WSU Tree Fruit Research & Extension Center

Postharvest Information Network

Sunday, March 24, 2019

WSU-TFREC/Postharvest Information Network/Evaluating Apple Firmness Sensors

Evaluating Apple Firmness Sensors


Apple firmness is one of the most important quality factors. In 1992, the Washington State apple grading standards were amended to include minimum firmness standards. The firmness measurement used in the standard is the Magness-Taylor Penetrometer Test (MTPT). For Washington Extra Fancy and Fancy grades, shipping lots may not contain more than 10% soft apples (12 pounds on the MTPT for Red Delicious and 11 pounds for Golden Delicious). Just prior to shipping, the lot is sampled and, if more than 10% of the sample is below the firmness standard, the entire lot is rejected.

Currently packers cannot remove soft apples from a lot except by hand sorting. This is a source of frustration and economic loss to the packer and grower because any lot with over 10% soft apples is rejected, and many good apples are sold to processing rather than to the fresh market. A nondestructive firmness sensor integrated into an existing packingline could identify soft apples and minimize the number of firm apples rejected as being soft.

Many inventors across the world have stepped forward to meet this need, making a number of performance claims using different testing conditions. The Washington State Tree Fruit Research Commission (WSTFRC) asked the authors to devise and conduct an objective evaluation that would allow the industry to compare the performance of proposed sensors in side-by-side comparisons with the MTPT, using Washington State apples. This article is a report on our evaluations to date.

Sensor Evaluation Procedure

We evaluated sensors each spring from 1991 to 1994. The number and type of sensors and evaluation procedures varied from year to year, but three procedures were followed in every evaluation. First, we tested many apples, about 3,000 per year. This large number of apples was needed to test both Goldens and Reds, to ensure we used apples from growing regions throughout the state, and that we used a range of apple sizes and a wide firmness range. Because most of the apples we obtained from packing houses in the state were above 15 pounds MTPT, we set about a third of the apples to soften for 2 or 3 weeks in a warm room prior to testing. To obtain very soft apples, we set aside a second third of the apples in a warm room for 4 or 5 weeks.

The second procedure we followed was to measure the firmness of each apple by all sensors under evaluation, carefully following the inventors' instructions and ensuring that all measurements were taken on precisely the same location on every apple. Third, the MTPT was performed on each apple using a standard 11 mm dia. tip mounted in an Instron force measurement system at a rate of 25 mm/min. Thus, we conducted side-by-side tests that enabled us to make direct comparisons of sensor performance.


In consultation with the WSTFRC, we measured sensor performance using three criteria:

  1. Could the sensor estimate MTPT values to within 1 lb in the critical regions of 8 to 14 lb (MTPT scale)?
  2. Did the sensor bruise any apples (even those below 10 lb MTPT)?
  3. Could the sensor be integrated into typical packing line systems?

None of the five sensors tested met this criteria. None of the sensors were able to estimate the MTPT in the critical region of 8 to 14 pounds.

Figure 1 is a typical plot of the sensor readings vs. MTPT values for a set of 150 apples. Close grouping of the data points around a sloping line (45 degree incline is ideal) would indicate a useful correlation. In all the sensors, the line is nearly flat, and the grouping of data points around the line is scattered. From inspection of this data, it is apparent that the sensors cannot predict MTPT values within 1 pound in the critical region of 8 to 14 pounds.

Figure 1. Typical output of the nondestructive firmness sensors evaluated. The wide vertical band of data points and horizontal orientation of the points indicate the ineffectiveness of the sensor in predicting Magness-Taylor pressure values.

Analysis of Sensor Evaluation

After the 1991 evaluation, when it was apparent that the sensor prototypes tested would not meet the criteria set by the WSTFRC, we worked with the inventors to determine if a sensor was measuring a physical property not related to the MTPT firmness, or if the sensor design was not calibrated to measure apple firmness. The distinction is important because, if a sensor was measuring a related physical property not calibrated for apple firmness, further work on the sensor would likely improve its performance. If however, the sensor was measuring a physical property not related to firmness, the sensor would never perform to industry expectations.

Many inventors questioned the use of the MTPT as a standard for apple firmness, wondering if the destructive test is an accurate predictor of firmness or is used solely because of its historic importance (Magness and Taylor described the procedure in 1939). The MTPT measures physical properties with destructive deformation. These properties may not be related to material properties measured with nondestructive deformation, making the design of a nondestructive firmness tester that predicts MTPT extremely difficult.

The sensors we evaluated were based on two approaches. The first is that the natural (or resonant) frequency of an object is based on the "firmness" (specifically, the elastic modulus) of the material and the shape of the object. The higher the firmness, the higher the natural frequency. The second approach is that the amount of energy an object absorbs during impact is dependent on the elastic modulus and shape of the object.

Both approaches involve a number of assumptions. Many of these assumptions are valid for materials used in manufacturing, such as teel, but may not be valid for apple tissue. Key assumptions include:

  • The "firmness" of apple tissue is the same as the elastic modulus of the tissue.
  • The elastic modulus remained constant as apple tissue is deformed.
  • The elastic modulus for apple tissue is the same in tension as in compression.
  • The resonant frequency of an apple is affected by the apple tissue's elastic modulus but not by typical variations in the size or shape of Red and Golden Delicious apples.

Measuring Physical Properties

Comprehensive elastic modulus
To test these assumptions, we conducted uniaxial compression experiments in parallel with the sensor evaluations to collect stress vs. strain data. A 10 mm diameter cylinder 15 mm long was removed from a subset of the apples tested each year. These cylinders were compressed between two parallel plates in an Instron Universal test machine at a rate of 25 mm/min (Figure 2). We recorded the applied force and deformation of the cylinder at 1/100 second intervals. From the data, we computed the elastic modulus of the apple tissue as a function of tissue strain.

Figure 3 is a typical stress vs. strain curve for the apple tissue. Tissue failure occurred around 1% strain, and tissue damage occurred at about 0.1% strain. The resonant frequency-based sensors had a good correlation (r2 = 0.80) with the elastic modulus measured at 0.01% strain. However, the elastic modulus at 0.01% correlated poorly (r2 = 0.40) with MTPT values from the apples. The elastic modulus at 0.01% strain also correlated poorly with the first peak on the stress vs. strain curve (r2 = 0.45). As discussed below, the elastic modulus at this level of strain correlated poorly with human perceptions of firmness.

All indications of apple firmness pointed to the conclusion that the elastic modulus at low levels of strain does not predict apple firmness. The results are consistent for Red and Golden Delicious varieties, striped and blush varieties of Red Delicious, and are confirmed across four seasons.

The elastic modulus at a level of strain (0.8% strain), close to the strain needed to bruise apple tissue, correlated well with MTPT (r2 = 0.79). The elastic modulus at 0.8% strain also correlated well with the first peak in the stress vs. strain curve (r2 = 0.82). And, as discussed below, the elastic modulus at this strain rate correlated well with human perceptions of firmness (r2 = 0.76).

Tensile elastic modulus
In 1994, we used a second test to determine if apple tissue has the same elastic modulus in tension as in compression. This assumption (valid for steel and other construction materials) has a direct effect on the vibrational characteristics of the apple. Sensor designs to date that use resonant frequency to estimate firmness use this assumption.

Measuring tensile material properties requires putting the tissue under tensile loads (pulling the tissue apart). Pulling apple tissue is challenging because it is difficult to grasp the ends of the tissue in a way that holds the tissue securely but does not concentrate stress at the mounting points. Additionally, the tissue sample must be tested soon after removal from the apple, before the tissue loses excessive moisture. We avoided the need to apply a uniaxial tensile load on the tissue by placing a block of tissue in a bending load (Figure 4). As the tissue bends, the vertical sides of the block will rotate the top of the side in and the bottom out. The center of this rotation is along the neutral axis-the point where the stress in the block changes from compression (in the top portion of the block) to tension (in the bottom portion). The location of the neutral axis' endpoint in the vertical edge is an indication of the relative size of the compressive vs. tensile elastic modulus. If the neutral axis' endpoint is in the center of the vertical sides, the elastic modulus is the same in both compression and tension. If the endpoint is offset toward the bottom of the slab, then the tensile elastic modulus is greater than the compressive elastic modulus. Because the compressive elastic modulus was measured in tissue near where the block was removed from the apple (compressive elastic modulus section above), the tensile elastic modulus could be estimated from the compressive elastic modulus and the ratio of the distance from the bottom of the slab to the neutral axis endpoint to the length of the vertical sides.

Figure 4. Bending jig and video camera lens used to measure apple tensile elastic modulus.

A 10 x 20 x 5 mm slab was removed from a subset of apples and placed in a bending apparatus (Figure 4). The tissue block was videotaped as the block was loaded to failure. Using computer-based image processing, images of the tissue just prior to loading and just prior to failure were combined to produce images such as Figure 5. The black region above the gray, deformed block outlines the dimensions of the undeformed block. The point where the black region meets the white area just outside of the lower portion of the deformed block is where the neutral axis intersects the vertical sides of the block. The ratio of the distance from the bottom of the block to the intersection of the white and black regions to the length of the tissue block, along with the compressive elastic modulus and the load applied to bend the sample, was used in a finite element model of the block to estimate the tensile elastic modulus.

Figure 5. Superposition of undeformed and deformed apple tissue digital images. Transition point between light and dark regions on the vertical sides were where the neutral axis intersected the edge of the block.

The tensile elastic modulus in apple tissue was about 3 times greater than the compressive elastic modulus. Apple tissue failed in tension at lower stress levels than in compression, indicating that apple tissue is weaker in tension than compression. Because apple tissue elastic moduli were different for tension and compression, the direction of the load (compression or tension) must be specified when discussing elastic modulus. More importantly for firmness sensor designers, analysis of the resonant frequency of apple tissue is much more complex because of the different elastic moduli.

Scanning electron microscope images
The cellular structure of apple tissue consists of cells bonded together in a ring structure. This structure suggests two failure modes: either the tissue cells rupture (allowing attached cells to move relative to each other) or the pectin bonds that hold cells together give and allow intact cells to move. One hypothesis describing why apple firmness decreases with age is that "firm" apples fail due to cell rupture, while "soft" apples fail due to cell debonding. We hoped to find a correlation between the ratio of intact vs. ruptured cells in the failure plane and width with respect to firmness.

Figure 6 illustrates the differences in cell failure along the shear plane which resulted from uniaxial compression of firm and soft apple tissue. The firm image contains a number of ruptured cells, while the image from the soft tissue contains few ruptured cells, indicating cell debonding. These images would support the theory that the cell bonds in firm apples are stronger than the cell walls, failure in firm apple tissue is due to cell rupture, and that the cell bonds weaken with age, resulting in cell debonding in soft apples.

Figure 6. Scanning electron microscope images of firm (left) and soft (right) apple tissue. Images were taken from the shear plane resulting from uniaxial compressive loading.

Sensory Evaluation

The validity of using the Magness-Taylor Penetrometer Tester as the firmness standard was questioned by some who wondered if there was a property that correlated more closely with human perceived firmness than the MTPT. To test the MTPT, we needed to establish a new base for comparison. We decided to conduct a sensory panel as the basis to evaluate the MTPT. A sensory panel differs significantly from a taste panel. A taste panel is questioned to discover likes and dislikes, while a sensory panel is trained to isolate quality attributes (such as crispness, mealiness, sweetness, sourness, and firmness) and rate samples in a consistent manner.

In developing the sensory panel, we tested 50 individuals for the ability to sense quality attributes in apples. The individuals were ranked by their ability to discriminate these attributes and their consistency in measuring the attribute among similar samples. Twelve individuals were selected for the panel. The panel received 10 hours of training over a 2-week period. They refined their ability to discriminate the attributes we wanted (firmness, crispness, breakage, and cohesion), and to agree on a common scoring system so that the panel members scored samples the same.

In 1993 and 1994, we conducted a sensory evaluation of the apples in parallel to the sensor instrument evaluations. After the sensors and the MTPT measured an apple's firmness, the apple was divided into wedges, peeled, and presented to the panel. Members of the panel sat in an environment designed to minimize outside influences and rated the wedges with respect to our four attributes. Each apple was scored by three panelists, and the independent scores averaged for the apple. Each panelist rated six apples in a session, intentionally ranging from very firm to very soft. Panelist scores were normalized (average score of 0.0 and a standard deviation of 1.0) to correct for any remaining inconsistencies between panelists, and then averaged by apple for a rating of each attribute.

Because the sensory panel was conducted in parallel with the sensor evaluation and measurement of physical properties, we could compare human sensory rating with both MTPT and compressive elastic modulus. The MTPT was a good indicator of human-perceived firmness. Figure 7 shows the close clustering (r2 = 0.70) of data about the regression line. The data indicate that the Magness-Taylor Penetrometer Test is a good indicator of human perceived firmness.

Because sensor designers were using elastic modulus to estimate firmness, we compared apple tissue compressive elastic modulus at three strain rates to human sensory perception of firmness. Figure 8 demonstrates that at low strain rates (~0.1%) there is little correlation between elastic modulus and human perceived firmness. The elastic modulus near the first peak of the apple compressive stress vs. strain (strain = 0.8%) had a much higher correlation (r2 = 0.66) between compressive elastic modulus and human perceived firmness. These data indicate that an effective nondestructive firmness sensor based on elastic modulus must deform the apple nearly to the point of causing a bruise in order to measure firmness.

Figure 8. Graph of apple material properties vs. human perception of apple "firmness" indicating a better relationship between human perception of firmness and material properties near the destructive region of the force vs. deformation curve. Initial Elastic Modulus = compressive elastic modulus at a strain of 0.1%. Elastic Modulus near Peak = compressive elastic modulus at a strain of 0.8%. Peak = maximum stress in the apple tissue during compressive loading.

Development of a Computer Model of an Apple

As an aid to sensor designers, we developed a computer model of Delicious apples. The model contains the actual geometry of the apple so that sensor testing is done on the real, naturally shaped apples that the commercial sensor may test. The advantage of the model is that various configurations of sensor parameters (such as materials, transducer heads, excitation signals) can be tested against a wide range of apple sizes and shape (typiness), without the expense of constructing a prototype and the time required to measure firmness on a large number of apples.

Forming the geometric model
To create the computer model, a typical apple was selected, and then sliced into 3-mm slices parallel to the core line. We entered the profile of the apple and the profile of the core into the computer. From the profile data, the finite element program formed a skeleton of the apple, stretched a "skin" over the skeleton, and filled the interior of the apple with "flesh" (Figure 9). By changing the apple skeleton, the user can change the size and shape of the apple model.

Figure 9. Computer model of an apple showing skin, tissue, and core cavity.

The next step in forming the model is to divide the apple into smaller elements. The model contains about 4,000 pyramid-shaped elements. The user can specify material properties (such as strength and stiffness) for each element. The user can specify a firm or soft apple, add a bruise or "sunny-side" regions on the apple, or watercore around the core. The finite element package will predict how various sensors will affect the surface elements, and how the signal from the sensor travels through the interior of the apple. This capability allows the sensor designer to experiment with different apple shapes and firmness values before building a prototype sensor.

Static verification
The computer model is useful only if the sensor designer is confident that the computer results are accurate predictions of how a prototype sensor performs. To check the correctness of the model, we applied two verification tests. The first was a slow, almost static compressive loading of the apple (Figure 10). An apple was placed between two parallel plates which moved together and compressed the apple about 4 mm. The force and amount of deformation of the apple were recorded. The apple's material properties were measured for use in the model, and the apple's geometry was entered into the computer as described above. The computer results matched well with the stress-strain curve of the apple (Figure 11) within the limits of the computer software.

Figure 10. Compression loading of the apple.

Figure 11. Stresses at the center of the apple model and comparison of the experimentally measured reaction forces and computer estimate of reaction forces on the apple.

Dynamic verification
We also verified the model based on vibration of the apple. The apple was vibrated at frequencies from 300 to 2500 Hz. The apple had natural vibrations at 340, 710, 890, and 1860 Hz. The apple model was loaded as shown in Figure 10 and resonance frequencies calculated up to 2500 Hz. Figure 12 shows results of the dynamic verification. The wireframe of the apple model shows how the apple vibrated at 890 Hz. The curve on the graph is the computer's prediction of the apple's response at various frequencies. The four vertical bars are the natural frequencies observed when vibrating the apple. The model predicted the three observed natural frequencies but predicted a fourth natural frequency which was not observed. The dynamic response is important in the design of sensors which measure resonant frequency.

Figure 12. Comparison of experimentally measured natural frequencies (vertical bars) and computer estimated frequency response curve for the apple model.


Our evaluation firmness sensors in 1991 through 1994 failed to identify a nondestructive apple firmness sensor which met industry specified performance levels. Research conducted in conjunction with the sensor evaluation work indicated that critical assumptions made by the sensor designers were not valid for apple tissue. These assumptions were:

  • That apple tissue elastic modulus is constant--the elastic modulus changes as a function of strain.
  • That apple tissue elastic modulus is same for compressive and tensile stress-tensile elastic modulus is about three times greater than compressive elastic modulus.
  • That the elastic modulus is a good predictor of "firmness"-only the elastic modulus at nearly destructive levels is related to human-perceived firmness.

Designing a nondestructive apple firmness sensor will be difficult because of the need to deform the apple nearly to the point of causing a bruise in order to estimate firmness. We have developed a computer-based model of Delicious-type apples that can help nondestructive apple firmness sensor designers optimize their designs.

Marvin Pitts(1), Ralph Cavalieri(1), Steve Drake(2) and John Fellman(3)

(1)Biological Systems Engineering, Washington State University
(2)Tree Fruit Research Laboratory, USDA - Agricultural Research Service
(3)Horticulture and Landscape Architecture, Washington State University

Tree Fruit Postharvest Journal 8(4):13-22
December 1997

Tree Fruit Research & Extension Center, 1100 N Western Ave, Washington State University, Wenatchee WA 98801, 509-663-8181, Contact Us