New Eddy-Current Sensor Setup for High-Resolution Lithium-Ion Cell Dilation Measurements

Lithium-ion cells exhibit a dilation during charge and discharge cycles and over their lifetime. Quantifying this dilation is a reliable method to characterize materials and observe degradation mechanisms. Currently, such experiments base on dial gauges or laser scanners, which are quite expensive and unable to measure spatial differences, to apply pressure simultaneously, or take up a lot of space. In this work, a new sensor and measurement setup is presented that significantly improves the state of the art. The first priority is a high resolution to measure the dilation accurately. While the absolute accuracy of the setup is around $7 \mu \text{m}$ , the resolution of the sensor is lower than 4 nm if the cell-to-sensor distance is less than 3 mm. The second priority is the design of a space-saving jig, which allows the application of a uniform pressure. A setup based on well-characterized foams, which can be compressed to a defined force, is combined with an integrated sensor in the compression plates. Finally, the third priority is to use only off-the-shelf parts to achieve a cost-efficient product. The sensor is thoroughly explained and simulated under different circumstances. A calibration function is developed to map the nonlinear eddy-current sensor output to the corresponding distance. In the last part, measurements on lithium-ion cells are conducted and the results were compared with earlier literature studies.


I. INTRODUCTION
T ODAY, lithium-ion batteries are the industry standard for full electric vehicles. On a closer look, lithiumion only describes a large family of different materials and systems where all have in common that lithium ions are responsible for the charge transfer inside of the battery. Every material enhancement has to be characterized under various circumstances by analyzing information. The most prominent information is the voltage and voltage progression as together with current and time almost all the electrical properties of the lithium-ion cell can be described. However, also mechanical information has to be obtained if devices or large battery packs are built. It is also important to collect such data not only at begin of life (BOL) but also during lifetime and at end of life (EOL) to ensure safe operation over the full life cycle.
Graphite is currently the most prominent anode material, and most research has been on investigating this material. Dahn [1] published one of the first phase diagrams of graphite and lithium and did dilatometry studies considering only the thickness change in graphite [1], [2]. Experiments on larger cells for electric vehicles were conducted by many research groups including [3], [4], [5], [6], [7], and [8]. Some of the interesting findings, which will be used for validation and comparison in this article, are as follows.
1) The observation of a dilation hysteresis which was previously observed for the voltage [3], [7]. 2) A dependence of the cells' dilation on the applied current rate [3], [9].
Another use-case of the cell dilation measurement is the observation of different aging conditions [4]. Especially lithium plating, a quite dangerous mechanism and a limiting for fast charging, can be detected using the measurement of cell thickness increase [7], [8], [10]. The scope of this work is a sensor, which can enhance the development and validation process of new lithium-ion cells. It can be used to screen different materials and cell designs to determine dilation and swelling under specified conditions or to regularly ensure material properties in production.
Most of the mentioned groups used dial indicators to measure the change in cell thickness. However, recent works also applied laser scanning techniques [8], [11], [12] and are able to obtain spatial data. The sensor concept described in this article is based on the eddy-current principle and was first applied to batteries in general by [13], [14]. Grimsmann et al. [15] also applied an eddy-current sensor using an off-the-shelf LDC1614 from Texas Instruments (TI) [16] to investigate recuperation limits. The sensor concept is redesigned and fully evaluated using simulation and experiments in this work. Recently, Pannala et al. [17] published another article and application of this sensor concept which measures the distance to a compression plate and not to the cell housing directly.
Eddy-current testing has been around for many years as a nondestructive and sometimes noncontact measurement method for various use-cases. It can be found in applications for material inspection or process monitoring [18], for corrosion detection [19], and many others as for example collected in the review articles [20] and [21]. The sensor used in this work was also recently applied as angle sensor [22] or as a force sensor [23] which shares some similarities with the approach in this work.
In Section II, we will introduce our sensor design and setup as well as the complete test jig. Furthermore, a setup used to validate the sensor and its calibration is presented. In Section III, the setup is modeled considering different materials and different setup properties. A theoretical, nonlinear calibration is obtained and compared with the results of the calibration setup. In Section IV, the sensor is applied to a real lithium-ion cell, and the measured results are compared to previous publications, showing the applicability of the sensor setup.

A. Sensor
The sensor system bases on an LDC1614 from TI, which is specified as an inductance to digital converter with four input channels. The sensor actually measures the oscillation frequency of LC resonators, which typically consist of a printed coil and a concentric capacitor. During operation, the coil produces an oscillating magnetic field, which is creating alternating eddy currents in a conductive material brought into the proximity of the sensor. The opposing magnetic field of the eddy currents then affects the inductance of the sensor coil, and hence the oscillation frequency changes. A more detailed explanation of the LDC1614 working principle can be found in its datasheet [16].
For an optimal sensor design, the guidelines of datasheet and application notes are followed [16], [24], [25]. To achieve a sensor frequency below 1 MHz and a quality factor of around 20 for the LC resonator, the sensor coil is designed using the parameters in Table I. As all the four measurement channels of the LDC1614 should be used and still fit on the limited surface of the investigated cell, the designed sensor diameters are 13.9 mm. In high-resolution applications, the optimal target distance should be below 20% of the sensor diameter d o [16]. The sensor-to-target distance should therefore be smaller than 3 mm for highest resolution.  All the four sensor coils are placed in equal distance of 22 mm around the LDC1614 chip as can be seen in Fig. 1(a). The design margins between the sensor coil and conducting materials were set to the specified minimum of 20% of the radius and the conducting planes are reduced to the minimum and do not surround the sensors. It is noted that every coil represents a single measurement area of approximately 2 cm 2 .
A regular Arduino 1 Uno R3 is used to communicate with and control the LDC1614 and its sensor coils. The sensor coils are sampled sequentially, each time until a stable sensor value with the highest possible resolution is achieved. No powersaving options are applied to obtain the highest possible resolution. The sample rate of all the four sensors is maximum 8 Hz. This is typically too high for the intended application as cell thickness does not change that fast. To reduce the amount of data, a sample rate of 1 Hz is used. The Arduino itself is connected to a PC which is also controlling the battery cycler, in this case a BaSyTec CTS [26]. The data of the sensors and the cycler are stored in parallel and are later on combined using their timestamps.

B. Pressure Jig
A unique feature of this setup is the homogeneous application of pressure while still being able to measure the If the cell is expanding, the measured distance between the sensor and the cell surface decreases and the foam is compressed. In (b), the foam is compressed on purpose by the applied force. The measured distance that changes is the distance to the lower Al fixture.
cell dilation. This is achieved using the setup illustrated in Fig. 2(a). One side of the pouch cell is placed on the surface of an aluminum (Al) or steel plate while the other side is covered with a technical foam. That foam is previously characterized regarding its stress-strain curve. Above, the sensor printed circuit board (PCB), a 3-D-printed, nonconducting holder, and the top plate are placed. The force on the cell is now set by a fixed distance between the sensor and the bottom plate using the precisely manufactured distance tubes in Fig. 2(a). This results in a desired compression of the foam and leads to an applied pressure on the cell surface. An applied pressure is beneficial to push away gas and for better reproducibility [6], [10], [27].
Summarized, the sensor is measuring the distance from the sensor coil to the next conducting target. In case of the used lithium-ion cell, that is the Al pouch foil enclosing the cell. This distance is in fact the thickness of the foam and could be used to determine the local pressure. As the distance between the sensor and bottom plate is fixed, the absolute cell thickness can also be obtained by subtracting the total gap from the measured foam thickness. For both the measurement properties, the application of a calibration function is necessary as the eddy-current method has a nonlinear sensor value to distance characteristic and depends on exact material choices. In Section III, calibration curves will be obtained from simulation. To validate the calibration and perform a reference measurement, the following setup is used.

C. Validation System
A simplified 2-D schematic of the setup is presented in Fig. 2(b). The used setup is similar to a previously published setup of Deich et al. [28] and von Kessel et al. [29]. These setups are actively controlled pneumatic presses to apply all kinds of mechanical stresses on a lithium-ion cell. To use it as a validation system, it is enhanced using three high-precision linear gauges of type GT2-P12K from Keyence [30] distributed around the setup so the cell thickness change in the center of the setup is well-known at all times. These displacement sensors are specified with a resolution of 0.1 µm and an accuracy of 1 µm.
The exact setup to apply a pressure is not displayed and can be found in the aforementioned references. Instead of a lithium-ion cell, the PCB, its fixture, and a piece of foam are put in the press. The sensor again measures the thickness of the foam, from the sensor to the lower Al fixture. If the press does apply a force on this setup, the foam compresses and the distance between the sensor and Al changes. This change is used for two reasons. First, the foam can be characterized over a wide range of its stress-strain characteristic. Second, the sensor value and the applied calibration are compared and validated using high-precision linear gauges.

D. Measured Cell
The investigated pouch cell has a rated capacity of approximately 2.5 Ah and is produced in our laboratory. A detailed description of how the cell is build can be found in our recent publications [31], [32]. It consists of six double-side-coated graphite anode layers and five double-side-coated NMC811 cathode layers. The anode and cathode are separated by a standard separator of 21-µm thickness. The overall cell thickness including the pouch foil is approximately 1.5 mm. As the sensor is measuring against the Al in the pouch foil, the thickness of the Al is of relevance. The used pouch foil consists of a 40-µm Al layer covered in protective plastic layers of approximately 55 µm on each side.

III. CALIBRATION A. Sensor Simulation
To estimate the achievable resolution with different coil designs, multiple variants were simulated using MATLAB and finite element method magnetics (FEMM) [33]. FEMM is a free and open-source software tool in which magnetic problems as, for example, eddy-current sensors can be analyzed. The tool is also recommended by TI [34]. Because of the highly symmetric circular shape of the sensor, the simulation is reduced to a 2-D problem as sketched in Fig. 1(b). It is important that the top plate cannot be neglected in this calculation, as it is typically conductive. However, the plastic holder can be neglected because it is isolating. FEMM provides the possibility of an improvised open boundary condition (IOBC) for magnetic problems [35] which enables the simulation of magnetic problems in a confined space. In this work an eighth-order IOBC is used to reduce errors of the setup's inductance.
A function that maps a specific sensor value to a target distance is desired. Therefore, in each simulation experiment ,  TABLE II  STUDY CONFIGURATIONS TO SIMULATE THE EFFECT OF SEVERAL  PARAMETERS ON THE CORRELATION BETWEEN TARGET DISTANCE  AND OBTAINED SENSOR VALUE. FOR THE STEEL JIGS, MULTIPLE  TOP PLATE THICKNESSES ARE INVESTIGATED TO ESTIMATE  THE IMPACT OF MATERIAL THICKNESS the target distance varies between 0.5 and 7.5 mm with a stepwidth of 0.25 mm. Although the target distance is adjusted, the results are presented with the sensor count on the x-axis, and the simulated sensor distance on the y-axis as this relationship is used for calibration. Multiple other properties affect the sensor frequency. The most obvious are the accuracy of the printed coil and the capacitor. An LC R meter [36] was used to determine the inductance of the coil. It was found that the inductance varies less than 0.2% while the capacitance spreads 1%, which is also the specification of the capacitor. Therefore, the variance of the inductance is neglected, and each simulation of a configuration is done for the following capacitances: For each sensor PCB, the values of C are determined with an error of maximum 0.1% using the LC R meter. This value is then used during calibration for each sensor.
The actual sensor frequency using a different capacitance can be calculated using 1) Simulation Matrix: Furthermore, different top plate materials, thicknesses and distances are varied and create the simulation matrix in Table II. The study matrix contains only those combinations that are presented in this work. For each of the six study configurations and five different capacitances, 29 target distances are simulated leading to 870 simulation runs.
A single simulation run is realized as follows.
1) All the parameters of the study configuration are assigned to variables.
2) The representation of the study is automatically configured as, e.g., in Fig. 1(b). All the material properties are assigned according to the study parameters. 3) FEMM is generating a mesh, which is denser in proximity of the sensor, the target, and the top plate.
Furthermore, the open boundary condition is created. 4) The initial sensor frequency in Table I is assumed, and the problem is analyzed by the FEMM solver.
5) The solver reports the terminal properties of the circuit from which the inductance is calculated. The new sensor frequency is calculated using (1). 6) Two iterations of sensor frequency and inductance refining are done until a stable result is assumed. The LDC1614 outputs the sensor frequency as a discretized sensor count. The sensor frequency of the last FEMM run is transformed to its respective count using The factor 2 28 is due to the fact that the sensor value is discretized to 28 bits.
2) Target Thickness: The thickness of the Al target showed to be a challenging aspect as together with the targets conductivity it determines which fraction of the magnetic field is absorbed at a given frequency. This is due to the skin effect, a phenomenon where alternating fields are not directly absorbed at the surface of a conductor but penetrate the object with a decreasing intensity [37]. In case of a sensor frequency of approximately 1 MHz, the skin depth of Al is 82 µm. After this depth, the current density in the target decreases to less than 36.8% of its value at the surface.
As described in Section II-D, the pouch foil contains Al of 40-µm thickness. However, the cell contains multiple layers of conducting current collector (CC) which will also influence the sensor. To assess the influence of target thicknesses on the sensor, several different bulk thicknesses and realistic multilayer setups were simulated. These multilayer setups consist of a single 40-µm Al sheet resembling the pouch foil followed by variable numbers of 20-µm Al sheets in 100-µm distance each, resembling the CCs of anode and cathode. To simplify the simulation, no top plates and no variation in the capacitance are included in these runs.
Evaluation of the target thickness shows that a thickness of 2 mm is a good phenomenologic approximation for the given setup and is therefore used for all the tests in Table II. The results of the target thickness will be discussed in detail in Section III-B1. Fig. 3 shows excerpts of the simulation results. In Fig. 3(a), the results of the first study configuration without a top plate are presented. For higher capacitances, there is a shift to lower frequencies and therefore lower sensor counts. This is a direct consequence of (1).

B. Results of Parameter Variation
The general trend in Fig. 3(a) is that for larger sensor-totarget distances, the sensor counts increase faster. To achieve a good resolution, the sensor should be used in the range below 3 mm. For small distances, the sensors' resolution rapidly increases. However, such small distances are not always feasible depending on the used foam and the actual dilation of the cell. For distances below 1 mm, only cells with a relatively small dilation can be measured under constant pressure. Otherwise, the foam will be subject to an additional compression and the pressure on the cell will increase. Depending on the actual application, this might be desired to simulate the behavior in a larger battery pack. In this work, the goal is to measure the cell dilation under roughly constant pressure which is why the sensor-to-target distance is set between 2 and 3 mm.
In Fig. 3(b), different configurations of Table II are presented for a capacitance of 1000 pF. It shows that the addition of a top plate with different parameters has a non-negligible influence on the sensor characteristic. On a large scale, the magnetic top plate (Steel jig 6M) is shifting the sensors' frequency to lower values having the effect of increasing the inductance of the sensor. Materials with higher conductivity as Al or some types of steel have the opposite effect and further reduce the inductance of the sensor. If not considered, the plate shifts the target closer to the sensor.
The influence of the top plate thickness is not very large as can be seen in the inset of Fig. 3(b). A 4 mm-thicker steel leads to an error of ∆d = 4 µm at an overall sensor-to-target distance of 2 mm. A deviation of the steels' thickness will infer a calibration error 1000 times smaller than the actual deviation. Considering the different materials and slightly closer positioning of the Al validation setup compared with the thick steel jig, both better conducting material and closer top plate distance do have the same effect of reducing the sensors' inductance.
1) Estimated Accuracy: Because of many different factors, it is difficult to estimate an accuracy of the actual sensor based on the simulations. It is assumed that all the mechanical parts such as the holder, the top and bottom plates, and the sensor PCB itself are manufactured with an absolute accuracy of at least 100 µm. The deviation for the coil and the capacitor is less than 0.1% where the coil is assumed to be accurate and the capacitance is measured and used as an input to the calibration.
In Fig. 4, the thickness values to a fixed sensor count of 5 840 000 are presented. The figure should be read as different association of a target distance to the fixed sensor count. In the middle, a realistic single-layer cell with appropriate spaced copper and Al layers including the outer pouch foil is simulated. On the right half, the target distances over varied bulk Al thicknesses are presented. Note that pouch alone could also be seen as a bulk Al of 40-µm thickness.
It can be seen that the addition of further CCs as described in Section III-A2 does decrease the distance until a certain level. After three additional Al foils, the distance is slightly increasing again. It can be observed that the influence of additional layers is small and the target thickness is assumed to be realistic at this value.
However, the simulation of five or more of these thin sheets is very computational expensive. A representation of the same value using a simplified bulk Al is desired. Interestingly the variation in bulk thickness seems to have the same effect as adding layers. Its target distance is decreasing until a certain level and starts increasing again. In this case it seems like it will increase to higher values again instead of reaching a Authorized licensed use limited to the terms of the applicable license agreement with IEEE. Restrictions apply. Using the setup in Fig. 2(b), a validation measurement is carried out. The press is set to different forces and therefore different thicknesses of the foam which is measured by the developed sensor and three Keyence sensors. The graph contains the mean values of the four LC resonators and the three Keyence sensors. stationary solution. The simplified representation is marked by the dotted line in Fig. 4 at 2-mm bulk Al. At this thickness, the system behaves similar to a system or cell with multiple separated conductive layers. The remaining difference between the "Pouch + 10 CC" and the "Bulk Al-2 mm" is approximately 50 nm at the target distance close to 2 mm. However, the worst case scenario would be the difference between the "Pouch alone" and "Bulk Al-2 mm" which is 20 µm.
This shows that the better the investigated target is understood, the better the accuracy of the sensor setup will be. At 2 mm sensor-to-target distance, the maximum absolute deviations of the considered parameters are as follows: 1) inductance and capacitance: 3 µm-from Fig. 3(a); 2) plate distance and thickness: 0.1 µm-from Fig. 3(b); and 3) target thickness: 0.05 µm leading to a maximum estimated absolute error of 6.3 µm. For this value, the first two points are multiplied by two, as, for example, both inductance and capacitance can lead to the inaccuracy at the same time.
2) Calibration Function: A trivial interpolation between the observed data points of the simulation is used as calibration function. The starting point is a sensor value obtained from a setup specified by the aforementioned parameters. For each of these setups, a set of calibration data over different distances and capacitances is obtained. The sensor value is used to define the possible set of distances [line (1) in Fig. 3(a)]. Afterward, the measured capacitance of the specific sensor defines the corresponding target distance [line (2)]. Values between the data grid are linearly interpolated.

C. Validation
In Fig. 5, the results of a validation measurement are shown for the interval between 2 and 3.2 mm. For calibration, Fig. 6. Long-term data of the static setup in Fig. 2(a).
the "Validation" configuration of Table II is applied which describes the setup in Fig. 2(b).
A step profile between 2 kN (foam thickness 3.2 mm) and 5 kN (foam thickness 2 mm) is applied on the setup with a step width of 100 N. This results in 31 steps. These steps can still be seen in the validation, although it plots the measured "true" distance of the Keyence sensors against the sensed distance and should therefore be linear. The reason for their visibility is a nonperfect match of sample rates between the two sensors and a pulse-like increase in pressure through the press. The mismatch error between the two sensors is shown in the lower plot. It is apparent that the steps occur more frequently at lower distances. This is due to the nonlinear stress-strain characteristic of the foam that can also be obtained through such a measurement as previously described.
The absolute accuracy of the sensor is better than 2 µm and therefore better as the estimated guess in Section III-B1. Unfortunately, the dimensions of the setup and the maximum way of the Keyence sensors did not allow for a larger validation interval. However, the measurement shows a great confidence inside the limits were most experiments are planned. It has to be noted that the accuracy of the Keyence sensors is 1 µm and its resolution 0.1 µm. Considering these limitations, the uncertainty of the validated accuracy is still up to 50%.
1) Estimated Resolution: The validation carried out ultimately validates the system calibration. Since the sensor is set to use its full 28 bits to discretize the sensor input, the minimal change in sensor output count is known and can be translated to a resolution using the calibration function alone. The resolution of the sensor can still be measured as minimum step height of measured sensor data during the validation (see Fig. 5) or noise measurement (see Fig. 6), but is found to be inherently equal to the value obtained from the calibration.
At 2-mm target distance, the resolution of all the simulated setups is between 2.3 and 2.5 nm. The mean resolution at 1-mm target distance is 1.8 and 3.1 nm at 3-mm target distance.
In Fig. 6, the sensor values of a static setup over the course of one week are presented. A slight decrease in values can be observed, which may be attributed to factors such as aging of clock or capacitors, or minor variations in temperature that could affect the expansion of the cell or measurement setup. The setup is therefore situated within a climate chamber at 25°C with an accuracy of 0.1°C.
The second finding from Fig. 6 is that for 168 h the maximum deviation from the mean value is 120 nm in each direction. All the four sensors show a similar trend, which could indicate that a common external influence is causing the deviation. Nevertheless, channel one shows a slight deviation at maximum 50 nm compared with other sensors. Such residuals cannot be completely avoided as, for example, the surface of the cell might not have the same distance to all the sensors, which affects the resolution. When evaluating small deviations over longer time periods, attention must be paid to the possible effect of the observed noise.

D. Comparison With State of the Art
There are three major differentiators: spatial resolution, measurement resolution, and cost. The presented sensor performs best in the categories measurement resolution and cost. Laser or optical sensors typically have a much better spatial resolution while maintaining a good measurement resolution in case of laser sensors [8], [11], but are quite costly.
Other setups using measurement gauges have the best absolute accuracy but suffer in spatial resolution and cost [4], [7], [10]. Ultrasonic techniques can monitor the state of the inner cell but have a worse resolution [38]. The application of pressure remains a problem for all other setups. Mohtat et al. [6] presented a setup, where the average expansion of the cell can be measured under different pressures. Schulze and Birke [12] presented a setup to apply pressure on hardcase cells while measuring with a laser sensor.

A. Measurement of Charge and Discharge Cycle
In the following, the test setup depicted in Fig. 2(a) is used for all the tests. The current is given in C-rates, where 1C resembles the constant current that needs to be applied to charge the cell in 1 h. In Fig. 7, the charge and discharge voltage and dilation are plotted against the charged capacity. The average among the four sensor positions of the cell dilation is preferred for better visualization. A significant difference between the charge and discharge trend can be observed for both the measurements.
In literature, two reasons can be found for this directiondependent difference. First, there is an overpotential due to the internal resistance of the cell. This leads to a positively shifted voltage during charge and a negative shift during discharge. However, there is also a persistent difference between the charge and discharge trends if very small currents are applied. This remaining difference is also called hysteresis as it depends on the previous cell treatment [39].
Opposing to the voltage, the thickness is less during charging than during discharging. This can be explained energetically as mechanical work has to be done during charging when the cell thickness increases. The discharge process does not need extra energy and therefore works differently. This difference in the two directions is the reason for hysteresis.

B. Current Dependence of Dilation
In Fig. 8, different C-rates are applied, and the voltage and dilation measurements are presented for the discharge case. It can be seen that the trend of all the applied currents differs in both voltage and dilation measurement. Due to the internal resistance of the battery, the voltage drops with higher currents and has less pronounced shoulders.
Two SoC ranges with different behaviors of the cell thickness change can be seen in Fig. 8. At the beginning of discharge, the cell thickness remains at a higher thickness for faster currents. All lines are crossing at approximately 1.45 Ah, and the cell thickness of the slower discharge currents is now higher in the following section. It can be seen that while for C/10 there is still a pronounced plateau between 1.5 and 2 Ah, this characteristic goes away with higher currents. According to Grimsmann et al. [3], this is due to an increased inhomogeneity throughout the cell. The mixture of different stages leads to a more fluent transition from start to end of discharging.

V. CONCLUSION
This work presents a new measurement setup for high-resolution dilation measurement of lithium-ion cells. The method bases on the eddy-current principle and uses only off-the-shelf parts, which makes it a cost-efficient solution. Our setup allows the setting of defined pressures and enables multiple contactless measurement points on the cell surface. In this work, the sensor is thoroughly described and simulated using FEMM. It is shown and explained which factors do affect the absolute and relative accuracy of the sensor and how very high resolutions can be achieved. A setup to validate the sensor and the simulation is presented and proves a good absolute accuracy of below 2 µm or 0.2%.
The article does further contain some measurement examples of lithium-ion cells showing the possibility to accurately measure the cell dilation. The shown use-cases are reaching from the detection of lithium plating, the observation of a hysteresis between charge and discharge, and can provide evidence of a current-dependent dilation behavior. However, this work mainly focuses on the actual sensing behavior while the new setup can also be used to investigate the influence of different pressures on the behavior. Further improvements on the sensor characterization should be done regarding its dependence on temperature and pressure. Although the latter is expected to be relatively small, next-generation batteries as, e.g., the so-called solid-state batteries do currently require pressures in the MPa range which could also affect the sensor behavior.
This type of low-cost but precise and also easy-to-use sensors and sensor setups will simplify the testing of cell dilation and mechanical properties as cell stiffness. They will also enable more data to be generated in long-term tests as cycling or calendric studies as the used sensors are much cheaper and take less space. Further developments will include evaluation of temperature dependence of the sensor and its applicability to other cell formats such as prismatic or cylindrical cells.