first and second inspection results based on the third inspection results. The system automatically generates information on the defective board control number 52a, the defect name 52b obtained from the third inspection results, the inspection name 52c (e.g., continuity inspection, insulation inspection, spark inspection, high-frequency diagnostic inspection), the target wiring number 52d, the waveform information control number 52e, the graphed waveform range 52f, the inspection conditions 52g, the inspection date and time 52h, and the production information 52i. The left diagram of the waveform range 52f shows the upper and lower limits of the measured waveform with dotted lines, and the waveform range of the wiring board P where the defect occurred with a solid line. The right diagram shows the standard deviation, indicating the deviation value of the wiring board within the upper and lower limits of the measured waveform. In this example, the inspection results are within the upper and lower limits, and the deviation value exceeds the standard. By simultaneously outputting the inspection results of the upstream process linked to the inspection results where a defect occurred in a downstream process, it is possible to accurately determine which process the defect occurred in. The example in waveform diagnosis report 52 is a case where there is no abnormality, but there may be cases where a defect is found in the waveform diagnosis. Even in this case, a waveform diagnosis report can be submitted at the same time as the defect is detected in the third inspection.

[0168] As shown in FIG. 19 , these results can be linked to a circuit design CAD system to visually display statistical information, such as the frequency of occurrence of defective locations. This visual display is useful for analyzing the cause of defects. While FIG. 19 shows statistical information using shades of color, it may also be displayed using color coding or numbers. The statistical results can be visually displayed, such as high probability of occurrence for region R1 of wiring Nos. X3 and X4, medium probability of occurrence for region R2 of wiring No. X3, and low probability of occurrence for region R3 of wiring No. X3. This figure is linked to the first to third inspection results shown in Tables 1, 2, and 3. The figure displays the wiring number, and clicking on the displayed wiring number or each wiring displays the related inspection results. Collected information, such as design information, production information, and environmental information, is also displayed in association with the wiring number.

[0169] That is, a display device that displays inspection results of a wiring board that is made of insulators and conductors and has electrical components mounted thereon includes a calculation device, a recording device, and a display unit, the wiring is a wiring board that is made of insulators and conductors and has electrical components mounted thereon, the inspection of the wiring board includes a memory that stores multiple inspection programs, the inspection programs include a first inspection that inspects the electrical state of the conductors or insulators and a second inspection that inspects time-series data of the electrical state of the conductors or insulators, the results of the first inspection and the results of the second inspection are recorded in a storage device, the recording device records the first inspection results and the second inspection results according to the wiring pattern and insulating areas of the conductors and records design information related to the wiring pattern and the insulating areas, and the display unit displays the inspection results linked to the design information by distinguishing them by different colors, shades, or numerical values ​​according to the wiring pattern and insulating areas of the conductors.

[0170] Furthermore, it is possible to simultaneously display multiple wiring boards P and check the individual differences specific to the wiring boards P, and it is also possible to discover singular points in statistical information by utilizing so-called AI.

[0171] <Data Clustering> This inspection system can cross-sectionally utilize the inspection results of wiring boards P based on the same design, as well as the inspection results of wiring boards P with different designs. Fig. 20 shows an example of clustering targets for defect analysis using inspection results of three different designs A, B, and C.

[0172] The waveform diagnosis DB of each inspection device stores, for example, design data (Gerber data) according to the inspection target, response waveform data of the second inspection, inspection information, environmental information, the results of the first and second inspections, as well as any defective conditions found in inspections in later processes, detailed information, etc. As described above, the design data includes the wiring shape (wiring number) of the inspection target, Gerber data of other adjacent wirings, clearance, etc.

[0173] In clustering, inspection targets X with similar design data and inspection conditions, including clearances, are extracted from these design databases and generated as a group of data (clustered data). Clustering can be performed by a worker using data extracted from software that defines inspection conditions to classify similar wiring patterns, or by machine learning. Both methods are also possible. For example, in machine learning classification, a learning model generation unit classifies multiple data sets of training data into training data sets and evaluation data sets. The learning model generation unit trains a model using deep learning with the training data sets of training data. The learning model can generate both design information similarity judgments and waveform data similarity judgments. Information stored simultaneously, such as environmental information and production information, can also be used as evaluation factors for feature extraction. The classification results generated by clustering are added to the inspection results in the waveform data database. A classification result is assigned to each wiring number. The classification results can be classified using both character classification and numerical values ​​representing feature quantities. In the case of classification by numerical values, similarities derived from analytical elements such as wiring length, wiring width, and angle can be reflected and distributed on an XY plane. X and Y values ​​are assigned to each wiring number, and wiring numbers with high similarities are placed in close locations. Further analysis and classification can also be performed, including fault conditions and detailed information. For example, a risk level can be assigned to each wiring number using a number from 1 to 100.

[0174] For clustering, empirically proven patterns that frequently cause defects can be used as training data. For example, in the conductive pattern of the P3 layer in Figure 1, there is a long, inverted L-shaped wiring pattern L that spans two consecutive inner edges and is connected to five through holes. This type of wiring is susceptible to warping due to stress changes at the edge of the board. Flags and feature names can be assigned to such wiring patterns, and the wiring features (CAD data, wiring length, position, through-hole information, etc.) can be used as training data.

[0175] In addition to the method of using extracted data, the classification used for clustering can also be performed by developing CAD data stored for each wiring number, visually recognizing the similarity of the wiring shapes, and using an image recognition program that uses machine learning to determine the similarity. As described above, the design information stored in the waveform diagnosis DB is stored to the extent that the wiring shape can be reproduced in two or three dimensions for each wiring number. Adjacent wiring is also stored, making it possible to determine the similarity between the target wiring and the adjacent wiring.

[0176] 20 shows waveform data, design data (Gerber data), inspection information, and environmental information for different wiring board designs A, B, and C. By performing the above-described clustering for each of designs A, B, and C, it becomes possible to link inspection results with similar wiring patterns, and comparison of waveform data, inspection information (first, second, and third inspection results), and environmental information associated with wiring patterns can be performed for similar patterns.

[0177] The recording device records first and second inspection results corresponding to the conductor wiring pattern and insulating region, and records design data information related to the wiring pattern and insulating region. The recording device also stores multiple inspection result groups related to design data of different designs, and records a first inspection result group of a first inspection device having first design information and a second inspection result group of a second inspection device having second design information. The recording device can further cluster the first design information and the second design information, and link the first inspection result group of the first inspection device with the second inspection result group of the second inspection device. The second inspection device can change the conditions for the second inspection based on the linked inspection results of the first inspection result group. The design information includes, in addition to the design related to the wiring pattern and the insulating region, information on at least one of the following: distance between wires, through-hole information, resist area information, materials used, manufacturing plant, transportation distance, manufacturing date, parasitic capacitance, and inspection device number.

[0178] The use of clustering data collected in this way is expected to produce the following effects.

[0179] The inspection conditions of the inspection device 10 can be optimized based on the similarity between the design data and inspection conditions of the clustering data belonging to other inspection systems currently in operation or new inspection systems.

[0180] In cooperation with the circuit design CAD, the design screen of the inspection object X can be displayed, and for example, as shown in FIG. 19, a heat map of defective areas and the clearance between adjacent wiring can be visually displayed. By linking with the circuit design CAD in this way, it becomes easier to identify the cause of defects specific to the inspection object X, which has a specific shape. FIG. 21 is a diagram illustrating the use of inspection results during board design. The board design CAD has a clustering unit 61, a waveform simulation unit 62, a feature extraction unit 63, a similarity determination unit 64 using machine learning, and a display unit 65, which can cluster input design information 66 (first design information), extract features, and output simulated waveform data. Past data 67, including the clustered design information of other boards (second design information), is provided and input to the circuit design CAD from the waveform diagnosis DB. The clustering information output by the board design CAD and the clustering data provided from the past data are provided in the same format and input to the similarity determiner. The similarity determiner can link new design data (first design information) to design information (second design information) included in past inspection information based on the similarity of feature quantities and waveform data. As described above, the past inspection information is linked to the first, second, and third inspection results, and the display unit can display the results by different colors, shades, or numerical values ​​according to the conductor wiring patterns and insulating regions. That is, the design device has first design information corresponding to the conductor wiring patterns and insulating regions, the recording device imports inspection results obtained from an inspection device that inspects wiring boards, and the imported inspection results include inspection results for inspecting the electrical conduction state of conductors or insulators of the wiring board and second design information corresponding to the conductor wiring patterns and insulating regions. The first design information and the second design information can be clustered, and the inspection results can be linked to the design information for a specific conductor wiring pattern or insulating region.

[0181] Comparing similar inspection targets X extracted from different designs may reveal complex factors, such as the tendency for defects in inspection target X4 (see Figure 19) to occur when inspection target X3 is located nearby. In such cases, establishing a new clearance setting between inspection targets X3 and X4 for the new design can prevent defects. This can contribute to improving the yield of new products. In particular, board design CAD can set a minimum clearance distance for adjacent wiring. A larger clearance increases the board area, while a smaller clearance can lead to defects. This system can issue warnings about high-risk areas based on previous inspection results, depending on the adjacent wiring conditions. On the other hand, for areas determined to be less risky, the minimum required clearance distance can be set. A fixed clearance distance can also be set, and a warning displayed for risky areas, prompting the user to adjust the appropriate clearance distance. Furthermore, the previously fixed clearance distance can be made variable, taking into account the adjacent wiring conditions and previous inspection results. In safe areas, lines can be drawn even if they are close to adjacent wiring, but in dangerous areas, the set distance is changed to a distance that ensures safety.

[0182] When inspecting a newly designed wiring board, inspection conditions must be determined. While an operator can determine inspection conditions such as test voltage and current, the procedure for inspecting multiple inspection targets, and the contact points for the probe while expanding the design data and checking the design drawings, inspection conditions can also be determined using inspection condition setting software. Like the board CAD software described above, the inspection condition setting software includes a clustering unit, a waveform simulation unit, and a feature extraction unit, as well as a similarity determination unit using machine learning and a display unit. It also includes an inspection condition setting unit and an inspection procedure setting unit. The inspection condition setting software receives new design data (first design information), extracts features using the clustering unit, and outputs simulated waveform data using the waveform simulation unit. Past data, including clustered design information for other boards (second design information), is provided from a waveform diagnosis database and input to the inspection condition setting CAD. The clustering information output by the inspection condition setting CAD and the clustering data provided by the past data are provided in the same format and input to a similarity determination device. The similarity determiner can link new design data (first design information) to design information (second design information) included in past inspection information based on the similarity of feature values ​​and waveform data. As described above, the past inspection information is linked to the first, second, and third inspection results. The display unit can display inspection results obtained from highly similar wiring patterns on a wiring diagram developed with the new design data, distinguishing them by color, shading, or numerical values ​​according to the conductor wiring pattern and insulating area. Furthermore, the inspection condition setting software can determine inspection conditions based on past inspection results. For wiring numbers that are prone to defects based on past inspection results, it is possible to set a longer inspection time, tighten insulation inspection conditions (applied power, inspection time), and set multiple inspection points and multiple inspections to thoroughly inspect the wiring. On the other hand, it is possible to simplify inspection conditions for areas and wiring that are less likely to develop defects based on past inspection results. It is also possible to inspect wiring numbers that are less likely to develop defects collectively.The inspection condition software determines the inspection order, contact points for the probe, etc. based on clustering and makes suggestions to the user. The user can confirm the suggestions from the inspection condition setting software and press the OK button if they are acceptable, or press the Modify button if they wish to make corrections, make the corrections, and then press the OK button to set the inspection conditions.

[0183] Similarly, when a design change occurs, new clearance settings derived from the accumulation of inspection results as described above can be used, which can contribute to improving yields.

[0184] This technology can be used not only for design changes but also for optimizing inspection conditions. It is possible to anticipate cases where inspection conditions make it easier to judge good products as defective, and it is possible to avoid inspection conditions that are likely to result in such erroneous judgments.

[0185] <Detection of Deterioration of Consumables> For example, contacts used in the first inspection, such as the pogo pins or wire probes 22b, which are components that come into contact with the wiring board, deteriorate over time, and the contact resistance increases until they reach a level where they can no longer be used, and replacement becomes an inevitable option. This inspection system can also predict when the replacement of such consumables will occur due to deterioration over time.

[0186] Using FIG. 22 , we will explain a method for predicting when consumables should be replaced due to aging. The aging of consumables is evaluated using test results that indicate a non-defective product. Taking the pogo pin or wire probe 22b as an example, if the new pogo pin or wire probe 22b is replaced at time T1 in FIG. 22 , the new pogo pin or wire probe 22b has low contact resistance and the voltage measurement time using the pogo pin or wire probe 22b is short, resulting in a relatively short measurement time close to the original theoretical value. Thereafter, with repeated use, the contact resistance gradually increases, and the time it takes for the voltage measurement using the pogo pin or wire probe 22b to reach the threshold value increases. FIG. 22 shows the delay time from the theoretical value (ideal measurement time) to the threshold value due to aging. It can be seen that the measurement time tends to become relatively longer over time.

[0187] Therefore, even if the measured voltage is within the range normally considered to be a good product, if it falls below a predetermined value at time T2, it is determined that replacement is due, and replacement can be performed before any problems occur. After replacement, the device returns to its initial state (the state at time T1), and the replacement time can be detected again in the same manner. Furthermore, the inspection machine may perform an inspection in a state of poor contact due to a sudden abnormality, such as a pogo pin becoming misaligned during measurement. In such cases, abnormalities can be detected by measuring the difference between the input voltage and output voltage of each of multiple pogo pins over time, or by measuring the contact resistance of the pogo pins and graphing and visualizing the results. By issuing a warning when a set threshold is exceeded, sudden defects during measurement can be discovered.

[0188] While performing a pass / fail judgment of the electrical conductivity test, the inspection device can also perform another test (second test) to check whether time-series data such as contact resistance deviates from the model range. The first and second test results are recorded in a recording device. The model range can be set by a user using a rule-based method based on the measurement conditions of past tests, or by using actual past test results as training data to predict the model using a machine learning model. The machine learning model range can also be set using the autoencoder described above. In this case, changes in measurement values ​​that do not cause sudden abnormalities such as pogo pin or wire probe disconnection are used as training data to extract features. A model range is set for each measurement value and displayed on a graph with a time range. Measurement values ​​are plotted by measurement time, allowing operators to easily confirm that the measurement values ​​fall within the model range. Abnormality warnings can be issued not only visually via a display, but also via a lamp on the inspection device, email, or SNS. The vertical axis of the graph in Figure 22 represents measured values, which can include not only voltage but also current, power, DC resistance, AC impedance, and the average value, variance, and coefficient of variation of DC resistance. By comparing the time series data of these measured values ​​with the model range, it is possible to determine the deterioration of consumables over time. Figure 22 shows the change in values ​​over time without specifying the measured values. The measured values, shown as solid lines and dots, are plotted against the model range indicated by the dotted line.

[0189] Subsequent inspections of PCBs can result in defects, and this information is linked to the first and second inspection results, as described above. Defects discovered in later processes may be overlooked due to defects in the inspection equipment's consumables. This inspection system not only stores the pass / fail inspection results and waveform diagnostics during inspection, but also stores time-series data on the consumables' measurements during measurement. By reviewing the time-series data on the consumables and analyzing whether or not there are any abnormalities in the consumables, it is possible to discover defects in the inspection equipment's components that were not detected during inspection. As described above, wiring boards that pass inspection due to defective inspection equipment components are subject to recall. To prevent such defects from occurring, the inspection system can retrain the model range. Specifically, the autoencoder's neural network can be retrained using the changes in measurement values, excluding the time-series data on measurements where consumables malfunction, as training data, to set a new model range. Measurement values ​​that pass the new autoencoder will not match the input and output values, allowing for the detection of consumable abnormalities. That is, the inspection device includes a computing device, a storage device, a display unit, a moving device equipped with components to be moved to a wiring board, a prober having contacts that are components that contact the wiring board, and a fixing unit equipped with components to fix the wiring board, the inspection object being a wiring board made of insulators and conductors and carrying electrical components, and multiple tests are used, the tests including a first test for inspecting the electrical state of the conductor or insulator and a second test for inspecting time-series data of the electrical state of the conductor or insulator, the results of the first test and the second test are recorded in the storage device, and the time-series data of the second test result is accumulated to determine aging deterioration of the components. The storage device pre-stores a relationship between temporal changes in the time-series data of continuously inspected wiring boards and wear of components due to movement and contact during inspection, and the display unit can display the relationship between temporal changes in the time-series data of continuously inspected wiring boards and wear of components. The storage device also records at least one of the following information that shares the time information of the first inspection or the second inspection: temperature, humidity, radio wave conditions, monitored images, noise, vibration, number of inspections, number of contacts, and part usage time.

[0190] If replacement is performed after a malfunction occurs, a good product may be judged as defective, and a certain degree of misjudgment is unavoidable. Furthermore, the inspection device 10 must be stopped while in operation, which can reduce production efficiency. On the other hand, if replacement is performed by forcibly setting a replacement time, the replacement frequency may increase unnecessarily, resulting in increased running costs due to unnecessary part replacement. In contrast, by detecting the aging of consumables as described above, replacement can be performed at the appropriate time, and since the replacement time can be predicted, it is possible to prevent a decrease in production efficiency by, for example, performing replacement before the inspection device 10 is put into operation or during inspection periods.

[0191] There may be cases where a sudden abnormality occurs, as indicated by a ▼ in Fig. 22. Even in such cases, it is possible to determine whether or not the abnormality is sudden by referring to the number of inspections, the number of contacts, the time the parts are used, and the like stored as production information.

[0192] The consumables mentioned above include not only the contacts, but also a moving device (auto sheet feeder) equipped with parts for moving the wiring board, a fixing part (wiring board holder) equipped with parts for fixing the wiring board, and a motor for moving the measuring part.

[0193] [Other Embodiments] The above-described embodiments do not limit the configuration of the present invention. Therefore, the above-described embodiments may include omissions, substitutions, or additions of components based on the description in this specification and common general technical knowledge, and all of these should be construed as falling within the scope of the present invention.

[0194] In the above embodiment, the first inspection unit performs both a continuity test and an insulation test, but the present disclosure also intends to perform only one of these tests. Also, in the above embodiment, the first inspection unit performs a continuity test and an insulation test, but the present disclosure also intends to perform only one of these tests. Furthermore, in the above embodiment, the first inspection unit performs a continuity test and an insulation test, but the present disclosure also intends to perform only an insulation test.

[0195] In the above embodiment, the second inspection device is described as having a display unit, but the display unit is not an essential component. For example, if an inspection system including the second inspection device has a host computer that controls the system, the pass / fail judgment result of the inspection object may be sent directly to the host computer and the judgment result may be displayed on the host computer.

[0196] In the above embodiment, an autoencoder is used to determine whether an object to be inspected is good or bad in the second inspection. However, the inspection method and the configuration of the second inspection device are not limited to an autoencoder. For example, a machine learning model trained using previously collected waveform data and the corresponding pass / fail judgments for that waveform data as training data may also be used. In this case, a machine learning model may be generated for each inspection condition, and the machine learning model for the corresponding inspection condition may be used to determine whether an object to be inspected is good or bad in the second inspection. Alternatively, the pass / fail judgment of an object to be inspected in the second inspection may be based on a predetermined rule without using a machine learning model.

[0197] In the above embodiment, weights are selected in accordance with the inspection conditions of the waveform data in the second inspection, and an autoencoder to which the selected weights are applied is used as a machine learning model, but the inspection method for the second inspection and the configuration of the second inspection device are not limited to this. For example, a plurality of autoencoders corresponding to the inspection conditions may be prepared in advance, and an autoencoder may be selected in accordance with the inspection conditions acquired by the acquisition unit.

[0198] In the above embodiment, waveform data obtained from the test object for data collection is collected in advance for each test condition, and a machine learning model is trained based on the collected waveform data. However, the method of collecting waveform data for training a machine learning model is not limited to this. Test objects for data collection may be inspected under various test conditions, and the obtained waveform data may be stored in a recording device linked to the test conditions. The stored waveform data may then be grouped based on newly set test conditions and used to train a machine learning model. For example, if it is expected that waveform data equivalent to that measured under the initial test conditions will be measured under broader test conditions than the initially set test conditions, the waveform data may be regrouped under the broader test conditions. In this case, the machine learning model can be trained using waveform data from various test objects, thereby improving the accuracy of the machine learning model. Furthermore, since there is no need to remeasure the waveform data, a new desired machine learning model can be generated in a short time.

[0199] As described above, the inspection method according to an embodiment of the present disclosure can reduce the possibility of overlooking defects when inspecting a program and an inspection device wiring board.

[0200] 10 Inspection device 101 Electrical inspection section 102 Frequency inspection section 103 Waveform inspection section 104 Inspection condition information acquisition section 105 Design information acquisition section 106 Image signal acquisition section 107 Environmental information acquisition section 108 Communication section 109 Input / output section 110 Display section 11 Other inspection device 121 Solder printing device 122 Post-solder printing inspection device 131 Mounter 132 Post-mount inspection device 141 Reflow furnace 142 Post-reflow inspection device 152 Finished product inspection device 162 Inspection management device 162a Inspection image 162b Pull-down menu 171 Dealer 172 Support center 12 First inspection device 21 Wiring board holder 22 Measurement section 22A Upper measurement section 22B Lower measurement section 22a Measurement section main body 22b Pogo pin or wire probe 22c Camera 221 Switch circuit 222 Voltmeter 223 Ammeter 224 Variable voltage device 23 Determination unit 231 Switch changeover unit 232 Applied power supply setting unit 233 Voltage detection unit 234 Voltage rise completion detection unit 235 Current detection unit 236 Current increase detection unit 237 Detection time setting unit 238 Control unit 24 Display unit 13 Second inspection device 31 Measurement unit 32 Acquisition unit 33 Determination unit 34 Machine learning model 34a Autoencoder 35 Display unit 41 Server 42 First and second inspection result databases 43 Waveform diagnosis database 44 Clustering 45 Post-process 46 Final product assembly 47 Market 48 Production data tracing 49 Defect analysis 4A Defective waveform data removal 4B Abnormality detection learning 51 Defect report 51a Defective board image 51b Display items 51c Various inspection results 51d Peripheral image 51e Comment field 52 Waveform diagnosis report 52a Board management number information 52b Defect name 52c Inspection name 52d Wiring number information 52e Waveform information management number information 52f Waveform range 52g Inspection conditions 52h Inspection date and time 52i Production information 61 Clustering section 62 Waveform simulation section 63 Feature extraction section 64 Similarity determination section 65 Display section 66 Design information 67 Past data C Inspection conditions D Waveform data D1 Normal waveform data D2 Abnormal waveform data E, E1, E2 Reconstructed waveform data K Cable L Wiring pattern P Wiring board P1 Component side silkP2 Component side resist P3 Component side wiring pattern P4 Board P5 Solder side resist P6 Solder side wiring pattern P7 Solder side silk PS1 Insulation area PS2 Wiring PS3 Through hole PS4 Via PD1 Glass cloth peeling PD2 Disconnection PD3, PD5 Peeling PD4, PD6 Corrosion PD7 Plating defect PD8 Ion deposition PD9 Delamination R1, R2, R3 Area S Wiring board sheet W Weight X Inspection object X1 First inspection object X2 Second inspection object X3 Third inspection object X4 Fourth inspection object

Claims

1. An inspection method for inspecting a wiring board by applying an inspection voltage or passing an inspection current through an object to be inspected, wherein the object to be inspected is a part of a wiring board made of insulators and conductors and on which electrical components are mounted, and wherein a plurality of inspections are used, and the inspections include a first inspection for inspecting the electrical state of the conductors or the insulators, and a second inspection for inspecting time-series data on the electrical state of the conductors or the insulators, the results of the first inspection and the second inspection are recorded in a recording device, and the method further includes a third inspection that is conducted after the first inspection and the second inspection, with the wiring board processed in a subsequent process and with the electrical components mounted, and the result of the third inspection is recorded in association with the results of the first inspection and the second inspection.

2. The inspection method according to claim 1, wherein the results of the second inspection recorded in the recording device are rewritten with information on the defective state of the wiring board determined in the third inspection.

3. The inspection method according to claim 2, wherein the second inspection uses a machine learning model that has been trained to pass time-series data that has been determined to be a non-defective product based on the results of the first inspection.

4. The inspection method according to claim 3, wherein the second inspection retrains the machine learning model using time series data excluding time series data containing defect information determined in the third inspection.

5. The inspection method according to claim 4, wherein the recording device records at least one of production information including temperature, humidity, radio wave conditions, monitoring images, noise, vibration, number of inspections, number of contacts, and part usage time, which share the time information of the first inspection or the second inspection.

6. The inspection method according to claim 5, wherein the machine learning model used in the second inspection is updated according to the third inspection result, and the first inspection conditions of the inspection device that inspects the wiring board are optimized by referring to the third inspection result and the production information.

7. The inspection method according to claim 6, wherein the recording device records the results of the first inspection and the second inspection according to the wiring pattern and insulating area of ​​the conductor, and records design information relating to the wiring pattern and the insulating area.

8. The inspection method according to claim 7, wherein the design information includes, in addition to the design related to the wiring pattern and the insulating area, production information related to at least one of the distance between wiring, through-hole information, resist area information, materials used, manufacturing factory, transportation distance, manufacturing time, parasitic capacitance, and inspection equipment number.

9. An inspection method according to any one of claims 1 to 8, wherein the results of the second inspection are rewritten using defective state information classified by an interface provided in the device that performs the third inspection.

10. An inspection system for inspecting a wiring board by applying an inspection voltage or passing an inspection current through an object to be inspected, wherein the object to be inspected is a part of a wiring board made of insulators and conductors and on which electrical components are mounted, the inspection system comprising: a first inspection unit that inspects the electrical state of the conductors or the insulators; a second inspection unit that inspects time-series data on the electrical state of the conductors or the insulators; an inspection device comprising a recording device that records the results of the first inspection unit and the results of the second inspection unit; and another inspection device comprising a third inspection unit that inspects the wiring board after inspection by the first inspection unit and the second inspection unit, after being processed in a subsequent process and in a state where electrical components are mounted, wherein the results of the third inspection unit are recorded in association with the results of the first inspection and the second inspection unit.

11. An inspection device that inspects a wiring board based on waveform data of an object to be inspected obtained from the object to be inspected by applying an inspection voltage or passing an inspection current, wherein a plurality of parameterized inspection conditions are used, and the inspection device comprises: a machine learning model that collects waveform data obtained from the object to be inspected for each inspection condition in advance and trains the model based on the collected waveform data; an acquisition unit that acquires the waveform data obtained from the object to be inspected on the wiring board and the inspection conditions; and a judgment unit that inputs the waveform data acquired by the acquisition unit and the inspection conditions into the machine learning model and judges whether the object to be inspected is good or bad based on the output result.

Images